One of the less-discussed effects of working with agents is how clearly they reveal the places where your own thinking is incomplete. You think you understand a task — you've done it yourself, you know what good looks like — and then you try to specify it for an agent and discover you can't. The parts that felt obvious turn out to be implicit. The judgment calls you made automatically turn out to be unjustifiable on inspection. The process you thought was clear turns out to have gaps you were filling with knowledge you didn't know you had.

This is uncomfortable and useful in roughly equal measure. Uncomfortable because it surfaces the limits of your own understanding in a way that's hard to ignore. Useful because those limits were always there — working with agents just makes them visible, which is the first step toward addressing them.

The specification problem is where this shows up most acutely. Writing a precise prompt for a task you know well requires translating tacit knowledge into explicit instructions. Tacit knowledge is the kind you can't fully articulate — the sense of when something is good enough, the feel for which edge cases matter, the judgment that makes a senior developer's work different from a junior's even when neither can fully explain why. Agents can't use tacit knowledge. They need explicit instructions. The attempt to make the tacit explicit reveals how much of your expertise lives in that inarticulate space.

The productive response is to treat agent failures as prompts for self-examination. When the agent does something wrong that you could have done right, ask: what did I know that the agent didn't? Then ask: could I have written that down? If yes, write it down — it belongs in the prompt. If no, you've found a piece of expertise that you hold intuitively but haven't yet made explicit. That's worth understanding for its own sake, separate from whether you can encode it for the agent.

Over time, working with agents makes you a more explicit thinker. The discipline of specification builds the habit of clarity. The habit of clarity makes you better at the parts of the work that don't involve agents at all.

The agent doesn't just do your thinking. It shows you where your thinking was.