专题
Agent design
Planning loops, tool contracts, stop conditions, and human supervision for autonomous workflows.
博客
3 implementation notes in this track.
GovernanceAgent designagent
Evidence packets for human approval
Approval flows work when reviewers get concise context, not raw transcripts.
6 min read
Agentic systemsAgent designagent
Designing tool contracts for reliable agent actions
Typed tool interfaces reduce hallucinated actions and make retries safe.
7 min read
Agentic systemsAgent designagent
Why agentic systems need escalation design
Autonomy without escalation policy turns small errors into operational incidents.
8 min read
知识库
2 references in this track.