How Temporal used one spec and agents to find hundreds of SDK integration defects

francesc · x · 2026-11-19

At PortCon, Temporal's Francesc Campoy Flores described scaling integration testing with agents. Temporal ships durable-execution plugins for OpenAI Agents, Google ADK, LangGraph, Strands and Pydantic AI. They wrote a single precise spec covering shared questions, letting coding agents generate runnable reproductions per plugin (many injecting deliberate failures). Basics held everywhere; agents found hundreds of defects, mostly in specific setups. Fixes landed within weeks, and lessons from one plugin became checks for the rest. Pitfalls included rules agents couldn't see and a review loop burning absurd tokens on one PR.

Original post →

More from coding & agent

coding & agent channel →