Post

What if AI memory systems look bad because we’re measuring the wrong thing? Task completion alone doesn’t tell us much anymore. Did the agent learn from past work? Repeat old mistakes? Need the same human correction again? Michael thinks our benchmarks need to change.