What should an AI coding agent be allowed to forget?
We talk a lot about giving AI coding agents more context.
But I’m starting to think the harder problem is deciding what should survive.
A project can contain:
an architectural decision made 6 months ago






