From Prototype to Production: Making Agentic AI Systems Reliable
summary
Argues that Temporal makes agentic AI production-ready by turning LLM + tool loops into durable workflows with durable execution, persisted state in a backing database, Helm/Kubernetes deployments, and worker replay that injects prior results so failed runs resume without reissuing tokens. Shows 10 parallel web searches merged, an activity that fails four times then succeeds under automatic retries (exponential backoff, 30s max), integrates with agent SDKs, supports seven languages, and scales to hundreds of thousands of workflows per second via replayable history.
AI-written summary · about
chapters (8)
- 0:00Introduction to agentic systems
- 2:33Levels of AI adoption
- 7:12Challenges in agentic AI
- 11:47Why Temporal for durable execution
- 13:15Temporal architecture overview
- 19:54Temporal SDK and features
- 24:52Proven scale and integrations
- 27:19Q&A session
related talks
45mTaming AI Agents With Output.Ai
24mDurable Agents: long running AI workflows in a flakey world
25mAgents: Building Production AI Agents with Temporal
14mTemporal, OpenAI, and the Future of Agentic AI
26mBeyond Sandboxes: Architecting Durable Runtimes for AI Agents
52mDesigning Production-Ready Agentic Systems: Lessons from the Field
01h 00mReliable Agentic Systems need Durable Execution 🤖
39mFrom Demo to Production: Why Agentic AI Systems Fail - and How to Fix Them