From Blind Spots to Merged PRs: Continuous Agentic Performance Optimization
summary
Claims a production system that runs weekly agentic analyses and surfaces verified fixes that turn into merged PRs, not prototypes; shows P90 latency improvements after deployment and fixes N+1 queries and missing indexes within the first week. Describes concrete stack choices: GitHub agent workflows and Actions, Cloud Code, HUD prod-to-code mapping, ClickHouse queries and skills, runtime evidence capture (MCP) and automated test reruns to verify impact. Reports numbers: hot endpoints invoked ~7,000 times a week and a scoring system for high-ROI, low-risk fixes.
AI-written summary · about
chapters (17)
- 0:00Performance blind spots
- 1:05Performance issues persist
- 3:34Ignored optimizations leak
- 4:59Agentic workflows
- 6:21Weekly insight reports
- 7:26Automated opportunity detection
- 8:32Unverified fixes
- 9:56Production analysis context
- 11:21Prod-to-code context
- 13:05Complex production queries
- 14:29Low-hanging fixes automated
- 15:53PRs don’t solve performance
- 16:54High-ROI performance priorities
- 17:57Human-friendly reports
- 19:01Four lessons learned
- 21:30Trust in autonomous agents
- 22:11Conclusion
related talks
41mPerformance Engineering in the Age of DevOps and Agile
26mLet Them Eat Bugs: Practical Showcase of Agentic Issue Resolution
37mAI-Driven Developer Productivity at Scale
21mRemote Agents, or: How I Fixed Our Page's Performance Without Really Doing Anything
36mAgentic CI/CD: from Pull Request to Production without the Paper Cuts
23mThe Self-Healing Applications Decade is Here
42mRevisit Old Problems with New Eyes!
42m30 to 70 PRs a day: How we managed to not wreck our systems