Comment by swatcoder

You might want to think about how that technique scales as your decisions.log accumulates more and more guidance.

Or at least maybe ask Claude what will happen when the md and log files that keep it on task start to dominate and someday even overflow its context windows.

Using AI doesn't protect you from thinking about resource constraints and algorithm tradeoffs like any other engineer might. It's just that the resource constraints and algorithms tradeoffs that you need to engineer around become those of the AI tooling rather than the project its generating.