AI tools reward the teams that already did the prep work
A short note on why AI coding tools widen the gap between prepared and unprepared teams, with the one writeup we keep sending people.
A pattern from the last few months of reviews, stated briefly because it does not need length.
The teams getting a real multiple out of AI coding tools are the ones whose houses were already in order: clear module boundaries, tests that mean something, documentation the tool can be pointed at. The teams getting noise are the ones hoping the tool would substitute for those things. The multiplier multiplies what is there.
The clearest short writeup we have found on this is a practitioner's notes on preparing for AI tools, from a practitioner's working notes: the observation that the assistant's output quality tracks the quality of what you can feed and check it with, so the preparation work is not overhead on the AI investment, it is the AI investment. We keep sending it to clients who ask which assistant to buy. The honest answer is usually: the one you like, after the cleanup you were deferring.