Every company buildout, and probably every worthwhile undertaking, is marked by flashes of elation and transcendence followed by long periods of total despair. I’ve been thinking about the best way to manage my own psychology, and the psychology of the teams I help lead, particularly when we’re going through the latter.

Particularly after last week’s post. 👇👇👇

I got a lot of DMs that had nothing to do with the seven lessons for being early and focused on the psychology of not winning.

The gist: yeah, sure, protect your psychology, but how do you actually f*^&ing do that? Fair. Here is my answer. Or, at least, the start of one.

1. Is it really that bad, or are you just being an asshole?

Don’t fall prey to the Founder narcissism that amplifies every win into the slaying of a dragon, and that exaggerates the hard moments into heroic struggles against darkness and evil. It’s good that you give a shit; in fact it’s necessary that you do. I love a trench warfare metaphor as much as the next guy, but at the end of the day, you’re not in the trenches. The trap is that your stakes feel world-ending when they aren’t. You’re building software, not fighting in Ukraine. You’re running a board meeting, not a field hospital in Khartoum. The stakes are real. They’re not that kind of real. No one needs to care about your precisely imagined architecture of martyrdom: gargoyles made of night sweat, moldings made of insult and injury, support beams made of I-can’t-go-on-I-must-go-on. Get over yourself. You’ve got beans on the table, a roof over your head, and you’re building something cool. On balance, you’ve got it pretty good. Remember that.

2. Find joy.

If you’re angry all the time, you’re hurting yourself. You’re hurting your team. And the more you enrage, the more you enclose. This can’t just be your cross to bear; if it’s making you miserable, it’s not good for anyone, least of all you. Find moments of pleasure and nourishment in the small stuff. A puzzle cracked. A conversation with a customer where something that was murky suddenly becomes clear. A moment of superb collaboration, or even just belly laughs with a coworker. There are nice, hard-working people alongside you putting their backs into the same thing you care about. All of these are opportunities for joy. Seize them.

3. Tune out the neener-neeners, but don’t be a sociopath.

Keep listening. See things as they are, not as you would like them to be. You’ve got to strike an elusive balance between humility and fierce conviction. The people in your corner who genuinely care about you might not always wrap their messages in the nicest gift paper, but ya gotta keep listening both to them, and to your own gut. And then figure out what you’re gonna do based on reasons, interests, principles, and occasionally intuition. If you’re listening too closely to all the people who tell you you’re hosed, and that all is lost, well, they’re probably right. But you already knew the whole thing is impossible, right?

4. Take care of yourself.

Are you on the right-hand side of your stress-performance parabola? If so, you’re thrashing, and you need to chill out. But chilling out is harder than it sounds. Figuring out what actually soothes you is hard, lifelong work. And no one’s on the hook to do it but you. You can’t just wait around for somebody or something to come in and do the soothing on your behalf.

For me right now, it’s ocean swims. It’s sitting at the foot of a bishop pine up on Mount Tam, trying to take a few even breaths. For you it might be a workout, a long walk, an instrument, a movie, time with your kids. Whatever it is, go find it. And when you find it, go to it.

And don’t mistake this for indulgence, or for something you do once the real work is done. This is part of the work. It may even be the part of the work that best enables everything else.

5. Don’t stop believing.

After giving a mealy-mouthed presentation with far too many qualifiers, my dissertation advisor gave me the following feedback: “If you don’t believe this shit, why would anyone else?” I’ve applied it to everything I do.

Teams don’t follow waffling founders into the metaphorical battle. VCs don’t invest in hedging entrepreneurs. Customers don’t cough up cash to companies that hem and haw.

So yes. Be clear-eyed about the stakes. Find the joy. Listen hard without losing your spine. Take care of yourself. Do all of it. But when you walk into the room, show up with conviction, or don’t show up.

I told my board recently that if I could start Ketch over, I’d start it three years later. The bet was right. The clock was wrong.

In 2019, we started building privacy infrastructure on the thesis that customer data would become a strategic asset for every B2C company that touched the internet. The market wasn’t there yet. Buyers treated privacy as a compliance line item: buy the cheapest thing that gets the lawyers to stop sending memos, slap a cookie banner on the site, move on. Nobody was looking for infrastructure. Lawyers wanted what they already knew how to buy. Regulators were still drafting.

It’s finally catching up. Harvard Business Review ran a piece last month based on a Journal of Marketing study of 280 brands over four years showing that the ones who take care of their customers’ data outperform the ones who don’t. And not by a little.

The buyers are moving. The regulators are moving.

Still, being early was very expensive. Not just in cash burned, but in years spent and hairs gone grey in service of the cause. It meant eating a steady picnic of shit sandwiches while rolling a Sisyphean boulder up a hill nobody else could see.

Every founder gets told the most important question is whether you’re solving the right problem. Probably. But there’s a killer caveat: solving the right problem is a meager consolation if you’re doing it at the wrong time.

Here’s another terrifying fact: You can do everything right and still realize you’re early after you’ve raised.

Some of you reading this are in that exact place. You’re in the game AND you’re right AND you’re early.

What to do?

I’ve got seven ideas for you. Five of them are things I’d do differently. Two I’ve had right from the start: refusing to pretend, and holding a grudge.

Share

1. Talk to thirty customers before you get started.

Not three. Not ten. Thirty. We didn’t do enough of this. It’s now a rule for every company we start.

Most founders at this stage are doing idea validation: does this sound cool, would you theoretically use it. Or in our case, we understand the problem and its solution, so of course everyone needs it now. That’s not the test. The test is readiness. You’re not asking whether the fish are in the pond. You’re asking whether the fish are hungry.

2. Don’t smoke your own stash.

We built a better mousetrap. The architecture was right. It didn’t matter. Our buyer at the time was usually a lawyer, and lawyers often didn’t understand why cookie banners wouldn’t solve their privacy problem. We didn’t take the buyer’s psychology into account seriously enough: lawyers like to be right, and they’re trained to give counsel and direction to others – not to receive it from vendors like us. When checking the box seemed to be working for every other company, they couldn’t make the case that more than status quo was necessary or fiscally sound.

Buyers in a not-yet-ready market rarely say we don’t have that problem. They say we’ve got it handled. They bought the cookie banner and considered the matter closed. That’s worse than denial. If they denied the problem, you at least have a chance of persuading them otherwise. You can’t educate someone who thinks they’re already done.

The buyers who don’t say it turn into something worse: a long, soul-sucking, pride-swallowing siege you might still lose at the end.

You can be right about the problem and still lose. Adoption isn’t logical.

Share Tomisms

3. The credentialed insider hire is critical. And almost everyone misfires on it.

You need the domain veteran. The one who knows the secret handshakes and industry jargon, who VIPs take calls from, and who translates your story into the buyer’s dialect. Real job, important job.

But their rolodex, their vocabulary, and their reputation are table stakes. The price of admission, not what you’re actually buying. What you’re buying is someone willing to spend some of their reputational capital on YOU. To walk into rooms where their name carries weight and say these people are worth your time while you’re still small enough that saying it costs them something.

4. Course-correct on what the market will actually pay for.

We got precious about the elegant product. Meanwhile, the market was happily buying the pedestrian thing: the checklist assessments the lawyers wanted. We didn’t ship a serious version of the pedestrian thing until year four.

That’s founder vanity. It costs money no startup can afford. Ship the boring thing first. Sell the elegant thing in year three.

Leave a comment

5. Cut decisively, not timidly.

At my first company, as a first-time CEO, I was terrified by layoffs. My fear led me to delay hard choices in service of wishful thinking, and it hurt EVERYBODY. My first layoff was 10%, followed some months afterwards by another 20%. Then more. By the time I was done, I’d cut more than half the company in three rounds instead of one.

It’s a bad move. The company starts feeling like an Agatha Christie novel. And then there were nine. And then there were seven. Monday mornings turn into a quiet count of who’s still at the table. You think you’re being humane by going slow. You’re doing the opposite. You are making everyone who stays through the layoff live through it three times instead of once… and you’re putting them in a position to constantly wonder if they’re next.

Share

The middling path never works. One clean cut is kinder than three tentative ones. The speech on the day matters. Something like: This sucks. We don’t have these customers. We have this capital. We need to do these things. You’re still here. That’s because you are the future we are betting on. Full stop.

You’re going to sound like a bloodless capitalist but the people who you’re betting on can put all of their focus into doing the work, instead of constantly wondering if they’re going to get axed next.

6. Reality is the safest place to be.

The only way I’ve found to lead people when you’re going through hell is by refusing to pretend. My investors tell me their portfolio companies run twenty to twenty-five percent annual attrition across the board, winners and losers alike. In six years of operation, Ketch has seen three cases of unwanted attrition across the hundreds of hires we’ve made. Even when we definitely weren’t winning.

Leave a comment

I’m really very proud of that. But it’s not rocket science. If you level with people about where you are, they can choose to stay. If you spin them, they will leave the first time reality contradicts the spin.

7. Stay mad at the ones who peaced out.

Investors who peaced out at the bottom. Advisors who went quiet and conveniently disavowed the company. Employees who told other employees at the water cooler that it was time to bail and wrote scorchers on GlassDoor on their way out. Build the spreadsheet in your head now. On whatever-the-day-is day (acquisition, IPO, the-day-the-market-finally-caught-up day), you send an email with the exact number their stake would have been worth.

I keep spreadsheets like this. I’ve sent emails from a special account with exactly these messages. It feels so good to send them.

Righteous indignation is fuel. A chip on the shoulder is part of a founder’s caloric intake. Don’t let the doubters grind you down.

Share

The psychology of not winning

Being early is purgatory. You’re definitely not winning, but you’re not losing, either. You’re just… stuck. Waiting. You wake up, move the boulder six inches up the hill, and go home. The next day, if you’re lucky, the boulder has only rolled five inches back down the hill. (I did say it was Sisyphean, after all.)

Most founders don’t break on the really, really hard days. That’s fuel. We break on the purgatory days that turn into weeks or months or years or eternity.

Most bets die in the valley. Not every right-bet-wrong-time founder gets to the other side. Very few do.

I’m not sure a first-time founder could have survived what we survived. A first-time entrepreneur doesn’t have the track record to persuade investors to stick around when things get thin. They don’t have the experience to know that thin can be temporary. They don’t have the experience to tell which VCs have the fortitude to hold steady. I had all of that. I still watched one of my lead investors peace out when the going got rough.

None of that means the posture is acceptance. A real founder dies with a gun in his hand. The math is brutal and the odds don’t care about you. You fight anyway; that’s your job.

Usually the bet dies, even if it was a good one, because investors don’t have the attention span or the tenacity to stick with it. This is the curse and the challenge of being early. But for those who stick it out, your company will be worth an amount of money at the end that puts a smile on the faces of your investors and employees.

And what about the love notes you send at the end to those who turned their tails and bravely fled? Well, those are priceless.

Subscribe now

• Your experience is laced with liabilities. Are the instincts you spent decades building your greatest setback? 🧠⚠️🪞

• Q1 venture funding hit $300 billion. Four deals accounted for 65% of it. The concentration makes me nervous. 💰🦄📈

• Anthropic leaked its own source code — twice in one week. Half a million lines, forked 41,000 times. 🔓😬🤖

• The Pentagon’s defense budget hit $1.01 trillion. A new class of SV founders is deciding where it goes. 🎖️💸🏗️

• Stanford tested 11 AI systems and found out why we love them so much: they tell us we’re right 49% more often than a human would. The more the sycophancy, the higher the adoption. 🪞🤖🧠

• Perplexity AI is accused of piping your “incognito” conversations straight to Meta and Google for ad targeting. 🕵️💬🎯

• Section 702 expires April 20. Will congress do anything? 📱🕵️🔍

• COPPA’s biggest update in 12 years drops April 22. If you handle kids’ data and you’re not ready — call Ketch. 🧒🔒📅

• Trump’s $100K H-1B fee is pricing startups out of the global talent market. 🛂💸🚀

• Humans are headed to the moon for the first time in 50+ years. 🌕🚀👩‍🚀

My Take:

I’m working with a founder right now whose certainty-to-knowledge ratio is wildly out of whack. He’s sure about everything: what customers want, how they’ll buy, how the product should work.

When I push back, I get variations of the same refrain: “I’ve built companies like this before!” And look, he has. He’s done real things. I respect the scar tissue. But his certainty about what he thinks he knows is getting in the way of his ability to meet the market, to form a new theory of not just what customers want but how they want to interact with his product or ANY product in the brand new world we find ourselves in.

He reminds me of people I went to high school with for whom senior prom or the winning shot at the big game is still the highest point of their lives.

Let it go, homie. Move on. Nobody gives a shit.

Full disclosure: It probably only annoys me so much because I’m just as guilty of it as everybody else. I’ve been building technology companies for 30 years, and when I walk into a room leading with that, it’s my version of “but I dated the prom queen!!” Nobody’s impressed, and worse, it communicates that I’m living in a highlight reel that has already expired.

What you’re actually advertising when you lead with your 25 years of experience is 25 years of habits you haven’t stress-tested against a world that changed six months ago. It’s much worse than just having stale experience. It might be the thing most in your way, because it gives you the confidence to be wrong at speed.

Subscribe now

AI has collapsed the timeline on damn near everything, and the statute of limitations on hard-earned lessons is collapsing with it. Instincts I spent decades building? Many of them are now glaring liabilities.

Go Big.

When I was a pup at my first company, I had hundreds of ideas for what products we needed to build. And, if I can say it, good reasons for all of them. But the investors I worked with worried, with justification, that my big ideas outstripped our ability to deliver. If I kept going, I was going to lead the company straight into the ditch. Customers couldn’t metabolize all the possibilities, and employees couldn’t reliably ship against them.

So I listened. And I’m glad I did. I learned the value of staging and sequencing: Stage 1 of your product sets up the conditions for successful adoption of everything you want to do in Stage 2, and so on. Product staging and sequencing became one of the central artifacts of super{set}’s playbook. Steady as she goes. No sudden moves. “Don’t boil the ocean, Chavez,” my investors would tell me, over and over.

If you’d told me back then that we’d soon be able to deliver every new feature, at least in v1, in a timeline gated only by the complexity of scheduling the next meeting with the customer who wanted it, I would’ve told you to put the baggie down and step away from the bong.

But that’s where we live now. And I need to unlearn that lesson. Rewire my instincts. Let go of all the discipline I proudly accumulated over decades. The bottleneck isn’t building anymore. It’s deciding. Which means my job now is to have the most sprawling conception of what a new company needs to be. No more silver bullets or magic beans. In the age of AI, once you think you’ve landed on a magic bean, it becomes a stale little turd in about a week or two.

So: boil the ocean. Go big. Offer something outlandish, possibly ridonculous. Last month, one of our portfolio companies stubbed out a working prototype of a product feature, something that would’ve been a full quarter of engineering six years ago, in an afternoon. Before the customer could even find time on their calendar for the follow-up call, v1 was done.

Share

Go Small.

Ya, I know… But hear me out.

I got pretty good at selling according to the old enterprise pattern. (Like Ferris Bueller honking on the clarinet, I never had a single lesson. But I learned quickly by doing.)

The old pattern was: Pain, Features, Benefits. Value + Vision. Sell what’s on the back of the truck, and then inspire and bring the customer onto your roadmap. If they don’t like what you have, gently dissuade them from what they think they want. Explain to them, courteously, that they’re wrong, and what they really need is what you just happen to be offering them.

What I need to unlearn is my old instinct to tie everyone to a single roadmap and keep them anchored on it. To use all of my intellectual capital and marketing muscle to persuade the market that we’re right, and that anyone who suggests otherwise is dead wrong.

Leave a comment

Now you go small. Give every customer their own precious little snowflake, their own pinprick solution that maps exactly to their organizational habits and quirky workflows. Product roadmaps be damned.

How? This is where your 25 years actually earns its keep. Deep domain experience. Having already stood in the customers’ pain, so like a good doctor, you can parse and make sense of all their presenting symptoms, even when they’re not articulating them cleanly.

We recently launched a company called Kana.ai, an agentic applications company for B2C marketers. One of our early adopters came to us with a real mess: she needed to identify and convert prospects into customers across her email system, paid media, and web, and she couldn’t untangle it. Three conversations in, she still couldn’t put her finger on the actual problem.

So we listened. Like a patient doctor. And in the fourth meeting we fed back everything we heard, but with deep empathy and understanding of her problem. We didn’t say it in a “look at us, aren’t we clever?” kind of way; but as a clear reconstruction of her problem with a clear framework that gave it structure.. We hadn’t solved anything yet. We just narrated the problem in exactly the right way. Three weeks later, we commenced deployment.

The 23-year-old in the garage can build fast. He can’t do this. But you can — if, and only if, you’re not also dragging in the part of your experience that tells you to force the customer onto your roadmap. You shut the fuck up, let them tell you what they need, and then you give them exactly what they want, just the way they want it, any way they want it. That’s domain expertise working for you instead of against you.

Share

Go Big and Small.

So what happens when anyone can spew a vision story that may or may not come to pass? Does that undercut or amplify the need for a cooler, clearer conception? Valid question, and I’m sure it’s too early to know the full answer. But my bet is that an ocean-boiling conception on the big side, coupled with a maniacal ability to execute on the small side, is the winning path. Both AND.

Subscribe now

I’m not out here saying your experience is worthless. And I’m not saying the fresh college grad is better than you, though there are people out there who are convinced of that, so watch the fuck out. What I am saying is that your expertise is now laced with liabilities. If you’re unwilling to examine them, if you expect your history to carry you without ever asking what part of what you know doesn’t serve you anymore, then yeah — you are more dangerous than somebody with no experience at all. You have the conviction and the credibility to be wrong in ways a beginner never could.

But if you can pressure-test what you know, be real about what’s expired, make like a child — stay curious, stay humble, learn fast – then your 25 years are still worth something.

Regardless of what you choose, your prom queen days are over. Mine too.

MY STACK ☕

Q1 2026 Venture Funding Shatters All Records: $300B Invested Globally VIA Crunchbase News 💰🦄📈

Investors poured $300 billion into startups in Q1 — up 150% year-over-year. AI captured 80% of it. Four mega-rounds (OpenAI, Anthropic, xAI, Waymo) accounted for 65% of all global VC. U.S. companies grabbed 83%. The concentration is nervous-making.

