What’s Your Word?

Tell me if this describes your area of expertise.

• Your work follows specific rules and conventions.

• You produce output — documents, deliverables, decisions.

• That output has to be evaluated for quality.

• The work product is, at its core, words and symbols arranged on a screen.

• Domain expertise is what separates adequate from excellent.

If you're a lawyer, you said yes. HR? Yes. Marketing, sales, finance, consulting? Yes, yes, yes, yes. (Seriously — try to find a knowledge worker whose job this doesn't describe.)

Now here's the turn.

If you're a software developer, you also said yes.

Only one of those fields is meaningfully far along in the AI revolution. And everyone in the other fields looks at what's happening to code and says, "Yeah, but that's not my industry."

I think that's exactly wrong.

What Most People Missed

Andrej Karpathy — co-founder of OpenAI, former head of AI at Tesla, one of the most technically credible people on the planet — said on the No Priors podcast in March 2026 that he hasn't typed a line of code since December. Not a single line. In one month, he went from writing 80% of his own code to writing almost none of it. He described the shift as "an extremely large change" and said he doesn't think most people realize how dramatic it was.

(That's not a LinkedIn influencer. That's the person who essentially invented the modern AI training workflow.)

If you're not a developer, you probably read that and shrugged. "Good for him — not my world."

He wasn't just talking about code. He was talking about you.

The Substitution Test

Here's the mechanism. Call it the Substitution Test.

Anywhere you see the word "code" in a sentence about AI, replace it with your industry’s work product, henceforth known as “your word.” Brief. Contract. Job description. Performance review. Email. Proposal. If the sentence still makes sense, the disruption applies to you.

Try it:

• "AI is now writing roughly 27% of production code." → "AI is now writing roughly 27% of production briefs." (And that's just what's measured in production code — developers self-report the real number is closer to 50%.) Does that still feel like someone else's problem?

• "Developers are shifting from writing code to reviewing AI-generated code." → "Lawyers are shifting from writing briefs to reviewing AI-generated briefs."

• "I haven't written a line of code since December." → "I haven't written a line of _____ since December."

Fill in that blank with your word. Now sit with it for a second.

This isn't a rhetorical trick. It works because code is not some magical arcane language only initiates can speak. Code is, at its core, alphanumeric symbols arranged on a screen that carry encoded meaning. Rules. Logic. Instructions. Output. Sound familiar? It should — because that's also what a legal brief is. And a sales contract. And a performance review. And an email. Code is one member of a much larger family, and that family includes your work.

Which means: what's happening in software isn't a story about software. It's a preview.

The Bellwether

Simon Willison — one of the sharpest observers at the intersection of AI and practical knowledge work — recently described software developers as "a bellwether for other information workers." (Lenny's Podcast, April 2, 2026.)

A bellwether is the sheep at the front of the flock. It doesn't know it's leading. It's just moving.

Developers became the bellwether for one specific reason: code is verifiable. You can run it. Either it works or it doesn't. The feedback loop is brutally honest. That made software the perfect environment for AI to move fast — and for humans to start figuring out what it means to work alongside it.

Here's Willison's crucial observation: code is easier to verify than almost every other form of knowledge work. If AI writes a function that doesn't work, the computer tells you immediately. If it drafts a legal argument or a negotiation strategy? You might not find out it was wrong until you're sitting across the table from someone. (Which, by the way, is exactly why your domain expertise is going to matter more in the AI era, not less.)

Coders went first not because they're special. They went first because their work was the easiest target.

"The future is already here — it's just not evenly distributed." — William Gibson

Your field is in the distribution queue. You just have the enormous advantage of watching somebody else go first.

What the Preview Looks Like

A company called StrongDM is doing something worth pausing on. They build security software — serious, high-stakes stuff, not the kind of company you'd expect at the extreme edge of AI-assisted development. And yet they've introduced two policies that sound, on first pass, completely insane.

Policy one: nobody writes code. It must be written by AI.

Policy two: nobody reads the code.

(At a security company. Let that sit for a second.)

If no one's writing and no one's reading, how does anyone know the software is any good? Their answer: a swarm of AI agents simulating employees in a simulated Slack channel, making requests around the clock — "Hey, could someone give me access to Jira?" — stress-testing the software 24 hours a day at a cost of roughly $10,000 a day in compute. (That's not a typo. Ten thousand dollars a day in tokens — and they consider it a bargain compared to a manual QA team that sleeps.) No human QA. No code review. Just a permanent, relentless simulation of reality. They even built their own simulated versions of Slack, Jira, and Okta — because the real platforms have rate limits that couldn't handle 10,000 fake employees at a time.

Wild. And it worked.

Now run the Substitution Test.

What does a StrongDM look like for law? Simulated opposing counsel, stress-testing your arguments around the clock. For sales? Simulated buyers surfacing objections you never anticipated before a human eyes the proposal. For negotiations? An AI counterparty that's already practiced against a thousand versions of your offer before you've had your morning coffee.

This isn't science fiction. This is what's already happening in code, right now, at a legitimate security company. And code went first because it was easiest. The rest of us are in the queue.

What to Do This Week

Two things, not ten.

1. Run the Substitution Test. Find three headlines this week about AI and code. Replace "code" with your domain term. Notice what happens in your gut — specifically the moment it goes from "cool tech story" to "wait, that's my job."

2. Start building the muscle now. The people who will thrive in your field over the next five years aren't the ones who waited for their field's "Karpathy moment." They're the ones who started early, when the gap between "I use AI" and "I work with AI" was still closable. That gap is closable right now, today, in your field. Not because AI is coming for you — but because the people who learn to work with it are going to be able to do things that sound, to everyone else, like they've been given an unfair advantage. (And yes, I chose those words deliberately.)

What's Your Word?

You started this post by saying yes. Yes, my work follows rules. Yes, I produce output. Yes, it gets evaluated. Yes, it lives on a screen as words and symbols.

You said yes because it's true.

Now finish the sentence: "I don't think I've written a single line of _____ since December."

What's your word?

The distance between that sentence feeling impossible and that sentence feeling inevitable — that's the gap. It's not a gap to be afraid of. It's a gap to close. And the extraordinary thing is: you can see exactly what's coming, because someone else is already living it.

The future is already here. It just hasn't reached your inbox yet.

Join over 34,147 creators & leaders who read Methods of the Masters each week.

Related: It's a Skill, Not a Pill
Related: From Authorship to Stewardship