Wordsmiths, Rise Up
AI was supposed to make writers obsolete. They turn out to be some of its most fluent operators. Update your funnel before the smart money beats you to it.
One of the most virtuosic people I know with AI studied creative writing at Sarah Lawrence. The other is a runaway success as a Hollywood showrunner.
They’ve both been told for the past four years that AI is going to imminently replace them. There is some irony, or maybe some justice, that they’ve turned out to be some of the machines’ most fluent operators.
What most people don’t seem to understand, and The New Yorker has done magnificent reporting on this, is that the engineers building modern AI don’t really know how it works. Sure, they know there are weights inside a multi-layer neural network, conjugate gradient descent estimating Q-functions. But that’s about as useful as me knowing there’s a capacitance layer somewhere in my calculator. It’s in there. It’s helping me get the answer. Beyond that, no f***ing idea.
So what are these companies actually doing? They’re hiring wordsmiths and philosophers to coach the machines toward what we want. Lab coats outside. Wordsmiths inside.
Your best coders do not necessarily your best AI thought partners make.
The best prompting isn’t zeros and ones. It isn’t engineering. You describe a future outcome that doesn’t yet exist. You explain it in prose. You give colorful examples. You organize your thinking. You provide deep context and a narrative throughline. You tell the machine a story about what you want the world to look like, and it walks you there.
A personal note. Years ago, I went to grad school thinking: if I want to be in tech, I need to learn hardcore math. I was an okay programmer. Never great. And I said at the time, half-joking but mostly serious: if programming ever becomes gestural, if it becomes systems thinking and clear thought, I’ll be back.
Well. That’s exactly where we live. The gesture is narrative. The systems thinking is storytelling. Clear thought is the whole game.
Three years on from Karpathy’s call that English is “the hottest new programming language,” the build side has started shipping like he was right. GitHub, the company that owns code, designed its 2025 flagship developer tool, Copilot Workspace, as a prose-driven workflow: you describe the outcome in language, the system builds the spec, the plan, the code. And Anthropic, the people who make the model I’m sparring with, published a piece last September formalizing the shift: prompt engineering is now a subset of context engineering. Strip the jargon: it’s storytelling.
The hire side hasn’t budged.
Hiring managers are still screening for the engineer-shaped silhouette. Writers are still convinced they’re about to be obsolete. The market is mispricing. But the arbitrage is visible if you know where to look. An arxiv survey of what AI roles actually require found that communication and creative problem-solving together account for more than a third of the skill profile. More than AI knowledge, more than technical prompt design. The signal is there. It just hasn’t reached the headcount plan at most companies.
Just as last month I argued that virtuosic data engineers don’t necessarily make virtuosic agentic engineers, this week I’m arguing that your best coders do not necessarily your best AI thought partners make.
The people you’ve been hiring to “do AI” may not be the ones who actually wield it. And the people who can? Wordsmiths, philosophers, showrunners, the operators who think in narrative. They’re in the labor market right now, underpriced and overlooked, because nobody quite realizes their skill is the asset.
Take the leap from Sonnet to Opus. Absurd. Opus shows up as a thought partner, not an executor. I can spar with it. It smokes out nuances I hadn’t properly computed myself. It pushes back. And that only works because I’m bringing it a story. Context, stakes, characters, a throughline. Not because I’m barking commands at a vending machine.
And it’s not a vending machine. You don’t drop in a coin and get a Snickers. You build a world in language. The machine works it with you.
So. My wordsmith friends. The ones I’ve watched gnash your teeth for two years. The ones who think you’re f***ed because you don’t converse in the the language of the machines.
You speak the only language the machines actually need.
Stop hanging around the hoop. Stop gnashing your teeth. Get in the game. Start playing. Start experimenting. Understand what context engineering is. Master this. Run right into it.
The skill they called soft turns out to be load-bearing.
If you’re hiring: the skill profile shifted. Did your JD?