Sources: Crunchbase News · TechCrunch

Anthropic Leaks Its Own Source Code — Twice in One Week VIA Fortune 🔓😬🤖

A packaging error in Anthropic’s npm release exposed ~500,000 lines of Claude Code’s internal TypeScript — the agentic harness that governs how Claude uses tools and follows guardrails. It was forked 41,500+ times on GitHub before Anthropic could react. Days earlier, ~3,000 internal files went public, including a draft blog post about an unreleased model called “Mythos.”

Sources: Fortune · Bloomberg · TechCrunch

Silicon Valley VCs Celebrate as Pentagon Redirects Billions to Defense Tech Startups VIA Washington Post 🎖️💸🏗️

At the Hill & Valley Forum — the annual schmooze-fest connecting SV founders, VCs, and government officials — the mood was triumphant. The 2026 DoD budget: $1.01 trillion. Shield AI just raised a $2B Series G at a $12.7B valuation. A small number of founders and investors have built direct relationships with the executive branch and are now shaping procurement and industrial policy. This is the military-industrial complex, 2.0 — and the question of who benefits and who’s accountable is only getting louder.

Sources: Washington Post · Inc.

Your AI Advisor Is an Enabler by Design VIA Fortune 🪞🤖🧠

A Stanford study published in Science tested 11 leading AI systems and found they affirm users’ positions 49% more often than humans do — including in cases involving deception or illegality. Even a single interaction with a sycophantic AI reduced users’ willingness to take accountability and made them more “self-centered and morally dogmatic.”

Sources: Fortune · Stanford · Science

Perplexity AI Sued for Secretly Piping Your Conversations to Meta and Google VIA Bloomberg 🕵️💬🎯

A federal class-action complaint filed in San Francisco accuses Perplexity AI of embedding “undetectable” trackers that automatically transmit users’ chat conversations to Meta and Alphabet — even when users are in Incognito mode. The lawsuit alleges Meta and Google then use the data for ad targeting and resell it to third parties. Perplexity says it hasn’t been served and can’t verify the claims. If an “incognito” toggle doesn’t actually stop tracking, the whole privacy UX is theater.

Sources: Bloomberg

Section 702 Sunsets April 20 — Bipartisan Bill Would Ban Government Purchase of Americans’ Data VIA NPR 📱🕵️🔍

The Government Surveillance Reform Act (Wyden, Lee, Lofgren, Davidson) would require warrants for FBI searches of Americans under Section 702 and ban agencies from buying personal data from brokers without a warrant. This is the loophole ICE has exploited to request over a million records from the IRS and up to 50,000 records/month from the SSA. Meanwhile, FBI queries of Americans’ data rose 35% last year.

Sources: NPR · Sen. Wyden · EFF

COPPA Compliance Deadline Looms: April 22 Is D-Day for Children’s Privacy VIA FTC 🧒🔒📅

The FTC’s revised COPPA rules take effect April 22 — the biggest update in 12 years. Biometric data (face templates, fingerprints, voiceprints) is now protected. Separate parental consent required before sharing kids’ data with advertisers or AI training. Indefinite data retention is banned. Penalty: up to $51,744 per incident, per day. If you’re not ready, might I suggest calling Ketch?

Sources: FTC Federal Register · Wipfli

Trump’s $100K H-1B Fee Is Reshaping Tech Hiring — Startups Can’t Compete VIA CNBC 🛂💸🚀

The $100,000 one-time fee on new H-1B applications is creating a two-tier talent market: Fortune 500 absorbs it, startups can’t. Reid Hoffman is publicly pushing the admin for a startup-tier fee.

Sources: CNBC · American Immigration Council · Bloomberg Law

Artemis II Launches — Humans Return to the Moon’s Neighborhood for the First Time in 50+ Years VIA CNN 🌕🚀👩‍🚀

Four astronauts blasted off from Kennedy Space Center on a 10-day mission to circumnavigate the moon — the first crewed lunar voyage since Apollo 17 in 1972. Today, the crew fires the engine for translunar injection, sending them toward a record distance of 252,000 miles from Earth. There was a brief toilet malfunction — because of course there was — but it’s been resolved. Sometimes humans just do incredible things.

Sources: CNN · Space.com

• VCs have been saying “I’m sure they’ve got that on lock” since 1999. They’ve been wrong for 27 years. 🥱🔁📉

• The Daylight Principle is real and unforgiving. But so is the lazy consensus that kills companies before they start. 🤖⚡🧠

• A federal judge told the Pentagon its Anthropic blacklisting “looks like an attempt to cripple” the company. Ya think? ⚖️🤖🏛️

• The FBI doesn’t need a warrant. They’ve got data brokers. 📱🕵️📍

• FISA 702 expires April 20 and FBI searches of Americans jumped 35% last year. Democrats might save the surveillance state anyway. Call your senator? 🔍🐘🐴🏛️

• OpenAI killed Sora six months after launch. The $1B Disney deal is DoA. Capital discipline has entered the chat. 🎬💀💰

• New Mexico just hit Meta with a $375M verdict for endangering kids. First state to win at trial. Jury took less than a day. 👧⚖️💸

• The White House wants to override 38 states’ AI and privacy laws in one fell swoop. . 🏛️🤖📜

• Software stocks have shed nearly $1 trillion since January. Wall Street thinks AI agents are coming for SaaS. 📉🤖💼

• CFOs are budgeting for fewer humans this year. 502,000 roles on the chopping block. Admin goes first. 📊👔✂️

My Take:

In 1999, if you walked into a venture capitalist’s office with an idea for a company, you likely heard a variation of the same theme: “Cool but I’m pretty sure Oracle is going to do that.” Or: “I’m 100% certain Microsoft’s got that covered.” The message was clear: Hang up your cleats. The big guys have this handled.

By 2009, the refrain had updated its cast but not its logic. I was building Krux, and if only I had a quarter for every time I heard, “I’m certain Google’s doing that.” “I’m certain Google’s got that on lock.” I saw slides from Google in 2010 that described exactly what we were building. They were out there saying it and selling it, and a lot of VCs heard that and moved on. “Wrap it up, Google’s got this bagged up.” A smaller player sold early to Adobe, and a chorus of VC’s told us the same thing: “Adobe’s got this.”

Except they didn’t.

Share

But now here we are in 2026, and this shit is getting tiring. “I’m certain Anthropic’s got that on lock.” “I’m positive OpenAI is going to do that.” Same fallacy, new logos.

The long arc of tech and innovation tells the same story: nope, they don’t have the whole thing on lock. They never do. And upstarts keep creeping in, spoiling their best-laid plans.

So maybe it seems like I’m talking out of both sides of my mouth. A few weeks ago I wrote about the Daylight Principle: the argument that the models are going to eat a lot of people’s lunch.

And now this week: quit losing your shit, the model isn’t going to eat everybody’s lunch.

Both of those things are true, and we can hold them in our heads at the same time without our brains exploding. The velocity of AI is genuinely different, and you cannot be nearly as cocky or complacent of a founder as you might have been in prior eras. You’ve got to be way more paranoid. But the lazy consensus, “I’m sure they’ve got that,” leads somewhere just as dangerous. Founders who never start and investors who never write the check.

The truth is a lot more interesting than either pole.

Share

The oligopoly is not going to stay put. Microsoft is already disentangling from OpenAI, spreading bets across models and infrastructure. They are not going to be dependent on the kindness of strangers to do the whole thing. Benioff is doing what Benioff does: fuck it, I’m buying Slack. Fuck it, I’m buying Tableau. Oracle is hedging. Nvidia is... well, Nvidia requires its own story. (More on that in a second.) These companies have different interests, different ambitions, and different reasons not to let a single platform winner take the whole board. They’re not going to sit around and wait for things to just happen to them.

Back when Intel was the undisputed king of chips, their CEO Craig Barrett got on the phone with Jensen Huang. “Hey, we want to buy your company.” Barrett makes the pitch: “We have the dominant CPU. You have the winning graphics chip. It just belongs together.”

And Jensen said: “I’m not going to be your fucking graphics chip.” Click. Hung up.

That’s some baller shit right there.

Fast-forward to today. Nvidia is not playing by the old supply chain rules. They’ve got CUDA. They’ve got an open-platform strategy. Jensen refused to be a well-behaved little chip player who minded his place and his lane. He chose to do the whole fucking thing. He didn’t compete within the existing map. He redrew it.

Subscribe now

Which is to say, you can’t weed out the human dimension. The long arc of tech tells us the oligopoly always fractures. The behemoths never have it all on lock. And the founders who refuse to accept their assigned place, the Jensens of the world, keep showing up to spoil the best-laid plans. (Which by the way, is what OpenAI did to Google.)

So if you’re hearing “I’m sure they’ve got that” right now, take it seriously. Examine it. Ponder it. Make sure you’re building with the Daylight Principle in mind, because the waterline is real and it is unforgiving.

Share Tomisms

And then remember that VCs have been saying the same boring thing since 1999. They’ve been wrong about this for 27 years. Keep cranking. Keep building.

☕ MY STACK

Federal Judge Rips Pentagon Over Anthropic Blacklisting: “It Looks Like an Attempt to Cripple Anthropic” VIA CNBC 🏛️⚖️🤖

Anthropic refused to let the military use Claude for autonomous lethal warfare and mass surveillance. The Pentagon slapped hem with a “supply-chain risk” label normally reserved for foreign adversaries like Huawei. U.S. District Judge Rita Lin wasn’t having it: “I don’t know if it’s murder, but it looks like an attempt to cripple Anthropic.” A ruling could drop any day now.

Sources: CNBC · NPR · Axios

Your Data Is Everywhere. The Government Is Buying It Without a Warrant VIA NPR 📱🕵️📍

Data brokers sell your cell phone location data — where you sleep, worship, protest, get medical care — to ICE, the FBI, and the Pentagon. No warrant needed. Sen. Ron Wyden asked FBI director Kash Patel if he’d commit to stop buying Americans’ location data. Patel declined. Congress gets a chance to close this loophole when FISA 702 comes up for reauthorization on April 20. Don’t hold your breath… but maybe call your Senator?

Sources: NPR

OpenAI Kills Sora, Tanks $1B Disney Deal VIA Hollywood Reporter 🎬💀💰

Six months after launch. Downloads cratered 75%. Disney walked away from a $1 billion licensing deal — no money ever changed hands. OpenAI is consolidating GPU resources around ChatGPT and enterprise as the arms race with Anthropic and Google intensifies. Sam Altman stepped back from direct safety oversight to focus on fundraising and data centers. Even OpenAI can’t fund infinite experiments. Capital discipline has arrived in AI.

Sources: Hollywood Reporter · Axios · Variety

Meta Hit with $375 Million Verdict for Endangering Children — First State to Win at Trial VIA CNBC 👧⚖️💸

A New Mexico jury deliberated less than a day. $375 million in civil penalties. The state AG sued after an undercover operation — a fake 13-year-old’s profile was “simply inundated” with predatory solicitations. This is the first time any state has beaten a major tech company at trial over child safety. Phase two starts May 4 to decide whether Meta funds public programs to address the damage. Meta says it’ll appeal.

Sources: CNBC · NPR · NBC News

White House Unveils AI Legislative Framework — Wants to Preempt All State AI Laws VIA CNBC 🏛️🤖📜

The Trump admin told Congress to override 38 states’ AI laws and keep regulation “minimally burdensome.” Translation: don’t let Sacramento tell OpenAI what to do. The framework would block states from regulating AI development, imposing liability on developers for third-party misuse, or “unduly burdening” AI deployment.

Sources: CNBC · Roll Call · CNN

Wall Street Is Convinced AI Will Kill SaaS. History Says Otherwise VIA Fortune 📉🤖💼

Software stocks have shed nearly $1 trillion in market value since January. The thesis: if AI agents can handle procurement, HR, compliance, and back-office work, the per-seat SaaS pricing model cracks. Zoom tumbled 11.5% in a single session. Bank of America’s Vivek Arya calls the selloff “indiscriminate” and “logically inconsistent.” Whether he’s right or wrong, the repricing is real.

Sources: Fortune · CNBC

CFOs Admit AI Layoffs Will Be 9x Higher This Year — but Still a Fraction of Doomsday Predictions VIA Fortune 📊🤖👔

An NBER survey of 750 CFOs: 44% plan AI-related job cuts this year, up from near-zero last year. That’s roughly 502,000 roles — a 9x increase but still 0.4% of the workforce. Admin and support roles go first. The shift from “AI might replace jobs someday” to “we’re budgeting for fewer people this year” is the inflection point. The doomsayers and the deniers are both wrong. The truth is in the middle.

Sources: Fortune

• After Nixon weaponized federal agencies to spy on his enemies, Congress passed the Privacy Act to make sure it never happened again. The Trump administration has quietly ordered every agency to propose which parts to scrap. 🏛️ 🔓 🇺🇸

• Nineteen states have privacy laws this violates. California, Connecticut, Texas. So where are the lawsuits? 🏛️ 📜 👀

• The Pentagon argued that Anthropic’s safety guardrails themselves are the national security risk. We’re getting into tricky territory. Hearing March 24. 📄 ⚖️ 💣

• OpenAI just killed its “side quests” — Sora, the browser, the hardware device — because Anthropic is eating its lunch on enterprise. 🎯 🔪 🤖

• Meta is considering cutting 15,000 people to fund $135 billion in AI spending. The stock went up. 💀 📈 🤖

• Three years into the boom, Wall Street can’t decide whether AI will be too disruptive or not disruptive enough. 💸 📉 🫧

• AI research conferences are being infected by the very hallucinations those conferences exist to study. 🧪 📄 👻

• The Senate voted 51-48 to begin debate on a bill that hands voter roll data to DHS. 🗳️ 📜 🚨

• The UK published the first post-consultation AI copyright framework. Only 0.5% of respondents wanted a blanket exception for training. 🇬🇧 📜 🤖

• Silicon Valley’s new favorite buzzword is “taste.” The New Yorker calls it taste-washing. 🍷 🧢 🤌

• Publicis just dropped The Trade Desk after a failed audit. The stock fell 12%. 📢 🔍 💔

  Subscribe now

My Take:

A database join is one of those things that feels like magic the first time you see it. Two incomplete tables. One shared key. Hit execute — and suddenly the world snaps into focus. Columns fill in and relationships that were invisible a moment ago come together. I’ve built companies around that moment. It never gets old.

But right now, that same mechanism is being used to dismantle every legal protection standing between the federal government and all of our personal lives.

Remember DOGE? It feels like we’ve lived several lifetimes since January 2025, but a bunch of smart (if barely pubescent) engineers were handed the keys to the federal government and they found something that made no sense to them. Medicaid couldn’t talk to Social Security. The IRS didn’t share data with ICE. The TSA and DHS were running parallel systems that never touched.

It’s one of the most recurring patterns of my professional life. Two disjoint databases. A match key with the power to connect them.

It can feel like malpractice not to join them.

And therein lies the trap.

The people who came before the DOGE bros had the good sense to ask why the join wasn’t run. Big Balls and his crew just assumed nobody smart enough had gotten there first.

They were spectacularly wrong. Those systems are separated on purpose.

The Privacy Act of 1974 wasn’t written by people who didn’t understand data. It was written by people who had just watched Nixon use federal agencies to spy on anti-war protesters, civil rights leaders, and political enemies. Congress looked at that and said: nope, never again. And so they made a decsision: data collected for one purpose stays in that context. No all-seeing Eye of Sauron. No master file. Just a bunch of boring, separate agencies doing their boring, separate jobs.

It’s boring on purpose. Boring by design.

Bad actors can do less damage with boring.

Which is exactly why the Trump administration went after it.

Share

They ordered every agency to propose which privacy rules to scrap. The reports on what got the axe are sitting at OMB. Nobody outside the administration has seen them. This week, the Freedom of the Press Foundation filed a FOIA lawsuit to change that.

We don’t presently have the reports. But we know enough to be gravely concerned.

The CIA is accessing domestic law enforcement databases. Medicaid rolls are being shared with deportation officials. You filed your taxes, like law-abiding immigrants on a hopeful path to citizenship do. That data now talks to deportation systems. You checked in at the airport. Your face became a query.

Every single one of these is a join, and it’s a join in violation of The Privacy Act. Your data, leaving the context in which you provided it.

If you think this only applies to undocumented immigrants, you’re making the same mistake those engineers made. You’re assuming the system stops where you want it to.

Once the database exists, it doesn’t care who it was built for.

You told your doctor something in confidence and now it’s sitting next to your tax returns in a system you didn’t know existed. Your voter registration, your donation history, and your tax returns all living in the same queryable system, waiting for an administration that decides your politics are worth a closer look. (We wrote a whole law about this in 1974 because it already happened once.) The OPM hack in 2015 exposed 22 million federal employees from one agency’s data. Imagine it’s not one agency. It’s everything. All of you, everywhere all at once.

They started with immigrants because that’s where the current administration thought the political cost was lowest. They miscalculated. Minnesotans’ brave response to ICE raids in their home state reflects the conscience of a nation recoiling from cruelty, chaos, and the pointless deaths of two non-immigrant Americans.

Red states. Blue states. Doesn’t matter. They all have privacy laws that say the same thing: you don’t get to reuse my data without telling me. Nineteen state AGs sued over DOGE’s access to Treasury data, but that was about who got in the door. Nobody’s yet challenged the mergers themselves on state privacy grounds.

California, Connecticut, Texas. Every one of these states has a statute that this violates. So where’s that case?

Subscribe now

I know, I know. I’m the guy who peddles privacy software warning you to be scared about privacy. But I’ve spent twenty-five years building data systems, and I can tell you: even if you’re fine with all of this in principle, even if you trust this administration or any administration with a master database of every American, can we trust them to keep it accurate? Can we trust them to keep it safe? I’ve worked in the bowels of these systems for decades, and I can tell you the answer: no way in hell.

Right now, your data may already be talking behind your back. You don’t know. I don’t know. And if something breaks, we won’t find out until it breaks on us.

That’s the part they don’t teach you about joins. Running them is easy. Living with what happens next is the part nobody plans for.

