Agent Observability and Profiling
Series · 1 posts
-
AgentPProf: Profiling Long-Horizon Agent Trajectories Across Runs
Advanced Agent runtime, safety, and evaluationAgentPProf maps agent prompts, tool actions, and system effects into semantic operation stacks that can be aggregated across sessions. The paper reports 0.764 B³ F1 and MAP gains on three localization benchmarks; a flame graph reveals concentrations, not causal proof.
Understand it in 90 seconds
For speaking invitations, internal engineering sessions, or architecture exchange, see the topics and public work I can bring into the conversation.
Speaking & contact