Every codebase has work that everyone knows should be done and nobody does. The tests that would catch that edge case but take an hour to write. The documentation that's six months out of date. The error messages that still say "something went wrong." The refactoring that would make the module cleaner but isn't blocking anything. This is the work that accumulates quietly and makes a codebase harder to work with over time.

AI assistants change the economics of this work dramatically. The hour it takes to write a good test suite for a module drops to fifteen minutes when you can describe what needs to be tested and have the scaffolding generated. Documentation that nobody writes because it's tedious to maintain can be generated from the code and reviewed rather than authored from scratch. The refactoring that felt like a weekend project becomes an afternoon when the mechanical parts are handled by the assistant.

The leverage is highest precisely where human resistance is highest — repetitive, tedious, important-but-not-urgent work. The assistant doesn't find it tedious. It doesn't have a backlog of more interesting work competing for its attention. It will write the fifteenth test case with the same care as the first.

The discipline is redirecting some of the velocity the assistant provides back into the work you've been deferring, rather than using all of it to go faster on the work that was already getting done. A team that ships features faster is good. A team that ships features faster and reduces its technical debt simultaneously is better — and that's achievable when the tedious work stops being the bottleneck.

Use the time the assistant saves on the work you never had time for. That's where the compounding value is.