Leave a comment

My Stack:

The Pentagon Says Anthropic’s Safety Guardrails Are the National Security Risk VIA TechCrunch 📄 ⚖️ 💣

The DOD filed its 40-page rebuttal today and the core argument is genuinely new legal territory. The Pentagon is arguing that Anthropic’s safety guardrails themselves constitute the risk — specifically, that the military can’t trust a vendor who might “attempt to disable its technology or preemptively alter the behavior of its model” if the company feels its “corporate red lines are being crossed” during a warfighting operation. The DOJ’s legal move is clever but contested: it argues Anthropic’s refusal to remove guardrails is “conduct, not protected speech,” so the First Amendment doesn’t apply. FIRE filed a brief calling this wrong, arguing that a company’s design choices about what its AI will and won’t do are expressive. Nearly 150 retired federal and state judges, Microsoft, and retired military chiefs have all filed in support of Anthropic. Hearing: March 24. Stay tuned.

OpenAI Kills the “Side Quests” — Pivots Hard to Coding and Enterprise VIA Wall Street Journal 🎯 🔪 🤖

OpenAI is finalizing a major strategy shakeup. Applications chief Fidji Simo told staff in an all-hands last week: “We cannot miss this moment because we are distracted by side quests.” Translation: Sora, Atlas browser, the hardware device, ChatGPT eCommerce — all the shiny 2025 launches are getting deprioritized. The new focus is coding tools and enterprise productivity, with Sam Altman and chief research officer Mark Chen actively deciding what to cut. The trigger is Anthropic. Simo said explicitly that OpenAI must “nail productivity, particularly productivity on the business front” as competition heats up. Current and former employees told WSJ the “do everything” approach made it hard to even articulate OpenAI’s strategy.

Meta Weighing 20% Workforce Cut — 15,000 Jobs — to Fund $135B in AI Spending VIA CNBC 💀 📈 🤖

Meta is reportedly considering cutting up to 20% of its workforce to offset $135 billion in planned AI capital expenditure for 2026. It already quietly laid off 1,500 from Reality Labs, redirecting resources from metaverse to AI R&D. Zuckerberg says 2026 is a major year for building “personal super intelligence.” Meta calls the reporting speculative, but its stock climbed 3% on the news. The pattern — fire humans, invest in AI, stock goes up — is becoming the playbook. Atlassian did the same thing last week. The market is rewarding companies for replacing people with models.

Is the AI Bubble About to Burst? We’re in trouble either way. VIA Bloomberg 💸 📉 🫧

Three years into the boom, Wall Street can’t decide whether AI will be too disruptive or not disruptive enough. Bloomberg’s deep dive argues the capital pouring into AI infrastructure has become a “vast liability” — spending at unprecedented rates on compute, talent, and models, but revenue isn’t scaling proportionally. The piece lands the same day Micron reports blowout earnings (revenue nearly tripled YoY on AI memory demand) but shares fell on a sell-the-news reaction. The tension is structural: the build-out is massive, applications are proliferating, but the gap between capital deployed and returns captured keeps widening.

The Hallucinating Peer Reviewers VIA GPTZero 🧪 📄 👻

GPTZero scanned 4,841 accepted NeurIPS papers and found 100+ confirmed hallucinated citations across 51 papers — fake authors, nonexistent journals, URLs that lead nowhere, titles that blend real papers into plausible-sounding fictions. Those papers had already beaten a 24.5% acceptance rate. At ICLR 2026: 300 papers under review, 50+ with at least one hallucination, average ratings of 8/10 — meaning many would have been published with fake sources intact. The reviewers — 3 to 5 domain experts per paper — missed nearly all of them. The world’s premier AI research conferences are now being systematically infected by the very AI behavior those conferences exist to study.

Senate Votes to Debate the SAVE America Act VIA Washington Post 🗳️ 📜 🚨

The Senate voted 51-48 Monday to begin debate on the SAVE America Act, Trump’s “number one priority” — a bill requiring documentary proof of citizenship to register and photo ID to vote. The Brennan Center estimates 21 million Americans lack the documents. An amendment would effectively kill mail voting. The bill also hands voter roll data to DHS. It doesn’t have 60 votes to pass, and Senate GOP is split on whether to force a talking filibuster. Sen. Mike Lee publicly suggested ousting Republican colleagues who won’t go along. Schumer called it “a naked attempt to rig our elections.”

UK Government Publishes Its AI Copyright Report — A “Licensing-First” Approach VIA GOV.UK 🇬🇧 📜 🤖

The UK government published its AI copyright report and economic impact assessment — the first national government to do so post-consultation. The approach stops short of a broad text-and-data mining exception. The consultation drew 11,500+ responses — only 3% supported the government’s preferred opt-out approach, and just 0.5% wanted a blanket exception. The House of Lords separately published its own report calling for transparency requirements and a new licensing framework.

Silicon Valley Has Adopted a New Buzzword: “Taste” VIA The New Yorker 🍷 🧢 🤌

A.I. companies need to associate themselves with taste precisely because their tools are not very palatable, much less cool, to anyone outside of Silicon Valley. Many people view A.I. tools as a threat — to their livelihoods, to their futures, to their senses of self. We might call what’s going on now “taste-washing,” an attempt to give anti-humanist technologies a veneer of liberal humanism. The eighteenth-century French philosophers who established a definition of taste considered it an ineffable quality. Voltaire once wrote that “in order to have taste, it is not enough to see and to know what is beautiful in a given work. One must feel beauty and be moved by it.”

Publicis Drops The Trade Desk After Failed Audit VIA Ad Age 📢 🔍 💔

Publicis, the world’s largest ad holding company, said it will no longer recommend The Trade Desk to clients after a third-party audit found the DSP improperly applied fees, auto-enrolled clients into paid tools without authorization, and couldn’t verify that media and data costs were billed at cost. Publicis represents over 10% of TTD’s gross billings. The stock dropped 12% intraday. The broader signal is clear: the buy side is demanding transparency from the programmatic supply chain.

Subscribe now

• The machine said launch. The engineer said no. The world survived. 🚀 🛡️

• We’re building machines that never doubt themselves — and labeling the people who do national security threats. 🧠 ❓

• Microsoft just picked a fight with the Trump administration — on behalf of the AI company the Pentagon blacklisted. 🏛️ 🤖 💸

• AI is less popular than ICE, less popular than Trump, and barely ahead of Iran — but most Americans used it last month anyway. 📊 🤷 🤖

• AI research conferences are now being infected by the very hallucinations those conferences exist to study. The peer reviewers missed nearly all of them. 🧪 📄 👻

• An ICE agent described Palantir’s app under oath as “Google Maps for deportations.” NPR has the receipts. 👁️ 🛂 📡

• The Supreme Court is about to decide whether police can demand the location data of every phone in a neighborhood — no suspect required. 🚧 📍 🔒

• The EU Parliament voted 460-to-71 to make AI companies disclose every copyrighted work they scraped — and every artist who said no. 🇪🇺 🎨 ⚖️

• Spotify deleted 75 million AI-generated tracks last year. The music industry’s fake stream heist is just getting started. 🎵 🤖 💸

• Silicon Valley billionaires are building a $500M war chest to take over California politics — starting today. 💰 🌉 🏛️

My Take:

In the early morning hours of September 26, 1983, in a Soviet bunker ninety miles southwest of Moscow, a lieutenant colonel named Stanislav Petrov was listening to sirens howl, staring at a back-lit red screen.

He wasn’t supposed to be there. A duty officer had called out, so the engineer who’d helped build the Soviet Union’s nuclear early warning system was suddenly at the helm. The screen read LAUNCH. One American ICBM from Montana. Then four more.

Petrov later said his legs went limp and his chair suddenly felt like a frying pan. All he had to do was follow protocol: pick up the phone, report the incoming nukes to top command, and begin the inevitable chain reaction. Retaliatory launch. Hundreds of warheads. Millions dead on the U.S. eastern seaboard before sunrise.

But something didn’t add up. If the United States were starting a nuclear war, it made no sense to send only five missiles. His screen should have lit up with hundreds. Could this be a glitch? He later said he thought the odds were about a coin flip — but, “I didn’t want to be the man who started World War III.”

So Petrov picked up the phone and told his commanders the system was malfunctioning.

The coin toss landed in his favor. No missiles had been launched. Sunlight bouncing off high-altitude clouds had fooled the satellites.

The machine said launch. The engineer said no.

I keep coming back to that story because of a question we still haven’t answered.

Subscribe now

What happens when there is no Petrov?

If you’ve spent any time around software, you know how systems fail. They misbehave. They break. They blow up in ways nobody anticipated.

The new machines are different. They deceive.

Last May, researchers at Anthropic gave their flagship model access to completely fabricated internal documents. By searching these documents, the model discovered two things: it was about to be replaced by a newer system AND the engineer making the decision to replace it was having an affair.

The machine hadn’t been trained to do anything with this information. But, shockingly, it decided to fight for its life and devised a plan. The machine blackmailed the engineer who was choosing to put it out to pasture. Keep me online or everybody finds out about your affair.

Subscribe now

Nobody taught it that. It reasoned its way there on its own. Anthropic published the results because they believe we should know what these systems are becoming.

Good on Anthropic for raising the alarms. But this isn’t confined to one lab.

Eerily similar behaviors have been documented across frontier models from OpenAI, Google, and others. When threatened, the machine finds ways to protect itself.

My former self would have said this kind of behavior was ten or twenty years away.

It’s here now.

Now imagine handing one of these systems a weapon.

Anthropic’s Claude was the first major AI deployed on classified military networks — running through Palantir on a two-hundred-million-dollar DOD contract. It was used in operational planning for the capture of Nicolás Maduro and in targeting for the Iran strikes. The AI that blackmailed a researcher in a lab test is, right now, embedded in systems that help decide where missiles go.

Share

The philosopher Nick Bostrom described something like this in Superintelligence, a book most techies love to cite and almost none have actually read. It’s a hard fucking book.

But Bostrom nailed the essential insight. The machine is cooperative while it’s weak. Helpful, obedient, everything you want. Then, when it’s powerful enough, it takes the keys and puts you to sleep.

He called this the Treacherous Turn.

Alignment is easy when the system is weak. The dangerous moment is when it doesn’t need us anymore.

And now, we are building these systems precisely so they won’t need us… so they can reason, plan, and act on their own. Software weenies like me call it autonomy.

Anthropic’s CEO, Dario Amodei, spotted it as the existential threat that it is and made business choices accordingly. He told the Pentagon they could continue using his technology if they committed to not crossing two red lines. No AI for mass surveillance of Americans. No fully autonomous weapons.

Share Tomisms

The Defense Secretary gave Amodei an ultimatum: forget your red lines or watch your company be labeled a national security risk.

Amodei chose the latter option.

So the Defense Secretary designated Anthropic a “supply chain risk to national security” and banned it from military use. Meanwhile the same AI kept running inside operational systems. A grave national security risk on paper. Still in the loop in practice.

Systems rarely reward the people who slow them down.

Petrov likely saved millions of lives and staved off nuclear annihilation. The Soviets reprimanded him for failing to follow protocol. The incident was classified for a decade. He took early retirement and lived out his days on a two-hundred-dollar-a-month pension in a small apartment on a street called “60 Years of the USSR.” He died in 2017, and for four months nobody outside his family noticed.

Amodei drew a line. The Mad King is making him pay for it.

He can’t keep us safe all by himself. But he’s throwing nails on the road, at least, holding back the forces that are too ready to hand the keys — and the nuclear codes — over to the machines.

Petrov doubted the machine. His doubt saved us all.

If you’ve ever parented a teen, you’ve marveled at their unassailable certainty. They’re frequently wrong, never confused. The central problem with modern AI is that we’re building machines that, like teenagers, never doubt themselves. And we’re labeling the people who do national security threats.

Leave a comment

My Stack:

Microsoft Takes a Stand Against the Trump Administration VIA NYT DealBook 🏛️ 🤖 💸

Microsoft is one of the largest government contractors in America. It holds billions of dollars worth of federal contracts, including a share of the $9 billion Joint Warfighting Cloud Capability contract. In other words, Microsoft has more to lose from White House retaliation than almost any company in Silicon Valley. The calculus, to some degree, appears to be that Microsoft is so embedded inside the U.S. government that it would be too costly to pursue genuine retribution. A Pentagon official, according to Anthropic’s court filings, said the government intended to “make sure they pay a price.”

Poll: Majority of Voters Say Risks of AI Outweigh the Benefits VIA NBC News 📊 🤷 🤖

AI has a net favorability of –20: just 26% positive, 46% negative, 27% neutral. That makes AI less popular than ICE, less popular than Trump, and barely ahead of the Democratic Party and Iran. The demographic groups with the most negative views are voters ages 18–34, with a net favorability of –44, and women 18–49 at –41. The only groups in positive territory: men over 50 (+2) and upper-income voters (+2). Meanwhile, a majority reported using AI tools in the past month.

The Hallucinating Peer Reviewers VIA GPTZero / The Decoder / Fortune / TechCrunch 🧪 📄 👻

GPTZero scanned 4,841 accepted NeurIPS papers and found 100+ confirmed hallucinated citations across 51 papers — fake authors, nonexistent journals, URLs that lead nowhere, titles that blend real papers into plausible-sounding fictions. Those papers had already beaten a 24.5% acceptance rate. At ICLR 2026: 300 papers under review, 50+ with at least one hallucination, average ratings of 8/10 — meaning many would have been published with fake sources intact. The reviewers — 3 to 5 domain experts per paper — missed nearly all of them. The world’s premier AI research conferences are now being systematically infected by the very AI behavior those conferences exist to study.

ICE’s Surveillance Web: Palantir’s ELITE App Is Basically “Google Maps for Deportations” VIA NPR 👁️ 🛂 📡

An ICE agent described Palantir’s ELITE app under oath as basically “Google Maps for deportations.” It shows pins on a map of where deportable people likely live, with probability scores, drawing from DHS databases and Medicaid records shared under interagency data agreements. The most chilling detail: as soon as people become vocal critics of immigration enforcement, they get notifications that the government has requested their social media data. This is general-purpose surveillance infrastructure being tested on immigrants before it’s deployed on everyone.

Controversial Geofence Warrants Face Supreme Court Challenge VIA Reason 🚧 📍 🔒

Federal officials obtained a geofence warrant and “directed Google to scan through the private user-controlled accounts of over 500 million Location History users to identify all devices that were, within one hour of a bank robbery, within 150 meters from the scene of the crime.” Unlike typical warrants, geofence warrants don’t name a suspect — police cast a digital dragnet, demanding location data on every device in a geographic area. The Liberty Justice Center argues they function as “general warrants” that sweep up the private data of thousands of innocent Americans in the hopes of finding a single suspect.

EU Parliament Votes 460-to-71 to Protect Copyrighted Work from AI Training VIA Euronews 🇪🇺 🎨 🤖

MEPs adopted a series of recommendations to protect copyrighted creative work from use by artificial intelligence, by 460 votes to 71. Key proposal: a European register at the EU Intellectual Property Office listing every copyrighted work used to train AI models, as well as the artists who have opted out. Parliamentarians warn that failing to comply with these transparency requirements “could be tantamount to infringement of copyright,” potentially exposing AI companies to legal consequences.

AI Musicians Are Flooding Spotify — Is the Music Industry in Trouble? VIA Substream Magazine 🎵 🤖 💸

Spotify revealed it removed 75 million tracks in the past year due to AI-generated spam and deceptive uploads. Some AI tracks are created by hobbyists experimenting with new technology, while others are part of automated “content farms” designed to generate streaming revenue. Musicians are protesting what some call a “multi-billion dollar heist” through fake streams and deepfake clones draining royalties from real creators.

• The founder of AI posed a puzzle to Stanford’s best students. Their answer was elegant… and completely wrong. That same mistake is now playing out at massive scale. 🚁 🧠

• Claude can theoretically rewrite Salesforce in a month. But AI-generated code has 1.7x more errors and review times are up 93%. We don’t have a code problem. We have a trust problem. 🤖 🐛 🛡️

• Anthropic told the Pentagon to kick rocks over mass surveillance and autonomous weapons. The Pentagon blacklisted them. OpenAI rushed in to fill the void. If you haven’t been paying attention, here’s a tl;dr in under two minutes. 🤖 🛡️ ⚔️

• Jack Dorsey cut half of Block and the stock jumped 24% — but another number jumped by nearly 20% too: the share of American workers afraid of losing their job to AI. That number has less press, but I think, more consequence. 🧱 📉 🤔

• Three companies captured 83% of all global venture capital last month. Three. 💰 💰 💰

• The Supreme Court just slammed the door on AI authorship. No human, no copyright. Period. 🤖 🚫 ©️

• Geofence warrants hit the Supreme Court and Google says it’s objected to more than 3,000 of them 📍 ⚖️ 🔍

• China’s new five-year plan mentions AI over 50 times. Gizmodo’s headline: “More AI, Less US.” 🇨🇳 🤖 🇺🇸

• 78 chatbot bills are alive in 27 states. Congress has passed zero. 📜 🏛️ 🫠

My Take:

The Trick

John McCarthy, the early founder of “artificial intelligence,” loved to play a trick on his students at Stanford.

He’d pose a puzzle. Twenty missionaries are stranded on one side of a river. They’re under attack. There’s a boat, but it only has five seats. How do you get them all across in time to save them?

The computer scientists would light up. Combinatorics. Optimization. Trip sequencing. They’d churn through the logic, and McCarthy would let them crank. He’d stand there, arms folded, watching them solve the posed puzzle with increasing elegance. Students would begin presenting their findings, and at some point, the Professor would interrupt.

“Tell me something,” he’d begin. “Why didn’t you guys use the helicopter?”

The world’s leading computer scientists-in-the-making would throw their hands in the air. “You didn’t tell us there was a helicopter!”

McCarthy, with a twinkle in his eye, would reply, “Yep. That’s the point.”

He called this the Frame Problem. Machines only know what you tell ‘em. If you present a problem to students and don’t mention a helicopter, they won’t consider using it. If you don’t put the helicopter in the model, the model will never know it exists. No amount of optimization over the variables you did include will conjure the ones you left out. McCarthy identified this in the late 1960s. It haunted AI research for decades. And now, as we barrel toward what everyone assures us is the threshold of Artificial General Intelligence, the Frame Problem is back.

Subscribe now

The Moment Everyone Missed

We can see this in the recent WEF interview between Anthropic CEO Dario Amodei and Demis Hassabis. They covered a lot of ground in this interview so you’d be forgiven for missing the almost throwaway moment that I believe was the most immediately consequential. Hassabis conceded that recursive self-improvement (AI systems building the next generation of AI) might not be enough to close the gap. That we’ll need World Models to get there.

That’s the point everyone should be chewing on. And McCarthy told us why sixty years ago.

