Building an AI Platform Your Engineers Actually Trust
summary
One Brief builds an internal AI platform with a skills repo and a sandbox plus guardrails that plug into Claude, Linear, and the git PR workflow; a bug diagnoser becomes a bug fixer that writes PRs, enabling feedback and a measurement pipeline that replays agents against past merged fixes as an answer key. Initial clean merged fix rate is 16% and benchmarked evals improve it to 36%, sandbox cuts startup prompts from ~25 to 1, and about half the org uses the tooling.
AI-written summary · about
chapters (9)
- 0:00Making AI Real at OneBrief
- 0:49Why AI Adoption Was Hard
- 2:17Building an Internal AI Platform
- 3:03The AI Bug Diagnoser Experiment
- 3:39Small Wins That Actually Mattered
- 4:33Why We Chose Measurement Over a Knowledge Graph
- 5:12Turning a Bug Diagnoser into a Bug Fixer
- 6:14Building Evals and Improving Success Rates
- 7:04What We Learned About AI Adoption
related talks
30mBuilding Tools with Tools: Modern AI-Driven Development
26mLet the Platform Build Itself: Using AI To Construct an Internal Developer Platform With CNCF Tools
28mAI for Everything After Code
17mFrom Silos to Synergy: How Platform Engineering Bridges Teams with AI
17mAI-Driven Platform Engineering
18mUsing AI Across the Dev Lifecycle
31mModern Architecture for AI-Augmented Development Teams
26mAI Agents as Your Co-Developer