With evidence
Agents
An evaluable action loop, not a chatbot with plugins.
Simple explanation
An agent observes, chooses a tool or action, observes the result, repeats. If failure is invisible, it is not a system.
Technical explanation
state → policy → tool/environment → observation. Metrics: completion, tool-use correctness, cost, latency, hallucinations.
Evidence
Decision Copilot (investigation loop) and Automation Agent (HITL, hitl-agent.v2). No “Agents 80%” bar.