Descriptions:
Cornelia Davis, developer advocate at Temporal and veteran of Cloud Foundry and distributed systems, delivers a technical conference session at AI Engineer on combining the OpenAI Agents SDK with Temporal to build agents that are reliable enough for production—not just demos.
The talk opens with a primer on the OpenAI Agents SDK (available in Python and TypeScript), explaining how it differs from basic LLM calls by giving the model agency over application flow. Davis then introduces Temporal, a durable workflow orchestration platform with roots in microservices infrastructure, and reveals that Temporal and OpenAI built a direct integration between the two products. The heart of the session is a series of live code demos showing how Temporal’s activity workers can handle the tool-invocation side of an agent loop—enabling retries, long-running operations, and reliable execution without the agent needing to manage failure modes directly. A key pattern demonstrated is dynamic tool dispatch: rather than statically registering all possible tools upfront, the system routes incoming tool-call names through a handler dictionary at runtime.
Davis is explicit about tradeoffs—when Temporal adds meaningful value versus when a simpler agent loop suffices—and addresses the gap in the OpenAI Responses API around auto-generating tool JSON schemas. The session is aimed at developers who have experimented with the OpenAI Agents SDK and are now thinking about what it takes to run agents reliably in production environments, making it a practical engineering reference for the agent infrastructure layer.
📺 Source: AI Engineer · Published January 12, 2026
🏷️ Format: Hands On Build