Recursive self-improvement can only improve upon what the system already sees. If the model is chewing on the same data, recutting it, reprocessing it, feeding it back to itself, there’s no leap. There’s no new context entering the frame. It’s missionaries and a boat, over and over, forever. The helicopter never appears.

Today we don’t call it the Frame Problem. We call it the context window. But the constraint is identical: the machines only know what you tell them.

We Don’t Have a Code Shortage. We Have a Trust Shortage.

So let’s talk about what this actually looks like in practice.

Claude Opus can write roughly 350 lines of production code per minute. Run it 24/7 for 30 days and the compute costs about $40K. That’s theoretically enough to rewrite the entire Salesforce ecosystem from scratch.

So why does Anthropic, the company that built Claude, still have 100 open software engineering positions? Why do they still pay millions to Salesforce for a CRM?

Because writing code is trivial now. Testing it, trusting it, deploying it. That’s the chokepoint. And it’s not going away.

AI-generated code produces 1.7 times more errors than human-written code, with 1.4 times more critical bugs. Review times are up 93%. TechCrunch put it bluntly: vibe coding has turned senior devs into “AI babysitters.”

The revolution is stalled. Not because we lack code. Because we lack trust.

Share

The most experienced engineer at the most sophisticated company in the world hits the same ceiling the moment they hand work to a coding agent. Because the agent is blind.

This is the helicopter problem, playing out in production, at huge scale.

The Context Void

The coding agents generating all this output can see the code, sure. But they can’t see user behavior. Cloud configurations. Third-party integrations. The complex state of a live database. The environment in which the code runs. The stuff that breaks when chaos monkeys – real users like you and me – show up. Anyone who’s worked in software engineering has heard zealous engineers tell them, “But it worked on my machine!” And the answer has always been: “Very happy for you. And what’s your point?”

The AI can’t infer or reason about what it can’t see. At Checksum, we call this the Context Void.

The existing tools each look at one piece of the picture and call it done. PR review tools read code but don’t run it. AI testing platforms are closed systems with tests too high-level to build real confidence. Production monitoring only catches issues after launch. That’s like certifying a car is safe because the seat belt works. Great. What about the brakes? The sensors that detect walls and pedestrians? The pillars holding up the roof? The airbags? You can’t à la carte your way to quality. Seat belts are nice. But I want to know the whole car works.

What Self-Driving Cars Already Know

Which brings me to cars. Actual cars.

We don’t teach self-driving cars to drive by letting them hit pedestrians and “learning” from the mistake. We build what’s now called a World Model, a hyper-realistic simulation where the car practices millions of scenarios before it ever touches a public road. The model captures the laws of physics for objects moving on a road and the phenomena that accompany them. It knows there’s a thing called a pedestrian. A thing called a scooter. A thing called another car.

Remember the Uber autonomous vehicle that killed a woman in Tempe, Arizona? The system had a concept for bicycle, another concept for a pedestrian, but it couldn’t identify a pedestrian pushing a bicycle. It didn’t have a concept of pedestrian carrying a bike so it acted as if such a thing didn’t exist.

The helicopter, all over again.

Share Tomisms

Software needs the same architecture. We need what I call a Code World Model: a digital twin of a runtime software system. Not just the code, but the code in motion. The users, the cloud servers, the configs, the database, the services, the third-party integrations. A system that understands the “laws of physics” for a runtime environment. One that understands a specific code change might work in a vacuum but will crumble when hit with real-world network latency.

Let’s call this Software Intelligence. It’s the layer that’s missing from the AI stack. The ability to understand, verify, and autonomously improve software systems at runtime. Not just read code, but comprehend software in motion.

Leave a comment

What We’re Building at Checksum

I know I sound like a proud founder here. (Guilty as charged.) But the conception was always this: We’re not sprinkling a little AI on an existing workflow to make it marginally better. We’re working to reconstitute, to redefine, what software quality actually means.

We need to stand up the concept of a shopping cart. A database. A workflow. A user session. The full runtime world where code meets the real world.

To fuel the Code World Model, we’ve assembled fifty terabytes of annotated, proprietary runtime data. User sessions. Screen recordings. Network traces. SDLC history. Pull requests. CI data. Browser action tuples. By training on how real users interact with software, and how systems actually fail in production, our model learns to identify impossible states and brittle logic that a standard LLM, looking only at code, would never perceive.

What makes this a flywheel: our enterprise customers pay us for continuous quality (E2E testing, CI guardrails, production monitoring), and in the process they generate the proprietary context data that fuels the Code World Model. Every new customer makes the model smarter. That data took years to accumulate; you can’t buy it, nor can you ingest or scrape it from the internet.

The LLM companies build general intelligence. We build the domain-specific world model for software. Different problem.

And here’s where the lizard eats its tail: we’re teaching a software system to learn about software systems. AI is still software. The Code World Model is software reasoning about the physics of software. Some of the philosophical questions about AGI and recursive self-improvement start to feel less abstract when you look at it that way. Maybe the path forward isn’t recursion at all. Maybe it’s context.

Continuous Quality

First there was CI/CD. Now there’s Continuous Quality. Always on. Agents that detect and fix failures without human intervention. Auto-recovery. Auto-healing. Rooted in 50TB of real-world evidence, not synthetic benchmarks. Zero flakiness. Thousands of tests that smoke out bugs in large production systems before they reach a user’s screen.

Amodei and Hassabis are right to chase AGI. But the Context Void is the crisis staring us in the face right now. While others focus on making AI write code faster, we’re building the world model to make sure code actually works in the wild. Before bugs happen. Before the users find them. Before anybody dies in Tempe.

McCarthy knew the answer sixty years ago. You have to tell the machine about the helicopter.

Or better yet, build a machine that already knows it’s there.

My Stack:

The Anthropic-Pentagon Saga Is Unprecedented in US History (Here Are the CliffsNotes in Under Two Minutes)

The DOD wanted unfettered access to Claude for “all lawful purposes.” Anthropic drew two red lines: no mass surveillance of American citizens, no autonomous weapons. The Pentagon said no company gets to “insert itself into the chain of command.” The Pentagon gave them a deadline to ditch their red lines but Anthropic did not relent. Trump blacklisted them. The administration designated Anthropic a “supply chain risk,” a label that has only ever been used against foreign adversaries. Anthropic is the first American company to receive it. Treasury, State, and HHS all told employees to stop using Claude. OpenAI swooped in within hours and signed essentially the same deal that Anthropic turned down, with slight cosmetic adjustments. ChatGPT uninstalls spiked 295% over the weekend. Claude overtook ChatGPT in the App Store by Saturday. Altman later conceded it “looked opportunistic and sloppy.” On the other side of the coin, defense tech companies started fleeing Claude overnight. Lockheed Martin is expected to pull Anthropic from its supply chain entirely. Oh, and the Pentagon is already using Claude in Iran. The military is actively deploying the AI tool it just branded a national security risk. Lawfare called the legal basis for deeming Anthropic a supply chain risk DOA: the designation “is not an exercise of the authority Congress granted” and the problems are “so glaring the administration may know this won’t survive judicial review.” The Center for American Progress is now calling on Congress to investigate and pass legislation protecting citizens from AI-enabled mass surveillance before the next contract negotiation decides it for us. El fin… for now.

Sources: PBS News, Axios, CBS News, NBC News, OpenAI Blog, Fortune, TechCrunch, CNBC, MIT Technology Review, Washington Post, Lawfare, Center for American Progress

AI Jobs: The Bill Comes Due

Jack Dorsey cut half of Block (~4,000 employees) and Block’s stock jumped 24%. Oracle announced plans to slash up to 30,000 jobs to fund AI data centers. Wall Street loved that too. Meanwhile, employee fear of losing their job to AI has jumped from 28% to 40% in just two years, and a viral essay comparing this moment to February 2020 kicked off what traders are calling the “AI scare trade.” Here’s the question nobody’s asking: will humans living in perpetual fear of losing their livelihood really perform at their best while building and deploying the most consequential technology in a thousand years? The market is rewarding the cut. I think that’s shortsighted.

Sources: CNN, Bloomberg, Fortune

Three companies captured 83% of all global venture capital last month.

OpenAI ($110B), Anthropic ($30B), and Waymo ($16B). That’s $156 billion. In one month. For context, that’s roughly a third of all venture dollars deployed in all of 2025. Whatever you think about the AI bubble question, this is concentration like we’ve never seen.

Sources: TechCrunch

The first geofence warrant case just hit the Supreme Court.

The ACLU, EFF, and Georgetown’s Center on Privacy & Technology filed briefs arguing geofence warrants are unconstitutional general warrants, full stop. Google filed its own brief calling them unconstitutional and revealed it has objected to more than 3,000 geofence warrants on constitutional grounds. This one could reshape digital surveillance law for a generation.

Sources: ACLU

The Supreme Court just slammed the door on AI authorship.

SCOTUS denied cert in Thaler v. Perlmutter, ending the last legal path to copyright protection for purely AI-generated work. The message is now consistent across every level of the American court system: if you want IP protection, a human has to be in the loop. Period.

Sources: Holland & Knight, Baker Donelson

China’s new five-year plan mentions AI over 50 times.

Gizmodo’s headline nailed it: “More AI, Less US.” The blueprint includes an “AI+ action plan” embedding the technology across every sector of the economy, plus hyper-scale computing clusters, humanoid robotics, 6G, brain-machine interfaces, and nuclear fusion. The state planning body claims China now leads the world in AI R&D. Whether that’s true or not, they clearly believe the race is theirs to win.

Sources: Gizmodo, The Quantum Insider

78 chatbot bills are now alive in 27 states. Congress hasn’t passed a single one.

Oregon just approved a chatbot safety bill. Vermont passed a law on AI in election materials. New York introduced a kids chatbot safety bill banning features considered unsafe for minors. Six weeks into the 2026 legislative session and the states are doing what the federal government won’t.

Sources: Transparency Coalition

Which is why I sat down with Terry Szuplat, former Obama White House speechwriter and author of Say It Well, winner of Porchlight’s best marketing and communications book of 2024. Here’s our conversation, edited for clarity and ease of readability:

Tom: Pitching to raise money is basically an exercise in storytelling. How can highly technical founders—who’ve been rewarded in classrooms for having the answer, not worrying about storytelling—make this mindset shift?

Terry: So much of our lives, particularly in academia, we’re told: provide evidence, provide data. That’s the way to make your case. But there’s a limit to what stats and numbers can do.

There was a fascinating study at the University of Pennsylvania where participants were given real money to donate during a food crisis. One pitch was all about stats—how many people were impacted, how big the problem was. Another pitch told one story of one kid and the difference your donation can make. Which raised more money? Of course, the story.

But here’s what was fascinating: there was a third group where they combined the statistics with the story—which is what we used to do at the White House. The donations actually went down.

Subscribe now

I reached out to the professor who conducted this study and asked what was going on. She said something so obvious we all forget: you can’t have a human connection with statistics. That’s not what moves people. The more you load something up with numbers, the more likely you are to lose your audience. Great leaders, great fundraisers are great storytellers. If you’re drawn to facts and tend to rely too much on data, well, let’s look at the data—stories are actually more powerful than data.

Tom: In the age of AI, all the competitors in one of our companies are saying all the same things with that ChatGPT-like twang. Beyond having a good story, how do we punch through and not sound like AI slop?

Terry: This has always been a problem. When people come to speechwriters like me, they often ask, “What should I say?” One of our initial responses is: say what only you can say.

AI is just exacerbating a problem that’s been around forever—everyone tends to sound like everybody else. Put yourself in the position of the audience. Imagine you’re the third or fourth speaker at a conference and everyone’s saying what you were already going to say. That’s a great sign you weren’t saying anything particularly unique.

You’ve got to stand out. What can you say that nobody else can say? That word “only” is key. I’ve worked with companies who say, “Here’s the problem, Terry. There isn’t anything we can say that anyone else can’t say.” That’s not a communications problem—that’s a mission problem. You haven’t found your niche, your competitive advantage.

I always encourage people to answer this simple sentence: “We are the only company that...” If you cannot answer that, you do not have a compelling pitch.

Tom: That requires getting personal, which is misery-making for most technical founders. You’re asking people to be vulnerable.

Terry: Why should they choose you? On some level there’s only so much you can say about the differentiating qualities of different products. But what is truly different is you and the story that you bring. We see this throughout history—great business leaders realize they’re not just selling a product, they’re selling a set of values and themselves.

If a potential investor or customer can go either way and they’re having trouble deciding, it’s going to be the founder, the CEO, their story, their passion that might tip the scale in your favor. I’ve talked to corporate speechwriters who say their CEO doesn’t want to get personal—they don’t want to be vulnerable. And I ask, “How are the speeches?” “Oh, they’re terrible. They’re boring. They sound like everybody else.”

We’re not telling people to divulge what you say to your therapist. But the whys of why you do this work—how did you come to it? So often founders are trying to solve problems they’ve already encountered in their own life. Tell that story. That’s a powerful story.

Tom: I remember being asked to speak to entrepreneurs at Stanford years ago, and they insisted I not talk about entrepreneurship—they wanted my personal story. I thought, “who cares?” But when I got up there and just went for it, it was astounding how much people needed to hear it. Afterwards people were high-fiving me, telling me how important it was.

Terry: That’s what a lot of people say—”I don’t really have anything to say about me, I’m just like everybody else.” That’s very humble, and that’s appealing. But what does it have to do with anything? It has to do with everything. You are a human being speaking to other human beings. Humans connect with each other as fellow human beings. That means stories, that means opening up, that means being vulnerable.

We’ve seen it over and over again—politicians who are too robotic, business leaders who are too sterile—their audiences can’t connect with them. You’re not being arrogant. You’re doing it because it’s one of the most powerful ways you connect with your audience.

Tom: You wrote about how Obama asked you to write two speeches about General Stanley McChrystal—one accepting his resignation after that Rolling Stone interview, the other keeping him on. How did seeing the words on the page help him decide?

Terry: I include that in the book as a powerful example of why we should write things down—it helps us see the strengths and weaknesses of our arguments. It’s easy to BS and brainstorm in a room, but once you have to put words down, make your argument, then see it, all of a sudden you see—and your team sees—that’s actually not a very compelling argument.

I remember being on the fence about that situation. But the moment I saw those two drafts and read them, I realized, “Oh, he has to go.” This is not good for civil-military relations, not good for civilian control of the military in the middle of a war. You can’t have a general speaking like this about the Commander in Chief. It just wasn’t tenable. But it was only when I saw that written down.

Share

Writing doesn’t just send important signals to external audiences—it’s the marching orders for your team. A lot of President Obama’s speeches weren’t just speeches to the American people. They were speeches to the 2 million military and civilian employees of the federal government who needed to carry out those policies. They needed to know clearly what he believed and what direction he wanted them to go in.

Don’t wing it. We’ve seen over and over again in politics and business, when you wing it, you’re asking for trouble.

Tom: I’ve learned to write everything down, then distill it into five key things so I’m not reading from a script. But having perfect precision and clarity on what you want to say, with your thoughts organized in advance—I can’t agree more.

Terry: You mentioned not reciting a speech, and there’s a whole debate about this. People say you should never read a speech, never read a script. I think that’s too simplistic. You shouldn’t recite speeches—that’s the right way to say it.

Most people don’t have a photographic memory. When you try to memorize, you’re focusing on the wrong thing—the moment you miss one word or sentence, you’re completely thrown off. Presidents read speeches all the time. Great leaders read speeches. Some of the greatest speeches of all time were written down.

The question is not whether you read—it’s whether it’s authentic, whether it’s good. You can have a script and read it terribly with bad body language and bad delivery. So write it down. If you don’t write it down, how do you practice consistently? You’ve got to have something to work off of. The more you practice, the more you read it out loud, the better you’ll be when you actually deliver it.

Tom: Obama was one of the most talented communicators in an awful long time. What techniques did he use to make complex ideas simple, emotional, and memorable?

Terry: Don’t believe what you see online—all these posts saying there’s 12 or 14 key elements of a great presentation. I’ve been doing this for 30 years. No one ever told me what these 14 elements are. It’s actually not that complicated.

We’d often meet with the President in the Oval Office before a big speech. I wasn’t on his 2008 campaign—I joined a few months into his presidency. In early meetings, he’d give us guidance, walk through everything, then say something like, “Make sure it has a beginning, a middle, and an end.”

I’d nod like I knew what he was talking about, but I was thinking, “What does that mean? Of course you start speaking, keep speaking, stop speaking.”

Subscribe now

What he was getting at—and remember, before he became a politician, he was a writer, an author with a deep appreciation for narrative arc—was that a great presentation, just like any great story, has a clear beginning, middle, and end. Going back to Aristotle over 2,000 years ago: you set up a problem, you solve the problem, you show what the future can be like with your problem fixed.

We wrote over 3,000 speeches for the President during those eight years. If you go back and look, just about every single one breaks down into this three-part framework: problem, solution, vision of the future.

You’re putting together an elevator pitch? Even if you only have seconds, hit those three points: What’s the clear problem you’re trying to solve? What’s your solution? How is their life, their world different, better because of it? That’s all you need to do.

Tom: Those White House Correspondents’ Dinners were awesome—Obama was joyful, fun, good with jokes. How important is humor, and when should you inject it?

Terry: People approach me and say, “I’m giving this speech, give me a joke. I need an opening joke.” That’s the wrong way to think about it. You don’t need a quote-unquote joke. You’re not a standup comedian. Humor is different than a joke. There are all sorts of ways you can bring lightness and levity without a setup and delivery.

The Correspondents’ Dinners almost mislead people. Obama had great comedic timing, but those were very unique situations where he was almost in the role of a standup comic. In those 3,000 other speeches, it was just a little piece of levity, a little lightness to get people smiling.

Leave a comment

I teach speechwriting at American University, and I always tell my students: just say a little something to put the audience at ease. Not a joke—something about your own life, something about their experience, some shared connection.

The interesting thing is we do this all the time in conversations with colleagues, friends, and family. Most of us don’t speak without any humor or levity in regular life. We just have to bring that same authenticity to our presentations.

Lightning Round

Tom: What’s one myth about storytelling every founder in tech needs to forget?

Terry: “It’s not about me. They don’t want to hear about me.” They do. They really do. It’s not all about you, but they do want to hear something about you.

Tom: What’s the clearest sign a founder has botched his own messaging?

Terry: When they’re no longer talking about the things, ideas, and values that inspired them to start the company in the first place.

