From developer to AI expert: which skills survive

There was one moment when it got uncomfortable. I'd let Claude Code build a feature — clean, in minutes, better structured than my first draft would have been. And instead of euphoria, up came the question a lot of people don't dare ask out loud right now:

If the agents do the boring work better than I do — what's left of me?

I have twenty years of software craft behind me. And it was still a genuine tipping point: go with it or step out. I went with it — but not the way I expected. The move from developer to AI systems expert isn't a training course. It's a shift in which skills still count at all.

The mistake: "I need to learn to prompt faster"

Most people meet the AI wave with the wrong reflex. They try to get better at operating the tool — prompt tricks, the newest model, the clever context hack. Useful, but perishable. Prompting techniques age out with every model generation.

The question isn't how to operate the tool better. It's which of your abilities gain value when code becomes cheap — and which lose it.

Skills that depreciate

Be honest with yourself. These things were worth real money for years, and are worth less every month:

• Typing boilerplate and CRUD from memory. The agent is faster, every day.
• Knowing syntax and API signatures by heart. That was never competence, just repetition — and the machine took it over.
• Being the fastest typist on the team. Writing speed stopped being the bottleneck.
• Framework trivia. "What was that parameter called again?" is a solved question.

This stings, because those exact things felt like skill for a long time. They were only partly. Mostly they were friction — and AI melts friction away.

Skills that appreciate

And here it gets interesting. The same wave that devalues the routine makes other abilities more valuable — because suddenly everyone can produce code, but almost no one can own it:

1. Specifying, not prompting. A precise, well-reasoned brief is the new core artifact. Someone who can decompose a problem cleanly steers the agent; someone who only wishes corrects it ten times.
2. Judgment at volume. You now read more code than you write. The ability to distrust generated code and convict it is the new primary discipline — I've written separately on why code review has to change.
3. Thinking in systems, not files. The agent sees the local slice. You see the architecture, the data flows, the layer boundaries. That can't be prompted away.
4. Translating between business and technology. What is this thing actually supposed to do, for whom, and how do you know it worked? No model asks that question — and it decides whether you end up with a product or just code.

Notice something? None of these are new skills. It's classic software craft — just freed from the typing that hid it for years. That's what I mean by the bridge: the last 20% was always the real job. AI just makes it visible now.

From producer to owner

The role change fits in one sentence: your value moves from producing code to owning systems.

That's not a downgrade. It's the promotion a lot of developers waited years for — just under a new banner. You become less the craftsman hammering every nail yourself, and more the architect telling a very fast, very literal team of agents what to build and whether it holds. That's the AI systems expert: not a prompt operator, but someone who lets AI speed in without losing control of the system.

What you can do this week

No course, no certification. One experiment:

Take your next task and write the spec before you fire up the agent. Inputs, edge cases, expected failure behaviour, the tests that must prove it — all of it before the first prompt. Then measure two things: how many corrective prompts you save, and how much better the first draft is.

You'll notice the work that matters happens before and after the generation. That's exactly where the AI systems expert grows.

This whole journey — from the uneasy tipping point to the new sense of the role — is what I unpacked in my audiobook From Developer to AI Systems Expert. If the topic grabbed you, you'll find it here.