TLDRNew
Why AI Needs Evals, Not Vibes
Takeaway
An eval runs your agent on the same tasks several times and checks every result, so you can compare the pass rate before and after a change.
More TLDRs
The Agent Doesn't Need to Know EverythingNew
The agent reads AGENTS.md on every task, so keep only what it can't find in the code, and move notes that only some tasks need into their own files.
Agents That Use Software Like You DoNew
Computer-use agents click, type and navigate software like a person, which makes them useful and risky to hand real tasks.
What Is Agent Orchestration?New
AI agent orchestration enables multiple agents to collaborate, delegate tasks, share context, and coordinate complex workflows to achieve better results.