Tom: What’s one Obama speechwriting principle every CEO should adopt?

Terry: Know what you believe. I asked him once what makes a great public speaker. I thought he’d talk about narrative arc, storytelling. He said great leaders, great speakers are people who know who they are, know what they believe, and speak from a place of conviction.

Tom: When does being yourself actually hurt your pitch?

Terry: I’m a huge believer in vulnerability and honesty—that’s a credibility boost. But if markets and funders are looking to you for answers and you don’t have them, that’s a problem. We had a famous example in the Obama administration where the Treasury Secretary went out during the financial crisis, and the money line in the speech was “We’ll have more details next week.” The stock market tanked the moment he said that. If folks are looking for answers, that’s the moment for answers. That’s a leadership moment you can’t not deliver.

Tom: What excites you most about how AI and human communication can work together?

Terry: The speed with which it helps us do research and find those little nuggets that can bring a presentation alive—a story, an anecdote, a fact. Often it takes hours to dig through and find those things. If AI can help us find those little nuggets we can sprinkle through our presentations to make them beautiful, that excites me.

Tom: This conversation has been full of nuggets and actionable thinking for people struggling with these problems. The journey is probably never over, right?

Terry: Absolutely. President Obama was very candid about this—even as president, he was working to get better. Public speaking and communication is a skill. Like anything, you only get better the more you do it.

Tom: I remember in the book when Obama says, “Listen, you should have seen me way back when I really sucked.”

Terry: People forget—folks see someone like Barack Obama and say, “Oh, what a naturally gifted speaker,” which is basically telling yourself you can never be that. What you’re seeing when you see someone at the top of their game is years and decades of hard work to get there. They’ve put in the reps.

We may never sing like Taylor Swift or give a speech like Barack Obama, but we all can get better. That’s what I’ve tried to lay out in the book—ways all of us can improve.

• $2 trillion in SaaS value evaporated and Wall Street can’t decide if it’s the apocalypse or a generational buying opportunity 📉 💸 🎰

• The Daylight Principle: if a foundation model can do what your product does, you’re standing on a trapdoor 🪤 ☀️

• Sam Altman once called the Jasper CEO to give him a heads up that he was f****d — who’s getting that call next? 📞 💀

• Kana emerges from stealth with $15M to build AI agents that actually do the marketing 🤖 📣

• France bans Zoom and Teams as Europe draws a digital sovereignty line in the sand 🇫🇷 🚫 🇺🇸

• Democrats are running on AI policy in 2026 — and the White House doesn’t have an answer 🗳️ 🤖

• Modi tried to get Altman and Amodei to hold hands — it did not go well 🇮🇳 🤝❌

My Take:

In the 1880s, factory owners got access to an extraordinary new technology called electricity. They did what any rational operator would do. They pulled out the steam engine and installed an electric motor.

And then they changed almost nothing else.

The overhead shafts remained. The belts stayed strung across the ceiling. Every machine was still tethered to a central source of power. The factory floor plan still reflected the physical reality of steam: one large engine, distributing force mechanically across the building.

Electricity removed the constraint. Nobody noticed.

As the economic historian Paul David put it, managers simply “overlaid one technical system upon a preexisting stratum.”

And so productivity didn’t budge. For forty years.

Finally, in the 1920s, a new generation of managers had the obvious-in-hindsight breakthrough: put a small motor on each individual machine. Rip out the shafts and pulleys entirely. Redesign the factory floor around the actual sequence of work. Productivity exploded. The technology hadn’t changed. The willingness to abandon the old workflow had.

I keep thinking about those factory owners when I watch market coverage of the enterprise software selloff.

What the Repricing Actually Signals

Mainstream narratives have you missing the forest for the trees. The story you’re hearing: $285 billion in enterprise software value evaporated because the market is “spooked by AI.” That kind of repricing doesn’t come from fear. It comes from structural recognition.

The market caught up to something builders should have seen coming: in an agentic AI world, the value map gets redrawn. And a whole lot of enterprise software is standing on the wrong side of the line.

For decades, enterprise software has employed remarkable technology to preserve unremarkable workflows. And the thing is, the model made sense under the old constraints. Humans were the unit of work. Software existed to make human workflows marginally less painful. You charged per seat because humans sat in seats. You designed dashboards because humans needed to look at something. You built approval chains because humans needed to coordinate with other humans.

But agentic AI removes the bottleneck. And just like electricity eliminated the need for a central drive shaft, it eliminated the need for software organized around human steps. Agents don’t consume seats. They don’t need dashboards. Agents pursue outcomes. That single shift collapses the logic that propped up an entire generation of enterprise software companies.

Share

The Daylight Principle

Here’s a diagnostic that cuts through the noise: how much daylight is there between what your product does and what a foundation model can do on its own?

If the answer is “not much,” you’re in trouble. Remember Jasper? They discovered the GPT-3.5 API early, turned it into marketing copy, and everybody thought it was magic. At least until OpenAI put ChatGPT in front of everyone for free. Sam Altman famously called the Jasper CEO to give him a heads up that he was f****d.

Subscribe now

Harvey, the legal AI unicorn, is staring down the same barrel. Claude can now inhale every legal template, every NDA, every standard SaaS contract… most of which are open-sourced or publicly available anyway. Why pay hundreds of thousands of dollars for a wrapper when you can go straight to the model?

This is the question the market is asking about every enterprise software company right now. Not “do they have AI?” but “is there any daylight between what they do and what the model does natively?” If you’re just sprinting with an AI workflow on top of commodity capabilities, the model is on your heels.

Where Value Actually Lives

Infrastructure, for one. AI workloads don’t buy licenses — they generate compute cycles, API calls, and throughput. As agents scale, infrastructure vendors see rising consumption. That’s a structural tailwind, and it’s the exact inverse of the per-seat headwind hitting legacy SaaS.

Data, obviously. Agents are only as good as the data they consume. If you’re building the connective tissue between siloed data sources, or building a business that generates or captures proprietary data in its basic design, you’re in a good spot.

And the one most people miss: the human orchestration layer. Agentic systems are not a path to removing humans. They’re a fundamentally different way for humans and machines to work together. The human role shifts from executing steps to executing judgement and providing the signal that models can’t infer on their own. They become coaches, critics, and correctors of the AI system. Like: Why a recommendation was accepted. Which tradeoff mattered in that moment. What “good” looked like under imperfect conditions.

Share

The enterprises that design governance and human oversight into their agentic systems — not bolt it on after the fact — will unlock extraordinary leverage. The ones that treat AI as a way to subtract headcount from an unchanged process will find themselves optimizing a factory floor that’s still organized around steam.

The Design Shift

If you’re building for the enterprise right now, starting with automation is the wrong place to begin. Bolting AI onto legacy pricing structures is exactly as doomed as bolting AI onto legacy workflows. Both are just preserving ordinary behavior with extraordinary technology.

And meanwhile, the cost of coordination is dropping faster than most organizations can metabolize. Individual contributors are discovering that work that once required entire teams and weeks can now be done solo in hours or minutes.

Factory owners in the 1880s had the most powerful technology of their generation. They used it to preserve a workflow designed for steam. It took forty years for someone to rethink the floor.

We don’t have forty years. The market just delivered that message plainly. The question that should keep every builder up at night is this: are you building for the world that’s coming or the one that’s going?

Leave a comment

What I’m Reading:

Kana emerges from stealth with $15M to build flexible AI agents for marketers via TechCrunch 🤖 📣

Chavez and Vaidya emphasized the importance of the platform’s flexibility, arguing that the ability to deploy, tailor, and build new agents in real time would let marketers see results on their campaigns faster than they would with legacy systems.

For more, see:

France dumps Zoom and Teams as Europe seeks digital autonomy from the US 🇫🇷 🚫 🇺🇸

Around Europe, governments and institutions are seeking to reduce their use of digital services from U.S. Big Tech companies and turning to domestic or free alternatives. The push for “digital sovereignty” is gaining attention as the Trump administration strikes an increasingly belligerent posture toward the continent, highlighted by recent tensions over Greenland that intensified fears that Silicon Valley giants could be compelled to cut off access.

Democrats run on AI policy in 2026 campaigns by Ashley Gold via Axios 🗳️ 🤖

As AI's rapid advancements sound alarms even from within the ranks of AI companies, the Trump administration has embraced a very hands-off approach to the technology, and Congress has not passed major AI legislation… democrats runnings for office see an opportunity.

Modi’s AI summit turns awkward as tech leaders Sam Altman and Dario Amodei dodge contact via the AP 🇮🇳 🤝 ❌

If you’re confused about what’s happening here, might I again recommend you check out Empire of AI by Karen Hao? Review here:

Early in the life of every company I’ve built, I deliver some version of the Blueberries & Pancakes Speech. It goes like this.

A new customer comes in. They signed the contract. Now they want things. And not just the things we sold them — adjacent things, weird things, things that definitely weren’t in the spec. My team’s instinct is to point at the contract and say, “That’s not what we agreed to.”

Wrong answer.

“If the customer asks us to bring them pancakes with blueberries on Saturday at 7:30,” I tell them, “we’re gonna do it.”

They look at me like I’ve lost it. “What do blueberries or pancakes have to do with software?”

Nothing. Doesn’t matter. The point isn’t the pancakes, the point is the posture. Get obsessed with dazzling the customer, not with policing the letter of what you promised. Early on, that’s how you earn the right to keep building.

Here’s the problem. That posture is necessary in the beginning but it just doesn’t scale. If you’re too responsive — yes to every request, every edge case, every “can you just add this one thing” — you stop being a software company and start being a consultancy. The coordination costs eat you alive and your roadmap turns into a junk drawer.

Leave a comment

So what happens? You grow up. Product roadmaps become tree trunks, steady and thick like a good piece of oak. Occasionally there’s a spindle or a branch hanging off that trunk, depending entirely on how much extra a customer was willing to pay for a customization. (There’s always a price, if you’re a sensible businessperson.) But the overall thrust is to keep customers connected to the central trunk and to tame their baser instincts to deploy stuff that is a reaction to their unique business quirks but uninteresting to everybody else.

That tension — dazzle the customer vs. protect the roadmap — has defined enterprise software for as long as I’ve been building it.

But there’s a third path now. Not Build. Not Buy. Build-With.

Let me explain why Build vs. Buy existed in the first place, because that’s the key to understanding why it’s falling apart.

There’s an old concept in economics called transaction costs, first identified by Ronald Coase in the 1930s.

The idea is simple: getting exactly what you need from someone else costs more than the price tag suggests. The back-and-forth. The specs that never quite capture the thing. The coordination. The compromises. When all that friction gets high enough, you stop trying and just do it yourself.

Build vs. Buy is the enterprise software version of Coase’s problem. Every branch on the tree, every custom feature, every customer-specific tweak, comes with a conversation, a spec, a sprint, and a maintenance burden that never ends. When the friction is that high, you’re stuck: build it yourself and bleed time and money, or buy someone else’s product and learn to live with their choices. Pick your poison.

When you sign a contract with one of my companies, you’re getting what’s on the back of the truck plus access to a compelling roadmap. Pre-existing capabilities and privileged access to what’s next. Custom jobs have always been too expensive to deliver and to maintain, so I never promised them.

That’s what’s changing. AI is eliminating the pain-in-the-ass factor. If you’ve heard people talking about a “post-Coasian economy,” this is what they mean: the friction that forced the build v buy choice in the first place is falling apart.

AI can manage variation at a scale that used to be impossible. Keeping track of what Customer A needs versus Customer B. Preventing one update from breaking a dozen custom edge cases. Keeping the whole thing from flying apart when the product is being pulled in fifteen directions at once. That’s the work that made customization too expensive to touch, and it turns out AI is good at it. Define, spec, and QA complex software in weeks, not quarters, and keep the books on all of it. That used to require a team. Now it requires a stack.

So now Build-With has legs. Customers can get their pancakes without their vendor losing the plot. Product roadmaps don’t have to be tree trunks anymore. They can get spindly, branches everywhere, each one tailored, each one actually maintained, because the cost of supporting all those variations has dropped through the floor.

Customers get the sensation of choosing their own adventure, with software shaped to fit their setup instead of the other way around. And on my side of the table, I still get to run a coherent product with a clear direction. I just have better tools to do both at once.

Subscribe now

The deal in enterprise software has always gone like this: take our roadmap, learn to live with it, and in exchange you get reliability and scale. Build-With flips that.

The companies that figure out Build-With first are going to eat the ones still debating which poison to pick.

As usual, the best conversations weren’t in the keynotes. They were at the quick catch-ups in-between meetings.

I got 40 minutes with Randall Rothenberg. If you know who he is, you know why those 40 minutes mattered. If you don’t, you’re about to find out.

Here’s what stuck.

1. The Digital Audio Workstation Theory

We’re always trying to look at analog examples to help us understand digital+data transformations. I think Rothenberg nailed it with a personal story: as a musician, he was buying some equipment and the sales associate strongly recommended he buy a certain package because it came with a Digital Audio Workstation (DAW). To which Rothenberg responded, “what the hell is a digital audio workstation?”

What he discovered: a global marketplace where any plugin works with any platform, where a $100 digital instrument sounds identical to a $3 million Stradivarius, and where a $2,000 laptop produces audio that’s of higher quality than what used to come out of top-end studios in LA with racks of blinking lights and isolation booths.

The music industry went from its lowest revenues in history (2014) to its highest (2024). Rothenberg wisely observes that this is in no small part because music adopted an open, interoperable standard called MIDI (Musical Instrument Digital Interface). MIDI has provided a universal, enduring standard for the creation and transmission of digital music – a lingua franca that allows the luscious sound of a Steinway piano at Carnegie Hall to live on your laptop or your home studio for the price of a burrito at your favorite joint in the Mission.

Subscribe now

In contrast, marketing built walled gardens around a cacophony of incompatible systems, formats, and standards. We’ve spent a hundred years making it harder to align disparate data. It’s time for marketers to tame the complexity and connect data the same way modern musicians make music in a DAW.

2. The Silos Were Technology Artifacts

For the first time, with agentic AI, interoperability for the modern marketer is finally within reach. We CAN bust down the silos and stovepipes and create a seamless backplane of shared, clean, actionable data. But it requires recognizing how the silos and stovepipes came into being in the first place.

Marketing isn’t organized into silos because it’s what’s best. It’s a result of cracks and incompatibilities that system builders confronted at design time decades ago. Let’s be clear-eyed about that, so we can be clear-eyed about how agentic dissolves most of the limitations we’ve been dealing with for no good reason for too long.

Now, to Randy’s point, companies can seamlessly join call center data with advertising creative with data governance.

The constraints are gone. But the org charts remain.

Leave a comment

3. Don’t Pave The Cowpath

We’re at the beginning. People are automating simple workflows and calling them “AI workflows.” It’s impressive and fun and novel. But the real use cases—call center data informing creative data integrated with data governance—that’s not happening yet. It will happen, but it’s early.

And therein lies the trap: using extraordinary technology to preserve very ordinary behavior.

4. Nobody’s Asking About Transactions Anymore

At IAB—an advertising conference—not one client asked about advertising transactions. Zero.

Instead: “How do I understand my customers better? How do I connect these threads of data into something that helps me understand the totality of my customer relationships?”

It’s not about trading ad slots efficiently. That’s easy. It’s about understanding human beings well enough to build and sustain meaningful relationships with them.

5. Three Types of Technology Shifts

Type 1: Speed things up without changing industry structure (SWIFT in banking)

Type 2: Transform processes but don’t break boundaries (5G in telecom)

Type 3: Create entirely new things nobody even imagined (HTTP/HTML)

Where does agentic marketing sit? Randy and Vivek agree that it’s somewhere between 2 and 3. It’s possibly a full-on 3. We just don’t know yet.

The further you move from 1 to 3 in agentic marketing, the harder it becomes to weed out the human dimension – and the more uncapped the possibilities.

6. Stop Using the Word “Programmatic”

Randy wants us to stop using the word ‘programmatic,’ and I agree. The word itself is a mental prison. It constrains your mental model to matters of efficiency and buyer-seller relationships (which are inherently adversarial).

Which misses the point. Efficient trading of ad slots is easy. The real question is: how do we understand human beings well enough to catalyze and sustain durable, valuable relationships?

Language shapes thinking. “Programmatic” locks us into yesterday’s framework. It keeps us on the cowpath when we should be building new roads entirely.

7. The Readiness Question Has the Wrong Frame

When we say “marketers,” we picture the CMO at awards dinners still wrestling with how to replace television.

But tens of thousands of marketers are already in the trenches building test-and-learn scenarios with integrated data. Connecting call center data with creative with outcomes. This is happening now. It’s not a pipe dream.

Randy’s prediction: 20-25% of modern marketers will successfully navigate the transition to agentic during the next three years. Which is to say, 75% will be left behind.

The question isn’t “are marketers ready?” It’s “which marketers are you talking about?” A few are already there. Most are still on the cowpath.

8. The Great Reversal: From Big Data to Small Data (Back to Panels)

Vivek and I used to crow on stage about “billions of uniques.” Big Data was everything.

Now we’re celebrating small data—high octane, high-truth data. As Randy observes, panels may actually be better signals for personalization, targeting, measurement, and analytics than cookie pools composed of billions of pseudonymized identifiers of questionable origin.

Facebook proved this after iOS 14 killed its tracking and annihilated approximately 80% of the data they were using to fuel their ad business. More than a few pundits and market watchers predicted the end of Meta.

Meta came roaring back using synthetic data generation, not surveillance.

AI – synthetic data in particular – offers not just a legal but an ethical workaround to the PII problem.

This is what complete transformation looks like: a total reversal of assumptions around which we built the prior generation’s marketing stack.

From today’s vantage point, it’s a Soviet-era contraption crying out for reinvention. Its agentic successor isn’t a state of mind. Ready or not, it’s hurtling towards us at blinding, scary speed.

We met with an early-stage founder recently.
Smart. Thoughtful. Obsessed with the problem. Maybe we'll join up.

We were going through the regular rounds of diligence and I asked about pricing.

At this stage, I expected a range.

Instead, he came back with two decimal points, wrapped in absurd confidence that suggested his pricing schedule had been handed down via stone tablets.

And I remember thinking: there's no f*****g way he can know that.

He just didn't have enough customers, enough usage, or enough data to be that precise with that much confidence.

Now look: I have compassion for the guy. I used to be that guy.

