Events are the Wrong Abstraction for Your AI Agents
summary
Argues that using Events as the core abstraction fragments business logic and increases latency, then proposes durable execution as the replacement: automatic state persistence, cross-process resume after crashes, and the ability to "sleep" workflows for days. Demonstrates a simpler architecture with automatic retries for LLM and tool calls, polyglot integration via Temporal's open-source runtime and its 7 language SDKs, claims a Raspberry Pi-to-space demo and that failures become inconsequential rather than needing complex event plumbing.
AI-written summary · about
chapters (5)
- 0:00The problem with events
- 2:26Event-driven AI architecture
- 4:46Issues with EDA
- 8:46Durable execution explained
- 12:45Conclusion and Q&A
related talks
32mEvent-Driven & Multi-Agent Architectures for Gen AI Workflows
24mDurable Agents: long running AI workflows in a flakey world
26mBeyond Sandboxes: Architecting Durable Runtimes for AI Agents
47mComplex event flows in distributed systems
30mHow Event Driven Architectures Go Wrong & How to Fix Them
33mFrom Prototype to Production: Making Agentic AI Systems Reliable
57mSimplify Your Architecture: Events and Publishing
54mHow Events Are Reshaping Modern Systems