DriftlessAF - An open source agentic reconciler framework proven at scale
summary
Open source reconciler framework Driftless.AF (Apache license) runs in Go with Terraform recipes for Google Cloud, a GCS-backed work queue, and a meta-agent layer using Gemini, Claude, OpenAI for LLM orchestration and a judge LLM for evals. It processes GitHub PRs and issues, auto-fixes titles, descriptions and code, manages cloned repos, and drives Chainguard Factory builds at scale: over 2,500 container images, 600M builds, 240K versions. Deployments produce rapid results — 900 test-improvement PRs in a day and ~2,100 Python fixer PRs (≈$27K token spend).
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:05Agentic reconciler framework
- 3:56Why we built DriftlessAF
- 5:20Core components
- 7:26Agentic AI infrastructure
- 11:00GitHub reconcilers
- 14:32Demo in action
- 18:23Chainguard Factory
- 21:33Scale and impact
- 22:59Results and improvements
- 23:21Future possibilities
- 24:46Resources
related talks
44mAgentic Delivery, Guardrailed: AI in CI/CD & Platform Engineering (Open Source Patterns)
32mFrom Toil To Triumph: Harnessing Agentic AI To Streamline Infrastructure as Code
01h 16mAutomating Large Scale Refactors with Parallel Agents
28mArchitecting AI-DLC for the organization
01h 23mOpening
26mOrchestrating AI Agents for Large-Scale Refactoring
19mNo Token Budget: Agentic Software Factories on Kubernetes
27mIntegrating Agentic AI Into Internal Developer Platforms To Minimize Errors and Downtime