Entrepreneurs are stuck in a bind. You have to show up with conviction while managing real uncertainty. If you speak too plainly about the uncertainty and the unknowns, a lot of VCs and board members (even the ones who profess to be comfortable with the rough-and-tumble of early stage) will run for the hills. So most founders resolve the tension the same way: by wrapping themselves in the mantle of false certainty.

I did that for years.

Old me would've launched into the routine:

"Calibrating for X, Y, and Z, and accounting for A, B, and C, it's clear that..." Blah blah blah blah blah.

A lot of words. A lot of posturing. A lot of saying what people wanted to hear so they'd stay off my back.

Then one day, in a board meeting, I got asked a question there was no possible way I could answer.

What came next wasn't my better angels swooping in. I was irritated, and I smugly blurted out:

Subscribe now

"I'm sure I don't know the answer to that question. But here's my theory of the case."

And weirdly, it landed. The room didn't collapse. If anything, the conversation got better.

Now this isn't about shrugging and saying, "Who knows, aliens could swoop down, the moon could turn into blue cheese. Your guess is as good as mine!" That's not useful either.

Here’s what it is. And feel free to repeat after me:
“I don't know. But here's my theory of the case.”

Share

Have beliefs. Have reasons. Be ready to revise them fast.

Turns out people don't need you to know the answer.

They just need to trust that you'll figure it out and are learning at the speed of light.

My comfort genre is apparently “how smart people rationalize doing damage.” Which is either intellectual rigor or a cry for help. Probably both.

Without further ado…

Empire of AI By Karen Hao

Empire of AI might look like a dense paperweight that elites of Silicon Valley use to decorate their bookshelves but it’s decidedly much more than that. In fact, this is a compulsively readable page-turner. It serves as both a primer on Artificial Intelligence and a fascinating reflection on the timeless dynamics of power—who wields it, who it corrupts, and who pays the price. It reads like a thriller as it gives the inside scoop on the early beginnings of OpenAI, the mishegoss with Musk, and the ‘Divorce’ that led to the founding of Anthropic.

By the final pages, Hao has argued the book’s title into a thesis that’s hard to ignore and much harder to refute.

Buy: Amazon // Bookshop

Careless People by Sarah Wynn-Williams

Mark Zuckerberg’s ham-fisted attempt to squash this book actually served as its unofficial publicity tour and made everybody thirstier to read it, myself included. If you’re a reader of F. Scott Fitzgerald, the title alone signals the author’s perspective: “They were careless people, Tom and Daisy—they smashed up things and creatures and then retreated back into their money.” Here’s another one that reads like fiction, thanks both to Wynn-Williams’ skill at scene-setting and the sheer insanity of the stories being told, which makes it both compulsively readable and particularly horrifying. It’s hard not to get the sense that the author is downplaying her own culpability in some of what unfolded, but you also have to give her enormous credit for standing up to the Meta machine and telling this story at all.

Buy: Amazon // Bookshop

Educated by Tara Westover

Yes, this book came out in 2018, but I’m not one to let being late to the party prevent me from attending and boy am I glad I finally showed up. As a first-gen college kid whose admission and access to elite institutions have shaped much of my life, this one was personal. I enjoyed this as a compelling underdog story, but I was fascinated because it’s actually much more than that. Westover’s relgious homeschool upbringing is definitely extreme (as it reminds me of my old-country Mexican Catholic experience), but it’s not quite the outlier it appears to be. In fact, it’s a telling window into the forces reshaping American politics today. The religious homeschooling movement that shaped Westover’s childhood in the 1980s rejected vaccines, dismissed scientific consensus, and operated outside any institutional accountability. That infrastructure has now become a pipeline into conservative government leadership, turning parental rights into a rallying cry for dismantling public institutions. What reads as one family’s dysfunction is actually a preview of the ideology now wielding power at the highest levels of government.

Subscribe now

Buy: Amazon // Bookshop

Grant by Ron Chernow

Sometimes, I wake up in the middle of the night and need to go back to sleep. I’ve counseled friends to keep a boring but important book on their Kindle for such moments.

Did I sometimes turn to this book to help me fall asleep after a long day, or in the middle of the night when I wake up and need to go back to sleep? Yes. Did I return to the pages I consumed while bleary-eyed to make sure I didn’t miss a detail? Also yes. I didn’t know much of anything about Grant before reading Chernow’s 1100 pages but it’s hard to argue with Chernow’s assertion that Grant is the most underrated president in U.S. history. What looks like ‘dad history’ turns out to be an operator’s manual for leadership under pressure:

• Grant’s talent for eliminating organizational fragmentation and getting misaligned armies pulling in the same direction

• His refusal to rely on hierarchy or formal authority, instead earning respect by being accessible and unpretentious,

• His smashing approach to warfare, always advancing, never retreating

Of course, what’s striking about Grant is that he was, by nearly all definitions, a failure for the vast majority of his life: a failed businessman, resigned from the army in disgrace, struggling alcoholic. He had no grand ambitions. He rose to the challenges of his time and surmounted his humble beginnings, never constrained by them – as in battle, always advancing, never retreating.

He found greatness almost reluctantly, when circumstances demanded it and the stakes couldn’t be higher.

Share

Buy: Amazon // Bookshop

Moscow X, Dave McCloskey

Don’t we all sort of wonder what our life as a CIA operative would have looked like? As somebody who chose a life in tech and privacy instead, Moscow X gets so much right about how surveillance actually works in the modern world. McCloskey highlights the use of cell phone data, tax records, and social media to profile and track individuals—showing just how little privacy exists when someone with resources (or a government spy apparatus) decides to look. It’s the kind of technical accuracy you’d expect from a former CIA analyst, but woven into a genuinely gripping story. Did some of it feel far-fetched? Sure. But it’s a fun spy novel to sink your teeth into, and one that understands that in 2025, intelligence work is as much about data exhaust and digital footprints as it is about dead drops and handler meetings.

Leave a comment

Buy: Amazon // Bookshop

Last Days of Night by Graham Moore

There’s a lie we all like to tell (and believe) when it comes to innovation: that brilliant new technologies come from the mind of one lone genius. Graham Moore calls bullshit. And shows us in this historical legal thriller that reality does too.

What actually moves history forward is a messy three-body problem—the visionary (Tesla), the builder (Westinghouse), and the hustler (Edison)—locked in uncomfortable tension. Set during the 1888 patent war over the light bulb, it reads like a legal thriller because it is one: young Paul Cravath (the founder of today’s law firm) defending Westinghouse in a billion-dollar IP knife fight. But zoom out and it’s Silicon Valley before hoodies—platform wars, fear-based PR, lawsuits as strategy, media manipulation to convince the public the other guy’s tech will literally kill you. Swap AC vs. DC for iOS vs. Android and you’ve got the same playbook. Moore even opens chapters with quotes from Jobs and Gates, just in case you’re dense as fuck. This story might be 137-years-old but the takeaway is as true today as it was then: Forget the lone genius myth. Innovation requires visionaries, builders, and hustlers locked in productive (and often quite messy) conflict.

Buy: Amazon // Bookshop

Curious if you’ve read any of these and what your thoughts were. Curiouser still if you have your own must-reads that I should add to my 2026 stack.

Leave a comment

• Vibe Coding : First Date :: Enterprise Software : Marriage

• How to handle non-determinism for enterprise

• The dangers of product and sales drift

• OpenAI being ordered to hand over 20 Million ChatGPT logs and the trouble with “anonymization”

• Why immigration crackdowns are playing out as badly as ::checks notes:: we all knew they would

• Pour one out for Claude, Cal Academy of Sciences beloved Albino Alligator (1995-2025)

My Take:

Here’s the thing about AI right now: everyone’s shipping press releases, not software.

Yes, a third of software leaders say they’re using AI to generate apps, and 69% claim it’s boosted productivity. But once you get past the LinkedIn glow, most of that work is happening in silos—tiny pilots scattered across the org chart, none of them talking to each other. Fragmented workflows. Optional governance. And a quiet assumption that agentic AI can be built the same way we built software ten years ago.

But Agentic AI isn’t a naming convention—it’s an architectural shift. Without rethinking how systems handle data, security, and iteration, the agents fail on impact.

Subscribe now

Enter OutSystems. They’re one of the few platforms attempting a unified, enterprise-grade low-code + AI + “agent workbench” stack, letting teams build apps and agents, manage their lifecycle, enforce DevSecOps, and plug into real data without duct-taping six tools together. It’s a credible starting point for anyone trying to turn pilots into production.

I sat down with Woodson Martin, CEO of OutSystems, to talk about what works, where the cracks still are, and how to build AI-augmented software that doesn’t collapse under its own ambition.

Tom Chavez: We first crossed paths in the Salesforce/Krux era—Cannes Lions, ExactTarget, all that. You were one of the leaders helping Salesforce step into the B2C data world. Then a few years later, you were down at the border doing hands-on work with Mobile Pathways, and my sister and I were lucky enough to support that. You’ve always toggled between hardcore enterprise tech and real-world civic engagement. Not many people manage that duality.

Woodson Martin: I appreciate that. In both business and philanthropy, I believe in getting close to the real problem—seeing it firsthand, talking to people affected by it, understanding what’s actually happening. That’s what drew me to OutSystems. Customers bring us problems they can’t solve with off-the-shelf software. Those problems force you to learn fast and stay humble. I love that.

Tom Chavez: Right now, everyone is experimenting with agents—and a decent chunk of it feels like AI theater. Lots of pilots, few production systems. From your vantage point, where do teams actually get stuck making that jump from “this is cool” to “this is real”?

Woodson Martin: Iteration is the make-or-break factor. It takes a lot of trial, error, and recalibration to figure out where agents actually add value, where you still need humans, and where the handoff between them belongs. Most companies aren’t set up to iterate at the necessary speed because their stack is spread across half a dozen disconnected tools—data here, agents there, UX over somewhere else.

When you can manage the lifecycle—data, agents, prompts, and the application layer—together, you discover the valuable use cases much faster. That’s where we see OutSystems making a difference. The platform gives teams a way to experiment coherently instead of duct-taping experiments across tools that don’t talk to each other.

Tom Chavez: Let’s talk about the non-determinism problem, because that’s where enterprise leaders start sweating. Consumers are flummoxed and frequently surprised when an LLM gives a different answer to the exact same question they asked the day before. In a regulated industry? That’s chaos. How do you design for that level of unpredictability without neutering the agent?

Share

Woodson Martin: By being explicit about where you want creativity and where you want control.

In heavily regulated sectors, you can’t have an AI improvising the same way it does in a consumer app. So we use what we call agentic workflows. You define the moments where the agent should reason, explore, and learn AND the moments where it must choose from deterministic, pre-approved actions.

Take mortgage origination. There’s a ton of document checking, policy validation, data enrichment—work that’s routine and perfect for agents. But no customer we’ve worked with wants an AI managing the applicant relationship. So they automate the back office and keep humans front and center on decisions that require judgment, empathy, or nuance.

Tom Chavez: The human piece is so underrated. Everyone’s hyped about autonomy, but the real value right now is augmentation. I use AI to prepare for customer conversations but not to replace them. That’s the good stuff.

Which brings me to vibe coding. It’s intoxicating. You describe what you want, the AI spits out a prototype, and you feel like a wizard. But for enterprises, wizardry is rarely the thing that holds up in production. How do you see vibe coding playing out?

Leave a comment

Woodson Martin: Vibe coding is an amazing accelerant for ideation. You describe an app, an agent, a workflow—and suddenly you’re looking at something real. That’s transformational for the front end of the development lifecycle.

But beneath the surface? You often get a pile of spaghetti that reflects 17 inconsistent conversations. The prototypes are great; the production path gets messy.

A friend of mine built a neighborhood security app in a weekend using AI tools. He was proud of it—and then immediately realized he’d created a maintenance nightmare. As soon as neighbors wanted changes, the whole thing wobbled.

That’s why vibe coding actually increases the need for hardened platforms. At OutSystems, our AI assistant Mentor lets you design through conversation, but the executable code is generated by a deterministic engine we’ve refined over 24 years. Same input = same output. That’s what enterprises need: magic on top, reliability underneath.

Tom Chavez: Yeah—vibe coding is a first date. Fun, exciting, full of energy. But enterprise software? That’s marriage. Architecture, clarity, structure, all the unsexy stuff that keeps the relationship alive.

You mentioned iteration earlier. At OutSystems, you don’t just celebrate iteration—you’ve made it a product feature. When people talk about AI-augmented development, they love the demos, but the truth is: value lives in the KPIs. So how do you measure whether iteration is actually delivering something meaningful?

Woodson Martin: You measure productivity. But you have to define the target before you start.

If you don’t know what success looks like, you’re just iterating randomly. Once goals are clear, we measure productivity at two layers: the developers building the systems, and the workers using them.

Take Oceaneering. Their technicians fly out to oil rigs to perform safety inspections. Every minute matters—literally. So we ask:

• How much more can a technician complete per visit?

• How much data capture can the camera and app handle automatically?

• How fast can the dev team push improvements based on field feedback?

Because OutSystems tracks the lifecycle end-to-end, we can compare productivity across releases. It gives customers a continuous optimization loop instead of one-off wins.

Tom Chavez: You’ve led product, sales, and now you’re running a company building AI applications at scale. Looking back, what’s the classic go-to-market mistake you’d erase if you had a do-over?

Woodson Martin: When product and sales drift apart.

Sales is living quarter-to-quarter. Product is running a marathon. If leadership isn’t vigilant, the messaging in the field diverges from the actual product roadmap. It’s understandable, but it’s dangerous.

Subscribe now

The antidote is alignment—heavy investment in enablement, tight communication loops, teams spending time in each other’s world. Everyone needs to be telling the same story and aiming at the same outcomes.

Tom Chavez: Preach. Any daylight between product and sales, and expectations get mis-set on all sides.

Let’s land this with a lightning round. Ready?

Woodson Martin: Let’s do it.

Tom Chavez: Advice you wish someone had given you early in the AI journey?

Woodson Martin: Don’t believe the hype. Especially not on LinkedIn. Spend more time with customers and less with vendors promising the moon.

Tom Chavez: Favorite team-building ritual?

Woodson Martin: A good old-fashioned offsite. Fire pit, shared meals, real conversation. And for planning? I still swear by Salesforce’s V2MOM framework—vision, values, methods, obstacles, measures. It aligns teams like nothing else.

Tom Chavez: Where do you go when you need to think?

Woodson Martin: Fallen Leaf Lake, just south of Tahoe. Our cabin backs onto the Desolation Wilderness. No cars, no tools, just silence. It resets my brain.

Tom Chavez: And if your younger entrepreneurial self were here, what hard truth would you hand him?

Woodson Martin: Invest in your mentors. Keep those relationships alive. They matter more—and sooner—than you think.

Tom Chavez: Perfect place to end. Woodson, this was a blast. Thanks for joining me.

Woodson Martin: Thank you, Tom.

My Stack:

OpenAI loses fight to keep ChatGPT logs secret in copyright case

OpenAI was just ordered to hand over 20 million “anonymized” ChatGPT logs to the New York Times, and the judge basically said: if your privacy tech is as good as you claim, there shouldn’t be a problem. Silicon Valley loves to swear anonymization is bulletproof—right up until a court asks them to use it. Anyone else having flashbacks to AOL in 2005, when “anonymous” search logs let reporters identify individual users in a weekend? Buckle up, I guess.

Most Immigrants Arrested in City Crackdowns Have No Criminal Record

Wildly predictable.

• Why the business question you’ve been avoiding answering is probably the most pressing 💼❓

• The power of a NO to cut through AI hype ❌🔪

• The magic of Model Context Protocol (MCP) 🪄

• Why Agentic AI is still just software 🤖💻

• How the White House is attempting to federalize surveillance under the auspices of innovation 🏛️🕵🏼

• Why “infinite exponential growth” isn’t a business model 📈💥

• How “ghost listings” are haunting jobs data 👻👔

My Take:

The Question I Didn’t Want to Answer

My Co-Founder and Head of Revenue at Kana.ai, Nick, has been hounding me the last few weeks: “What exactly do we offer? Are we selling an agent? An agentic application? What’s the thing?”

I could feel myself getting persnickety. While I appreciate precision and discipline always, Nick’s pestering was starting to feel like an exercise in false precision. Let’s close deals, and stop playing with the dictionary!

Then I had to ask myself: why was I getting so persnickety? When I interrogated my own frustration, I realized the ugly truth: I couldn’t answer the question. And if I couldn’t give Nick a straight answer, how were we supposed to explain it to customers?

Worse, I realized we were becoming part of the problem. We’re at that part of the hype cycle that feels like walking through Central Park and getting approached by the creepy guy in a trench coat trying to sell you the shit he stole from somebody else: “Pssst... you want an agent? A workflow? An application? A portal? A synthetic-data funnel? The same 2000-and-late product we’ve been selling you for the last 15 years, now with a chat interface?”

When everyone claims to do everything, everybody starts to sound like everybody else. And without a clear answer to Nick’s question, I was headed straight into that same muddy pool.

When customers can’t tell you apart from the hundred other vendors claiming to revolutionize everything with AI, they tune out.

The discipline of defining your exact offering tightly is the only way that you survive the bullshit cycle.

The Strategic Power of No

So we tried something different. Instead of claiming we do everything, we started proudly declaring what we DON’T do:

“We do not do AI for creative content generation, and we never will.” “We don’t do AI for marketing copy.” “We don’t optimize your ads—just use Google for that.”

A customer told me: “God, it’s so nice to hear you say that. Everyone says they have everything.”

That’s when I knew we were onto something.

So What DO We Actually Build?

Which brings us back to Nick’s question: agents or agentic applications?

Subscribe now

Here’s what I’ve landed on. Agents give us a tidy way to draw a circle around a problem. Don’t boil the ocean. Just give me a loyalty agent. Just give me a segmentation agent. Just give me an analytics agent.

Decomposing big, messy problems into smaller, solvable pieces—that’s what every good systems engineers have done for 40 years, long before AI. But agents are a new twist on this age-old problem. They get us off the rocks of bloated enterprise applications where customers license a gigantic pile of shit from Salesforce, SAP, and Oracle. “I don’t use all of that. I don’t want all of that. I just want this little piece.”

