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
a constraint that must never be violated
an approach that was tried and failed
a temporary implementation detail
an old requirement that is no longer relevant
a conversation explaining why something was built a certain way
Should an AI agent treat all of these as “memory”?
I don't think so.
Some information should remain part of the project's long-term knowledge. Some should expire. And some should remain available only as historical context without influencing current decisions.
That makes me wonder:
If you were designing memory for an AI coding agent, what would you make permanent, what would expire, and what would you keep only as history?
I’m curious how other developers think about this.
Top comments (0)