The VEP Police: An AI Agent That Governs an Open-Source Project
summary
An AI agent built with LangGraph state machine, MCP and LLMs autonomously discovers and monitors enhancement proposals across GitHub, performing parallel checks for deadlines, activity, compliance and cross-domain risk, then posts prioritized alerts to Slack/email and updates four columns on the release project board. It scales a small-maintainer team from a surprise growth to 41 VEPs (roughly 100 PRs), uses a two-tier LLM strategy with deterministic code for routine checks, snapshots for debugging, and catches real governance issues in real time.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 2:22Chaos of unstructured contributions
- 4:41VEP process for governance
- 6:26Growth and challenges
- 9:15Building AI agent
- 13:36VEP Police architecture
- 15:02Lessons and LLM limits
- 18:41Debugging agent decisions
- 21:59Testing and improvement
- 22:20Future plans
- 25:11Agent rules in open-source
- 27:20Final thoughts
- 28:03Q&A
related talks
50mPanel: Beyond The Vibe: Code Quality First
12mHow to maintain a codebase – when everyone can code
33mGoverning the Ungovernable: Security and Compliance for AI Agents in Open Source Projec...
01h 02mPlatform Engineering in the age of Generative AI
22mGepAI in Development Processes and Dev Tools
35mThe Agentic AI Engineer
41mImprove the Management of Open Source Community with GPT4
11mFrom Prompt to PR: Vibe Coding Meets AI Code Review