Fast Doesn’t Mean Fragile: Delivering AI-Powered Software at Scale
summary
CircleCI runs 7 million builds/month across ~200 engineers and uses DORA metrics to quantify delivery: a sample elite profile at three deploys/day, 45 minute lead time, 10% change failure rate and 30 minute MTTR. Introducing AI raises deploys deploy frequency ×10 and — holding other stats constant — drives ~90 minutes of daily brokenness. Practical fixes shown: a code quality agent that auto-finds and fixes flaky tests, dependency-aware test selection to only run impacted suites, and pushing fast security, linting and test feedback into the IDE/inner loop.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:54Subaru Outback analogy
- 5:08DORA metrics explained
- 12:37AI risks in delivery
- 16:15AI agents reduce drag
- 18:24Faster test suites
- 22:18Inner and outer loops
- 26:11Vibe coding limits
- 28:18Collapsing loops with AI
- 30:26AI code generation layers
- 33:57Redefining design for AI
- 37:29Summary
- 38:10Q&A
- 39:13Zero UI vs traditional
- 41:02Technical debt management
- 44:09AI tools adoption timeline
- 46:57GitHub Actions impact
- 48:46Closing remarks
related talks
37mCI/CD Observability, Metrics and DORA: Shifting Left and Cleaning Up!
33mHigh performing engineering teams and the Holy Grail
29mHow to use AI to fight your DevOps transformation!
32mFast, Smart, and Fragile: Rethinking Testing Leadership in the Age of AI
25mContinuous Delivery, high-performing engineering teams, and the Holy Grail
27mThe Need for Speed (in your CI/CD setup)
26mYou can write code faster. Can you deliver it faster?
24mUnlocking CI Performance in the Age of Agentic Development