Beyond Request/Response: What AI Agents Actually Need From Your Event Streams
summary
Agents subscribe to AsyncAPI v3 descriptions that map Kafka topics to Kafka-backed WebSocket channels and carry OAuth2 client‑credentials scopes per channel; the spec gives schema contracts but does not enforce them, so an event gateway must map addresses to scopes and an AI gateway must proxy LLM calls. Real incidents show LLM bills jumping (examples: $80→$400 overnight, $1,200 over a weekend). Governance combines per‑agent credentials, gateway‑enforced scoping and throttles, local filtering/batching, and monitoring of dollars per subscription to close discovery, access, and observability gaps.
AI-written summary · about
chapters (18)
- 0:00AI agents and event streams
- 1:13Ungoverned AI agent consumption
- 3:20Event stream readiness checklist
- 4:46Documentation discovery gap
- 5:48Agent authentication scoping
- 6:50Agent cost tracking observability
- 7:52Closing governance with AsyncAPI
- 9:16AsyncAPI contract for order events
- 13:09Scopes via AsyncAPI operations
- 15:57AsyncAPI specification limits
- 16:38Access control governance
- 18:27Cost governance with LLM budgets
- 20:11Observability metrics for agents
- 20:52Agent interaction scenario
- 22:40AsyncAPI implementation design
- 24:23Runtime agent policy enforcement
- 26:33Summary
- 27:15Q&A
related talks
27mChallenges of Ambient Agents: Building Governable AI Systems at Scale
19mAPI Management for AI: Governing Ingress and Egress Traffic.
45mRun, Secure, and Govern Your AI Traffic with an AI Gateway for Multi-LLM
19mAI Agents for Incidents: The Good, the Bad, and the Ugly
26mBeyond the Gateway: Fine-Grained Authorization for AI Agents
29mHow Auth0 Secures AI Authentication & Authorization for AI Agents
25mWho Let the Agents Out? Securing AI Workflows the Right Way
40mIn AI We Trust? Securing the Future, One Agent at a Time