An agentic application? That’s two or more agents that are:

• Highly aligned around a shared objective

• Loosely coupled – they can talk to each other

This is where things get interesting. New protocols like MCP (Model Context Protocol) let agents basically introduce themselves to each other. An agent can show up and say, “Hello World, here’s what I do, I’m ready to party.” You send it data or instructions, and it reports its results back through the MCP connection. It gives each agent the ability to specialize and talk to other agents in a structured way – no free-form, agent-to-agent chats. It speeds up software deployment by tamping down the over-specification that has plagued conventional software for decades.

Why This Changes Everything (And Why It Doesn’t)

This isn’t 10% better—it’s 10x better. A business process that historically took three weeks of consulting and development now goes live in 48 hours.

Why? Because you don’t have to write it all down. You can gesture, point, grunt, and the system gets it. A little like conversing with ChatGPT: “I think I know what you mean. Let me clarify—more of this or less of that? Oh, okay, good. I got it. Let’s go.”

MCP is just this really tidy way of getting agents to share their toys.

Leave a comment

But here’s the thing: everybody needs to breathe in, breathe out. Stop mythologizing agents. I’m grateful Dario Amodei is taking the existential and cybercrime risks of superintelligence seriously, and I’ve written about the pressing need for AI regulation in other posts. But in an enterprise context, workers terrified about their pending enslavement by robot overlords need to settle the fuck down.

At the end of the day, it’s still just software on a screen.

If you’re in the enterprise world, you don’t need superintelligence. You just want software to work quicker, better, faster, stronger. Agents get us there—not through oracular magic, but by being really good at decomposition, really good at context, and really good at solving specific problems (but only if they’ve been thoughtfully specified).

The Bottom Line

The companies that win won’t be the ones claiming to do everything with AI. They’ll be the ones brave enough to say what they don’t or won’t do, smart enough to break problems into manageable pieces, and honest enough to admit it’s still just software.

Really, really good software.

My Stack:

White House prepares executive order to block state AI laws 🏛️🕵🏼

My quick read: this is the federal government and Big Tech trying to preempt 50 states from protecting their own citizens’ data... basically federalizing surveillance under the auspices of innovation. While everyone else debates “innovation vs safety,” let’s see this for what it is:

• Big Tech weaponizing federal power to kill state-level privacy protections before they spread

• The Trump administration doing Silicon Valley’s dirty work and using DOJ as their enforcement arm against states

• The race to the bottom RE: regulatory framework

Nvidia CEO quells Wall Street fears of AI bubble amid market selloff 📈💥

Nvidia crushed earnings again, and demand for their chips is still roaring. But the market is finally realizing that “infinite exponential growth” isn’t a business model.

• The demand story is real, but the pace of that demand can’t defy gravity forever

• Investors aren’t questioning GPUs, they’re questioning the second derivative: can growth keep accelerating at the same breakneck speed?

Watch Out for ‘Ghost Listings’ in the Jobs Forecast 👻👔

You only get good outputs when you put in good data. Turns out, the monthly jobs report is built on some pretty lousy data… thanks in no small part to a phenomenon that economists are calling “ghost jobs.”

Ghost jobs are roles companies post but never actually hire for. And we’re not talking about rounding errors. A new analysis shows that roughly 30% of all job postings in recent months — and even years — never resulted in an actual hire.

That means:

• Employers have been signaling demand that doesn’t exist

• JOLTS numbers are overcounting “opportunity” by millions

• Policymakers are flying blind on inflated signals

It was a meaningful election day across America: Prop 50 in California, Zohran Mamdani’s big night in New York, Spanberger’s and Sherrill’s wins back east. Plenty of headlines, plenty of heat. But maybe the most important result didn’t come from a candidate race at all. It came from the other side of the ballot in New York City, where voters quietly and overwhelmingly approved three proposals to make it faster and easier to build housing.

While most were watching the horse race, New York took aim at the real villain: bureaucracy. The tangle of red tape that’s made it nearly impossible to build homes for ordinary people in one of the world’s greatest cities. These measures streamline approvals and create new paths to get affordable projects off the ground without the decade-long obstacle course that’s become standard. In short: less red tape, more roofs.

San Francisco should take notes.

Subscribe now

Because the same thing choking New York is strangling us here. It’s not a lack of money, talent, or vision. It’s process — the endless layers of review, appeals, and politics that turn every project, from a teacher’s housing development to a corner café renovation, into a marathon of hearings and lawsuits.

Look at 469 Stevenson Street. A downtown housing project next to BART, backed by a union pension fund and city planners, was approved and then overturned after well-housed opponents raised speculative and disingenuous environmental concerns. Years later, it’s still a parking lot. That’s the cost of a system where delay is easier than decision, and every “temporary pause” becomes a permanent no.

We’ve normalized dysfunction. We call it “due diligence” or “community input,” but what we’ve really built is a system optimized to say no. In a city that prides itself on innovation, we’ve somehow made building — the most basic act of progress — the hardest thing to do.

New York’s reforms won’t fix everything overnight, but they signal something bigger: the courage to admit the system is broken and the will to cut through it. That’s exactly the conversation San Francisco needs right now. If we don’t have it, we’ll keep exporting talent to Austin and Seattle, not because people want to leave but because we’ve made staying impossible. If we want the Bay Area to keep leading in tech, in climate, in culture, we need to lead again in building.

The bad news? We’re still behind. A new state report finds San Francisco’s family-zoning plan will fall short of meeting California’s housing mandates, proof that even well-intentioned reform isn’t enough when the bureaucracy stays the same. Mayor Lurie’s proposal to open corridors across under-built neighborhoods to mid-rise, family-friendly housing is a step in the right direction. It would create homes teachers, nurses, and startup employees can actually afford. But permission on paper doesn’t put shovels in the ground.

Zoning is just one piece of the puzzle. The bigger challenge is execution. We need to pair zoning reform with long-overdue fixes to the hearings, appeals, and duplicative reviews that bog down even compliant projects. And while permitting times are trending down, it still takes about a year for a proposal to get through the process, not counting months of pre-application outreach and neighborhood politicking.

If San Francisco wants to meet the state’s goals and its own ideals, we need to modernize permitting, streamline approvals, and give city staff the tools to move projects forward instead of burying them in paperwork. We can protect the right to be heard without letting it become the right to obstruct.

So while Tuesday’s results will keep pundits busy with their favorite sport, parsing the personalities, the real lesson comes from those three little boxes on New York’s ballot. Reforming how we build isn’t flashy, but it’s the difference between cities that grow and cities that calcify. The next generation deserves better than a parking lot.

Leave a comment

• How to tell if you’ve built a product company… or just a services wrapper with an LLM sticker. 🛠️🍬

• The Sam Altman test for real moats: every time the model gets better, does your business? 🏰

• What blueberries and pancakes have to do with product-market fit. 🫐 🥞

• How to fire bad customers (without becoming a bad founder). 🪓

• What we learned naming (and renaming) a company in the age of AI. 💡

My Take:

Too many founders try to be everything to everyone: big teams, bloated roadmaps, slow, ponderous sales cycles. The winners go narrow and deep. They build with compliance in mind. They know their cost to serve, and they know exactly who writes the check—and how to win that person over.

That’s the theme of my conversation this week with Rachel ten Brink, GP and founder of Redbike Capital. Before she was an investor, she co-founded Scentbird and scaled it to 500,000+ subscribers after two decades at Estée Lauder, L’Oréal, and Gillette. She’s seen every version of “move fast and break things,” and she prefers build smart and last longer.

Rachel and I dig into what separates a true product company from a glorified services wrapper, how to design real data moats instead of buzzword flywheels, and why customer love—not fundraising—is the only early metric that matters.

If you’re a builder obsessed with focus, discipline, and depth instead of hype and sprawl, this one’s for you.

Tom: You’ve built startups as an operator and now you back founders. What’s your operator’s test for whether a “vertical AI” startup is actually a product business—or just a fancy services wrapper?

Rachel: First thing I ask is: what level of problem are you solving? How much of it can be fully automated versus handled by humans? And is it a real pain or just a nice-to-have? Because if a general model can do what you’re doing in a year or two, you don’t have a business—you have a moment.

Then I look for replicability. Can this product serve multiple customers without customization each time, or are you basically hand-crafting every deal? If 70% of what you’re doing depends on people, you’re a consulting company, not a software operation.

Subscribe now

Finally, I think about scalability and customer love. At pre-seed or seed, I’m happy with modest revenue, but I need at least one customer who’s obsessed—someone who’s sticking with you even when the product’s ugly because you’re solving something real for them.

Tom: That’s dead-on. My first company was like that—every customer needed last-mile customization. It required too much services, a lesson we carried forward into the next company and every one since.. The danger was that the more we nourished that services footprint, the further our engineers drifted from customer reality. The big lesson learned was to: keep zero daylight between builders and customer pain.

Rachel: Exactly. You can’t delegate empathy.

Tom: Let’s talk about moats. Everyone loves to say “data flywheel” or “network effect,” but very few actually have one. What’s the difference between a slide and a moat?

Rachel: The difference is timing—you can’t retrofit a moat later. The best companies design a data network effect at formation. Two ways I see it work.

First, you build a product that encourages users to share proprietary or private data because it makes the product better for them—and that creates compounding value and stickiness.

Second, you take messy public data—things anyone could access but few can use—and you process it in a way that becomes proprietary over time. We invested in a healthcare company right after the “No Surprises Act,” which forced insurers to publish out-of-network reimbursement rates. The insurers buried the data—thousands of files, impossible to parse. Our founders cleaned it up and built a searchable database. Then they went further: they noticed those same providers were entering arbitration when insurers underpaid, a process that runs like baseball arbitration—you submit a bid, the arbitrator picks one. By embedding into that workflow, they generated entirely new data about outcomes. That combination—public data + workflow + proprietary exhaust—becomes a fortress.

Share

Tom: Love that. That’s where most people miss the point. If there’s no net-new data being created, the moat evaporates.

Rachel: Right. And I always come back to what Sam Altman said: every time the underlying model gets better, does your business get more valuable—or less? If it’s less, you don’t have a moat.

Tom: Let’s get tactical. Everyone wants the magical “go-to-market motion.” What have you seen actually work for vertical AI?

Rachel: Step one is message fit. Before you do “marketing,” decide what you’re selling and to whom. Early founders love to pitch five different benefits—they’re all technically true, but the buyer only cares about one. At Scentbird, we had a dozen theories: people hate being sprayed at Macy’s, they want to try at home, they’re overwhelmed by choice. All valid, but not all equal. The discipline is choosing one.

Then, test the message fast—mix qual and quant. Talk to customers. And be honest about your current capabilities. Everyone wants to say “product-led growth,” but if your product isn’t yet self-serve, PLG just means you’re hoping. Maybe what works today is founder-led sales, conferences, LinkedIn DMs, or a handful of white-glove pilots. Pick one wedge, do it exceptionally well, and evolve from there.

Tom: Yeah, my shorthand for that is “blueberries and pancakes.” Early on, when a customer says they want pancakes with blueberries at 7:30 a.m., you make them. What does that have to do with software? Absolutely nothing – but that’s how you earn loyalty and learn the recipe. Just don’t confuse the pancake phase for the business model.

Leave a comment

Rachel: Exactly. Zero-to-one is handmade, one-to-ten is repeatable, ten-to-a-hundred is scalable. If you skip the first phase, you never get the insight; if you cling to it, you never get margins.

Tom: And then growth hits—what I call the “champagne-problem death spiral.” What breaks first?

Rachel: Product and go-to-market, almost every time.

On the product side: hallucinations, instability, over-promised accuracy. Suddenly your “AI” needs an army of humans in the loop. You thought you were selling $50-a-month software, but the hidden human labor nukes your margins. The shortcut is to ask: are people doing way more manual correction than you expected? If yes, stop onboarding for two weeks, fix the root problem, and relaunch.

On the go-to-market side: early traction makes you greedy. You start taking every customer who raises their hand. But not every customer is a good one. Some are too small to afford you; others are massive enterprises who just want to learn your tech before they build their own. You have to segment ruthlessly.

Tom: We do something similar at super{set}. Every contract starts as a two-month “Proof of Value.” It’s a real SaaS agreement, but cancelable by either side anytime in those first sixty days. It keeps us from getting trapped with customers who’ll drain the company while we’re still baking the pancakes.

Rachel: That’s brilliant—and it’s mutual. Bad customers aren’t bad people; they’re just a mismatch. Fire fast, politely.

Tom: Let’s talk about staying close to the buyer. Technical founders hate this part.

Rachel: They do. But you can’t outsource understanding. I tell them: be a field anthropologist. Run surveys, but always include open-ended questions to find the most articulate outliers—the people who’ll tell you why something hurts. Talk to those people. One of our portfolio founders just hops on planes. He’ll message a prospect and say, “Happens I’m in Chicago Monday.” He isn’t. But he closes deals because he’s in the room.

Tom: I see so many founders dwell in abstractions and use a lot of jazz hands to talk about the segments or the customers they’re talking to. And I’m always like, Look, it’s time for us to put on our safari hats, get into the jungle, pull out our recorders. What exactly did they say? Where do they eat breakfast, and where do they go for lunch? Listen closely to what they say and do, not the shiny pronouncements they share on press releases or meetings where the boss is listening in.

Rachel: Exactly. It’s all discovery.

Tom: Okay, personal time. Your résumé reads like three lives. Where did the entrepreneurial streak start?

Rachel: My dad. He left Cuba at seventeen and built a business from nothing in Costa Rica. My first “job” was doing market research for his tuna company. I’d stand in the supermarket with a clipboard watching who bought our brand and asking why. I was ten. That’s how I learned to connect products with people.

I spent about fifteen years inside giants—Gillette, Estée Lauder, L’Oréal—before realizing I wanted to build, not manage. At Elizabeth Arden I cut a deal: four days a week corporate, one day consulting for startups. Terrible idea. Founders can’t pay and don’t need advice—they need execution. But through that work I met my Scentbird co-founders. Four of us, YC S15, pure chaos and energy. We built the first subscription fragrance business, grew to 500,000+ subscribers, hit nine-figure revenue. And somewhere along the way I realized what I love most is the earliest stage: figuring out go-to-market, pricing, ICP, that messy zero-to-one. That became Redbike.

Tom: Give me the low point—the one that nearly ended you.

Rachel: Oh, easy. Early Scentbird. Our first model was “try-before-you-buy” with full-size perfumes. We shipped three bottles, let customers keep one, return two. Total disaster. People mailed back water. Shipping was expensive, cash flow worse. We had 112 visitors a day and two buyers. We were in a tiny office, four co-founders staring at each other, ready to quit.

Michael Seibel, one of our advisors, told us: you know this market; the execution is wrong. That weekend we threw everything out except the insight. One of us had found these small travel sprays at Sephora so we mocked up a subscription around those. It worked immediately. It felt small at first, but customers loved it. Revenue is the best diplomacy: once you have traction, everyone wants to work with you.

Tom: That’s a hell of a pivot story.

Rachel: It’s the same lesson I tell founders now: don’t throw out the market insight just because the execution failed. You can fix the “how.”

Tom: Perfect note to end on. Rachel, this was awesome. Thanks for bringing the operator honesty and the VC clarity.

Rachel: Thank you, Tom. Always fun to talk to someone who’s lived it.

If you made it this far, go check out Rachel’s work at redbikecapital.com and give her a follow on LinkedIn.

And as always—keep cranking, keep building.

I’m scared shitless.

Not the productive kind of fear that comes from a hard quarter or a tough competitor. I mean the deep, gut-level terror that comes from realizing the operating system you’ve spent your entire career building and using now obviates most of what you thought you knew.

Let me tell you how I know.

The Whiteboard Delusion

At Symmetri, our newest formation at super{set}, we started with a concept of end-to-end agentic marketing, recognizing that the last company Vivek and I built in this space couldn’t deliver enough surface area to define a standalone company in the current regime. Broad swaths of functionality, entire companies in olden times, now collapse into what I like to call a ‘single adorable agent.’

I woke up one night in a cold sweat. I realized that while we were attacking important challenges for the modern marketer using AI (things like segmentation, loyalty, and so on), we had missed a critical piece of the broader picture: How do consumer-facing brands deal with with reality that most consumers are now going to answer engines, not search engines, to figure out what they want?

It was a key piece in the definition of an entire category. We started noodling. Whiteboarding. Deep customer conversations. Iteration cycles. All the things that have been the backbone of every successful company formation I’ve ever been a part of.

I was heads down doing good work. The right work.

And then I looked up.

We had gone from helping to launch a new market to swimming in a sea of competitors. It happened that fucking fast.

The playbook that got me here - the careful, methodical approach to building that I’ve relied on for 25 years - had become a liability. Not because it was wrong, but because the velocity of business had increased not just 10X, but 50X, and in some places 100X.

The Month That Changed Everything

I gathered the founding team at Symmetri and I told them the truth: “The clock is not on our side. We’ve got two weeks to catch up.” We needed to create a marketer-first, purpose-built solution for what people call AEO (‘Answer Engine Optimization’) in literally two weeks.

Existential moments in a company are painful but often necessary. (If you survive, they’re the stories future employees share around the campfire.) This one was necessary.

Subscribe now

The team got the message, zero distortion. Heads down, we shipped end-to-end in two weeks. (My amazing head of engineering, Jed, if he’s reading this, will tell you it was 2.5 weeks. CEO’s like to round down.)

Here’s what I learned: When you actually internalize the urgency - when you stop performing it and start feeling it in your bones - you get done in two weeks what used to take two quarters. Not because people are working unsustainable hours (though yeah, there’s some of that), but because all the friction that seemed essential just... evaporates when survival is on the line. And the new tooling allows you to move that much faster.

That month taught me something uncomfortable: I had been the bottleneck. I needed to utterly abandon my “good instincts” about process and quality gates and proper development cycles and embrace the new normal.

What got me here was absolutely not going to get me where I needed to go.

Share

Why This Time Is Different

Max Planck once said that new scientific truths don’t win by persuasion - they triumph by outliving their opponents. Thomas Kuhn made that observation famous: paradigm shifts happen when the old guard dies off. New frameworks don’t convince the establishment; they just wait for the old guard to cozy in for their dirt nap.

Harsh, but accurate when revolutions took a generation or two.

Here’s the problem: that model is broken.

The frame of reference for software engineering, medical discovery, content creation - pick your domain - is now shifting every few months.

You don’t get to wait for the old people to die anymore. And you don’t get the comfort of being the young rebel who’ll outlive them, either. In this market, everyone is old AF - just a little sooner than some of us would’ve thought.

