The Developer's Ladder in the Age of AI and Agents

“AI can write code now—so what are programmers even for?” I’ve heard that line more than once, and lately I catch myself thinking it while I’m coding. Honestly, a few months back I went through a stretch of real panic. Features that used to take me half a day to wrestle with now come out of an Agent in minutes, and often cleaner than I’d have written them. If “being able to type out code” is itself losing value, then is everything I’ve spent years accumulating still worth anything?

Then it clicked. Code was never the value itself. It’s just the step that turns intent into a running system, and what AI has taken over is precisely the most mechanical part of that step. The real value didn’t disappear—it moved up. It shifted from “can produce it” to “knows what to write, how to steer it, and why it should be this way.” So I drew myself two ladders: one for the act of coding itself, and one that’s more general. Each rung is a gut check—can I stand one step higher than I do now?

The Coder’s Ten-Rung Ladder

1. Can ask — You can describe a requirement clearly to AI and get back code that basically works.
2. Can verify — You can tell whether the AI’s output is correct, gets it running, and stands up to tests.
3. Can decompose — You can break a complex task into small steps the AI can handle, and move it forward piece by piece.
4. Can constrain — You can write a solid CLAUDE.md or spec file so the Agent keeps producing consistent code.
5. Can architect — You can design module boundaries and data flow; AI just fills in the implementation while you set the skeleton.
6. Can orchestrate — You can run multiple Agents in parallel, merge their results, and resolve conflicts.
7. Can judge — When the AI hands you several options, you can make the right call based on business needs, performance, and maintainability.
8. Can build tools — You can write MCP Servers, custom Skills, and Hooks for an Agent, extending what the AI is capable of.
9. Can design systems — You can define an entire AI-assisted development workflow so a team of people plus Agents collaborates at peak efficiency.
10. Can define problems — You can spot problems and opportunities no one else sees; the AI becomes an extension of your will, and you decide what gets built.

Everyone’s Ten-Rung Ladder

This one isn’t just for programmers. AI doesn’t care what your job is—anyone can walk the path below.

1. Can use — You can use AI tools to get daily tasks done and offload repetitive work.
2. Can ask — You can pose good questions and pull high-quality answers and advice out of AI.
3. Can filter — You can tell what’s real and what’s good in the AI’s output, and you don’t get led astray by hallucinations.
4. Can combine — You can chain several AI tools into a workflow and multiply your output.
5. Can distill — You can turn experience into prompt templates and SOPs, so the AI reuses your judgment.
6. Can define rules — You can set roles, boundaries, and standards for AI so it runs on its own inside your framework.
7. Can connect across domains — You can use AI to get up to speed in unfamiliar fields fast, grafting knowledge from one domain onto another to create new value.
8. Can build assets — You can create reusable Agents, knowledge bases, and automated pipelines that keep producing while you sleep.
9. Can design collaboration — You can design how humans work with AI and AI works with AI, so the whole organization levels up.
10. Can define direction — You can see which problems are actually worth solving; what AI amplifies is your vision and your taste.

The Higher You Go, the More It’s You

Put the two ladders side by side and the pattern jumps out. The bottom is all execution—asking, verifying, getting things to run—and AI keeps getting better at those itself, which is exactly why the lower rungs are the easiest to replace. Move up and it turns into judgment—architecture, trade-offs, setting the rules. AI can offer options, but you’re the one who decides. At the very top, all that’s left is definition—deciding what to build and which direction to look. That’s something a machine will never hand you, because it doesn’t want anything. In a sentence: the higher you go, the more it’s you, and the harder you are to replace.

If I’m honest, right now I’m stuck somewhere between rung six and seven. I can constrain and orchestrate, and I’m comfortable running several Agents at once, but “can judge” is still rough—faced with a few options, I tend to go on gut feel rather than a real understanding of the business. The rung I want to climb toward is the tenth one, “can define problems”: not waiting for requirements to land on my desk, but seeing for myself what’s worth doing.

So I’ll leave you with a hook too. Stop here and point at these two ladders—which rung are you standing on? Then ask yourself: which rung do you want to climb to? Working out those two answers will do you far more good than learning ten more AI tools.