Rising · Manual
AI Agents
Builder and research activity is accelerating faster than evaluation evidence.
Why this is here
Agents are the Lab’s Data → systems bridge. SignalLab gives per-source volumes and the collect vocabulary. Status (Rising) stays a dated human read.
What I am watching
Evaluable loops (tool → observation → visible failure) versus relabelled chatbots. Tool protocols (MCP) beside a homegrown FastAPI workflow. Eval suites: completion, tool-use, cost, hallucinations.
Current read
Rising in discourse and public repos. Lab evidence: Decision Copilot, Automation Agent (hitl-agent.v2, a single agent) and the profiling A/B/C table.