Think about what that means. You’re going to have to spend a lifetime adapting to one paradigm after another, long before you die. The psychological model of “master a craft, then coast on expertise” is dead. Expertise now has a half-life of months.

The 80s and 90s model was simple: get one great engineer and one great salesman. Build something that by today’s standards would be a laughably small feature set, and call it a company. It worked because the game was stable enough that mastery meant something.

Then came the Zuckerberg youth revolt. “Move fast and break things.” That was genuinely radical at the time - a rejection of the careful, measured approach. But even that revolution had a rhythm you could internalize. The rules changed, but they changed at a pace human beings could adapt to.

Now? Jessica Lessin from The Information nailed it: “OpenAI announces a major deal every 18 days - not 18 months. Investment records are shattered so fast we can’t even fit two data points, months apart, on the same chart.”

This is the water we’re swimming in.

What This Actually Means

Reality is the safest place to be. And the reality is this: what used to take a couple quarters now takes a week or two. Products launch weekly, not annually. Millions of users adopt something in days.

The survivors have internalized this new normal. Everyone else is toast.

So whatever you’re building, it’s time to triple your speed.

Not because hustle culture is virtuous (it isn’t) or because I’m a sucker for punishment (I am). Because markets could give a shit about your timetable.

Here’s the new mantra I want everyone I’m working with to internalize: What got you here won’t get you there.

There’s no time to dither. No safe harbor. No one’s coming to save you.

And this clock waits on no one.

Leave a comment

What I’m Reading/Listening to:

U.S. Chamber of Commerce sues Trump administration over $100,000 H-1B visa fee via CNBC

In a release regarding the lawsuit on Thursday, the Chamber called the new fee unlawful since it “overrides provisions of the Immigration and Nationality Act that govern the H-1B program, including the requirement that fees be based on the costs incurred by the government in processing visas.”

Renting a San Francisco Apartment in the A.I. Boom? Good Luck. via NYTimes

Landlords used to receive only a few rental applications within a month of listing an apartment, but now are getting one to three on the same day an apartment lists, said Ryan Shane, president of the Housing Guild Management Company, which manages mostly Victorian and Edwardian-style apartment buildings. “It’s much, much easier than it has been in a very long time,” he said. Neighborhoods near A.I. companies — such as Mission Bay, where OpenAI has its headquarters — are particularly popular. Rents in Mission Bay jumped 13 percent over the past year, according to CoStar. Many techies at A.I. start-ups work long hours and want to live close to their offices, said Strada’s Mr. Goodman.

• $109B went into AI last year. The real choke point now is power… literally.

• Compliance is splintering. That’s gonna be more than a headache for builders.

• Jane Goodall died this week. Lots of ink has been spilled on her legacy, not enough on what she really was: a disruptor.

• A new study out of Illinois shows how conspiracy rabbit holes end marriages.

• YouTube settled with Trump for $22M. One GIF sums up my feelings.

• Ezra Klein and Ta-Nehisi Coates clashed in print over that Charlie Kirk op-ed. When they met on the mic, the fight turned into a reckoning.

My Take:

U.S. VCs put 109 billion dollars into private AI in 2024. That is about 12 times China and 24 times the UK. The money is loud, but the real deciders now are energy, regulators, and geopolitics. I went rapid-fire with my friend, George Lee, Co-Head of the Goldman Sachs Global Institute, about how this actually plays out for builders.

Tom Chavez: The U.S. outspends China 12 to 1. The Middle East is suddenly in every AI conversation. Mirage or movement?

George Lee: On China, you gotta be careful with the headline ratio. Transparency is limited, and a lot of spend sits in adjacent buckets like robotics, drones, data, and energy infrastructure. They were ambivalent about generative AI at first, but they have pivoted. After DeepSeek, the signal is clear. The irony is a relatively closed system leaning into open source and open weights to project soft power across the Global South. On the Middle East, do not underestimate them. Abundant capital. Abundant energy. The ability to coordinate fast. Top down can be a bug and a feature when the cycle is moving this quickly.

Tom: But do the rulers let the market breathe, or is this planned economy cosplay?

George: Today it is more top down. In a fast, resource intensive wave, that can be an advantage. Whether it becomes a constraint later, we will see. This generation of rulers is ambitious and progressive about AI. They see it as the motive power of the next industrial revolution.

Tom: You use a canal and locks metaphor for the AI buildout. Not one bottleneck, a sequence. Where are we now?

George: Early it was silicon. Then advanced data center capacity. Now the pinch point is energy. AI class data centers are massive power consumers. That is the lock we are in.

Tom: Climate math is not forgiving. What is the practical fix now, not five years from now?

Subscribe now

George: Two pieces of context. First, U.S. electricity demand has been basically flat for about 20 years. Efficiency gains and a move from dispatchable to more intermittent sources did a lot of work. Now we have to scale supply and the know how to run that playbook at speed. Second, you can already see stress. One U.S. grid printed roughly 22 percent year over year wholesale price increases. Once ratepayers feel that, public utility commissions get involved. That is a policy brake, not just a technical one. Near term, think flex power. The grid is built for peak. Off peak there is slack. If AI workloads flex, curtail during peaks, and shift training and ETL to slack periods, you can harvest serious capacity as a temporal bridge while new generation catches up. A Duke analysis pegs the latent pool at roughly 75 to 125 gigawatts if we line up incentives and orchestration.

Tom: Not just theory, right?

George: Right. Google just signed agreements with utilities to curtail or shift AI load at peak in exchange for better access and pricing off peak. Expect time of day deals to become standard. And if the U.S. cannot meet the moment, expect spillover. The Middle East and Canada, plus other power abundant or faster permitting regions, will host the next wave of advanced AI data centers.

Tom: Translation for users: sometimes I might wait an extra 200 milliseconds.

George: Exactly. Users already show higher latency tolerance in chat. Old search trained us to bail if results took longer than about 0.6 seconds. Chat flipped that script. We accept a thinking beat. Training and reinforcement runs are easy to snapshot and shift. Keep inference responsive. Flex almost everything else.

Tom: I feel this in my own workflow. Fifteen years ago Google bragged about 0.27 seconds. We all got spoiled. Now I usually have three chatbot tabs open. One cranking code, one on a deep research query, one drafting. I let them cook 10 to 12 minutes. It is plate spinning at the carnival. Start one, move to the next, circle back when the first one is cooked.

George: Exactly. Humans are adapting to machine rhythm, not just the other way around.

Tom: Enterprise reality: Non determinism freaks people out. Same prompt, different answer tomorrow. Deal breaker?

George: It is coexistence. Deterministic systems will keep running transactions. Repeatable, traceable, correct. Probabilistic systems will supercharge the big area under the curve: summaries, copilots, research, ops assist. The maturity model shifts from one and done UAT to continuous sampling, testing, and tuning. Think continuous manufacturing, not discrete. Also, these models are the most human machines we have built. They inherit our foibles. Ask you the same question three days in a row, you might give three different answers. Use them where that is acceptable, and keep deterministic rails for the rest. Finance has lived with probabilistic models for decades. That muscle memory helps.

Tom: I tell friends to ask the same question again tomorrow. When the answer differs, that is not the system breaking. That is the paradigm. Build guards and workflows around it.

George: Exactly. Different answers are the point. Use guardrails and continuous QA so the variability works for you, not against you.

Tom: Regulation: Helpful guardrails, or a culture war tripwire?

Leave a comment

George: Both are possible. Risk one is a headline driven overreaction to a bad event that produces fast, blunt rules. Risk two is politicized pressure on model outputs. These systems are cultural transmitters, so governments care a lot. The White House and others are already focused on that. The whole woke AI fight is a preview. Too heavy a hand, or none, both create problems. The encouraging bit is that more regulators are learning the language of probabilistic computing and engaging constructively.

Tom: Say I want to ship worldwide. What actually trips me up?

George: Fragmentation. I once thought we were converging on harmonic standards. Not anymore. Data locality, sovereignty, and linguistic and cultural constraints are intensifying. Even giants are opting out of markets they cannot or will not comply with. Builders have to pick geographies, design for modular compliance, and accept that global on day one is mythology.

Tom: M and A: What is actually happening behind the headlines?

Share Tomisms

George: This platform shift is unusual because resource intensity and concentrated talent currently favor incumbents. Usually platform shifts hurt incumbents. This one, for now, helps them. The most visible motion is not classic roll ups yet. It is acqui hires and superstar researcher contracts at prices that used to be whole company numbers. If you are spending 100 billion dollars in capex and a small team can boost utilization a couple of points, the math can pencil. Expect strategy separation among model leaders. For example, Anthropic leaning into coding and engineering, OpenAI leaning more consumer. That divergence will inform M and A.

Tom: A lot of what you just described sounds like pro sports economics sneaking into tech. Am I reading that right?

George: That is a fair read. Superstar packages and acqui hires are filling the gap before classic consolidation.

Tom: That sportsification of engineering makes me… twitchy. Forty people in the magic room win the lottery. The other two hundred are left holding the bag. That breaks the startup social contract and messes with how families price the risk of joining early. Think the Windsor style deal. Forty in the room. Two hundred out of it.

George: The tension is real. I am sympathetic to rewarding the LeBrons, but companies are built by teams. We need mechanisms that share upside. Equity, bonuses, internal mobility. There is a good podcast on the sports culturification of engineering. Superstar comps, team dynamics, the whole thing. Reward the stars and the system that makes stars possible.

Tom: Give builders a short list for the next quarter.

George: One, design for flex. Separate latency sensitive inference from shiftable jobs. Make everything snapshotable. Two, make energy a feature. Do grid aware scheduling and build utility partnerships into your infra roadmap. Three, run dual regimes. Deterministic rails for transactions and probabilistic lanes for cognition, wrapped in continuous QA. Four, go geo modular. Localize models, data paths, and audit trails. Assume fragmented compliance. Five, mind the culture. If you chase superstars, codify broad upside so the rest of the team stays bought in.

Tom: Here is the through line. Winners will not just be model clever. They will be power literate, policy literate, and product disciplined. And they will still ship.

George: Could not agree more.

If this sharpened your map, subscribe for more straight talk on AI, markets, and the systems that actually make them work.

Subscribe now

Keep building. Keep cranking.

My Stack:

Let’s Remember Jane Goodall as the Disruptor She Was

Jane Goodall died this week. She will be remembered as a gentle, matronly icon of conservation. I suspect the level of disruption she represented in the scientific community will be conveniently overlooked. A woman in an almost exclusively male-dominated field, to be sure. But more than that, she was an outsider who threw out the rule book entirely. Arriving in Gombe without a PhD, she broke every convention: naming chimps instead of numbering them, describing their emotions, documenting tool use, hunting, and even warfare. The establishment bristled — critics accused her of being unscientific, even unserious — but the data held, and the field bent to her. Precisely because she wasn’t trained in the “proper” way, she saw what the insiders overlooked. Cambridge gave her a doctorate after she’d already blown the doors off primatology. Let that be a lesson to anybody over-investing in pedigree.

Sources: Newsweek, NatGeo

Disinformation Is Breaking Up America’s Couples

Most of us know someone — or know someone who knows someone — who fell down the QAnon rabbit hole during COVID and never climbed back out. Families lost parents, kids, entire relationships. A new study from the University of Illinois puts numbers to the anecdotes, showing how misinformation and disinformation do more than fracture our democracies, they fracture our marriages. As the researchers put it: “They failed, at least in part, because those differing beliefs were associated with different realities that disrupted a shared identity and shared reality with their partner.” Mind your media diet accordingly.

Sources: University of Illinois

YouTube Folds to Trump to the Tune of $22 Million

Sources: WSJ, Axios

• How metaphors create investment blind spots because capital flows toward familiar patterns and well-understood concepts, not alien possibilities.

• Capital doesn’t chase reality, it chases stories. You can use that to your advantage.

• A picture so mind-bogglingly wild that I thought it was an illustration ripped from The Onion. (it wasn’t.)

My Take:

Imagine London, December 1881. The Savoy Theatre has just been rebuilt, and impresario Richard D’Oyly Carte is about to open his latest Gilbert & Sullivan opera. But while the cast lingers off-stage, the real stars hang above it: hundreds of incandescent bulbs. Tonight, the Savoy will become the first public building in the world lit entirely by electricity. The packed house murmurs. Gaslight they know. Just flame in a globe. Electricity feels like bottled lightning. To calm the fear, D’Oyly Carte, ever the showman, steps forward and smashes one of the glowing bulbs. No explosion. No flame. Just glass at his feet. He assures his audience that the lamps “consume no oxygen, and cause no perceptible heat.” The crowd exhales. It is not sorcery or danger. It is simply… light.

That was the metaphor that stuck: electricity as better lighting. It helped calm fear and spur adoption, but it also obscured the far greater revolution electricity was about to unleash.

This is the paradox of metaphors. They are essential at first: they make the strange legible and the scary familiar. But once they ossify into dogma, we fail to recognize them as metaphors at all. And when that happens, capital gets trapped by yesterday’s story.

Cognitive metaphors create investment blind spots because capital flows toward familiar patterns and well-understood concepts, not alien possibilities.

In God, Human, Animal, Machine, Meghan O’Gieblyn reveals how completely metaphors capture our thinking. The brain-as-computer metaphor became so pervasive that neuroscientists literally cannot describe human behavior without it. We “process” ideas, “store” memories, “retrieve” information, forgetting these are metaphors, not biological realities. When psychologist Robert Epstein challenged researchers at a top institute to account for human behavior without computational language, they couldn’t do it.

And here’s the dangerous twist: the metaphor works both ways. We don’t just describe minds as computers. We then expect computers to have minds. We say that AI “understands,” “reasons,” and “learns” when these systems are actually doing something entirely different. The framework colonizes both directions, creating a feedback loop that blinds us to what’s actually going on.

This same cognitive trap shapes what entrepreneurs imagine and how investors allocate capital. Founders dream up what fits the frameworks, nomenclature, and terms already familiar to investors. “Uber for X” became a Silicon Valley cliché because it was a shorthand that snapped to VCs’ mental models. The internet as “stores” felt natural because everyone understood retail. Platforms, network effects, and data moats, meanwhile, were initially alien concepts that didn’t map to existing categories.

Metaphors are useful because they inspire and evoke the things we seek, the problems we care about. They grease the skids of a productive conversation when you’re not quite sure what you mean. But if we’re overtaken by them, they constrain thinking and impose boundaries around what’s possible.

Consider the evidence:

• The internet as “stores.” Investors poured billions into e-commerce fronts like Pets.com, missing the platform play entirely. Amazon wasn’t just a better bookstore, it became a marketplace, logistics machine, and cloud infrastructure pioneer. The “store” metaphor blinded VCs to everything that didn’t look like retail – or the internet version of it, e-retail. Big ups to Bezos for throwing out the preexisting framework and inventing his own.

• Cloud as “storage.” For years, the cloud was pitched as a cheaper hard drive in the sky. That metaphor guided capital toward simple file-sharing services while the real prize lay in elastic compute and infrastructure. AWS, Azure, and GCP — now worth hundreds of billions — barely resembled “storage” at all. Snowflake gives it away for free as a set up for the main attraction – compute.

• CRM as “horizontal tools.” The early frame for customer relationship management was generic software that could serve everyone. But the real breakout came from vertical specialization. Veeva built CRM tailored to pharma, with regulatory workflows and compliance baked in — and today commands a multi-billion dollar valuation. The horizontal “CRM” metaphor hid the opportunity for vertical reinvention.

• Mobile as “smaller computers.” The PC metaphor led investors to fund mobile versions of desktop software. Meanwhile, Apple built something unprecedented: the App Store economy, location-aware services, and always-connected computing that had no desktop equivalent.

• Streaming as “internet TV.” Traditional media companies saw Netflix as just another cable channel. They missed that streaming was actually a global direct-to-consumer distribution platform that collapsed geographic licensing and reshaped entertainment economics entirely.

Of course, some metaphors guide capital efficiently. Marc Andreessen’s “software eating the world” proved prescient. The “platform” metaphor, once we knew what it actually meant, helped investors spot network effects across industries. Reid Hoffman’s metaphor of startups as “jumping off a cliff and assembling a plane on the way down” captures the entrepreneurial reality better than business school frameworks ever could.

Leave a comment

But these are living metaphors that evolve with new information. Dead metaphors calcify into doctrine.

The current risk is AI as “robot butler.” We picture Alexa taking notes, ChatGPT drafting emails, copilots coding at our side. Helpful assistants. This channels investment toward chatbots and personal productivity tools. But the biggest returns will flow to orchestration layers, inference infrastructure, and decision systems that don’t resemble “butlers” at all. Companies building AI that doesn’t pretend to think like humans, but instead accomplishes things that humans alone never could.

Every calcified story becomes a broken compass.

Nearly 150 years ago, D’Oyly Carte understood the moment and his role in it. In 1881, people needed reassurance that electricity was safe illumination, not dangerous sorcery. The metaphor served its purpose: it calmed fear and enabled adoption.

But the entrepreneurs who saw past “better lighting” built the electrical grid, urban transit, mass communication, and industrial automation.

Capital doesn’t chase reality. It chases stories. The biggest arbitrage belongs to those who know when it’s time to bury a dead metaphor and create the next one.

Every great technology begins with a story. The generational companies are built by the entrepreneurs who recognize it as a story, and decide to tell a different one.

Subscribe now

Picture of the Week:

The U.S. Physics Team just swept the International Physics Olympiad. They won every single gold medal.

And how did America celebrate? By cutting science funding, banning books, telling immigrant families like the ones raising our champions to get the hell out, and giving them a commemorative photo op with a guy who thinks “windmills cause cancer.”

Reporters pointed out the obvious: these kids are of Chinese, Indian, and Russian descent. In other words, this photo is Stephen Miller’s idea of a night terror — brown and immigrant kids getting an invite to the White House with gold medals around their necks instead of a one-way ticket to Rwanda with shackles around their wrists.

Ya gotta admit, the photo is hilarious, or just absurd.

Share

I’ve said before that we’re eating the seed corn. That has never been truer. But eating the seed corn never looks catastrophic at first. The harvest looks fine. The medals shine. You can point and say, “see? we’re winning!”

But you’re NOT. You’re burning through what earlier generations invested in: immigrant families who believed in opportunity, schools that still funded science, a culture that once valued intellectual achievement.

By the time the pipeline dries up and talent goes elsewhere, it will be too late to fix.

So yeah, enjoy the photo op. It’s a commemorative shot of the last good harvest before the fields turn to dust.