Taming AI Agents With Output.Ai
summary
Argues that agent reliability is a distributed systems problem and shows an open source TypeScript framework built on Temporal called Output.ai that teams extract from production (Lovable, Webflow, Airbyte) to recover from API failures, replay runs, pause long jobs, and avoid duplicate side effects. Workflows live as folders with prompts, scenarios, tests, types, and trace files; a coding-agent compiles plain-text plans into Temporal workflows, includes cost tracking, CI-friendly eval steps, composable micro-workflows, and a Cloud Code plugin for deployment.
AI-written summary · about
chapters (10)
- 0:00Introduction and speaker intro
- 1:07Company journey and automation phases
- 10:11Challenges with drag-and-drop tools
- 15:26Problems with unchecked agent power
- 18:18Four principles for reliable AI agents
- 22:27Introducing Output.ai framework
- 28:44Workflow structure and components
- 31:31Live demo of workflow creation
- 38:51Replay.output.ai environment
- 41:17Q&A session
related talks
33mFrom Prototype to Production: Making Agentic AI Systems Reliable
24mDurable Agents: long running AI workflows in a flakey world
23mBuilding AI Agents in TypeScript
57mWhen AI Agent Teams Work (And When They Just Create More Work)
18mFrom Reactive to Resilient: Agentic AI
29mWhat exactly are AI Agents?
30mArchitecting Integrity: Responsible Software Evolution in the AI Era
28mAgentic AI