From Issues to Commits: Autonomous Agents for Smarter Software Traceability
summary
Builds an autonomous agent called LinkAnchor that takes an issue URL and repo, uses chativity 40 nano plus a middleware with data-extractor functions (issue text, commit lists, code snippets), iterative pagination and feedback pruning to fetch only useful context, and returns a resolving commit. Evaluation finds 107/120 unseen issues linked correctly, per-issue cost ≈ $0.01 and 23s runtime, and reported improvements of ~60%–700% over six baselines, showing the agentic approach works at scale.
AI-written summary · about
chapters (14)
- 0:00Autonomous agent project intro
- 1:26What is trace link recovery
- 2:52Issue-to-commit links hard
- 3:57Limitations of ML approaches
- 8:38LinkAnchor autonomous agent
- 13:37Agentic design solves challenges
- 15:29Performance gains over baseline
- 16:55Data sources agents use
- 18:23Agent cost and efficiency
- 19:05Lessons from building AI agents
- 22:23Efficiency of agentic solutions
- 23:23Q&A data leakage
- 27:11Q&A issue creation
- 29:05Q&A commits best practices
related talks
50mPanel: Beyond The Vibe: Code Quality First
16mTackling tech debt
29mPairing with the machine: How AI Agents are rewriting the rules of software development
16mCoding Reinvented How AI Agents Are Shaping the Future of Software Development
26mFrom autocomplete to agents: AI coding assistance state of play
42mFrom Logs to Fixes: Teaching AI to Maintain Software
30mAI Agents for Incident Management
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP