AI SRE agent self-heals a bad canary deploy in 112 seconds, ArgoCon demo
HowDevelop · x · 2026-11-20
At ArgoCon, a live demo showed a self-fixing canary deploy: Argo Rollouts aborted a bad canary, an AI SRE agent diagnosed the cause from live cluster state, and shipped the fix in 112 seconds through the same canary gate. Key insight: SLO alerts miss bad canaries — a 31.25% canary failure at 1% traffic only moves service-wide error rate to 0.0196%, so a 1% SLO alert never fires; the canary-scoped gate catches it, so the agent was given that same scoped signal. Setup: Argo CD + Rollouts across staging/prod clusters, failure notifications wake the agent, which works via three MCP servers — read-only Grafana (Prometheus + Loki), a release ledger, and GitHub.
More from coding & agent
- OpenAI: keep conversation prefix unchanged when using GPT-6.2 in Responses API — OpenAIDevs · 2026-11-20
- OpenAI warns prefix edits silently drop agent reasoning, ships detection tools — OpenAIDevs · 2026-11-20
- OpenAI updates Responses API so agents can build on earlier reasoning with fewer tokens — OpenAIDevs · 2026-11-20
- Semantic zoom: a pinch-to-adjust prose granularity interaction, one-shot with Gemini — illscience · 2026-11-20
- Thermal camera footage proves AI agents can program custom AI chips directly — yaroslavvb · 2026-11-20
- Critical CVSS 9.8 heap overflow in OpenSSL EVP_EncodeBlock, fixed in 4.0.4/4.1.1 — rickasaurus · 2026-11-20