Main idea

An assistant like Claude Code works with what its user brings to it: their questions, their distinctions, their convictions, their way of framing a problem. It multiplies that state with real power — more texts, more analyses, more variants, starting from who that person is today.

It guarantees nothing beyond. A product manager who writes ten times more framing notes than before writes them with the distinctions they already had: their state is amplified, not surpassed. Moving to the next point takes something else — time, curiosity, reading, and the confrontation of what you believe with what contradicts it.

The confusion comes from the fact that amplification is immediately visible and evolution is not. The volume produced gives the impression of progress when it only measures the speed at which you exploit what you already knew.

Why it matters

It separates two curves that are often read as one: productivity and competence. Nothing links them automatically, and the first can climb for months while the second stays flat.

It's also what justifies protecting work that produces nothing immediate. Without it, the tool endlessly amplifies the same starting point.

Nuances and limits

Using an AI can itself drive progress — if you use it to be contradicted, to discover an objection or to understand a field, rather than to obtain a deliverable.

Open questions

By what sign do you recognize, from the inside, that you are amplifying instead of evolving, when both produce the same sense of fluency?