mohammed elshriefmohammed elshrief•

agentic qa + observability for ai agents — runs them through realistic tool / rag workflows, verifies outcomes, isolates where the workflow became unrecoverable, and turns failures into regression tests. built at wat.ai w/ composio + magic hour. catching agents when they hallucinate.

stanford's meta-harness paper had a linear loop — i mapped it onto langgraph and made it a tree. two state machines, postgres-backed checkpointing, time-travel forking, cross-run memory. self-improving agent harnesses, by construction.