Cafe(s) and the Context You Can't Write Down
- Source
- metalbear.com
- Author
- ioanarebeca
- Date

Context
CAFE(S) is a framework, published by a group of researchers that includes DORA co-creator Nicole Forsgren, for judging the handed to a coding . MetalBear's article applying it argues that agent context has two kinds of source: decisions and constraints that only people know need documentation, while current service responses and integration behavior can be observed directly.
MetalBear presents its mirrord development tool as a way to provide that second kind of context. The code an agent is working on runs against the real services in a staging cluster, so the agent sees actual responses instead of relying on documents that need ongoing maintenance. This is a vendor's engineering rationale, rather than a measured comparison of agent success rates.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…
