Next.js Team Deploys AI Agent to Slash GitHub Backlog by 56%
The Next.js team at Vercel has shared a remarkable engineering story: they built an AI-powered agent called closability that helped them close 1,462 GitHub issues in under a month, bringing the open issue backlog from 2,244 down to just 995. Published on September 4, 2026, this is one of the most compelling real-world examples of AI agents being used in open source maintenance.
The Problem: A Backlog That Kept Growing
With millions of developers using Next.js, the GitHub issue tracker receives an average of 36 new reports every week. The backlog peaked at 3,109 open reports in January 2025. By August 10, 2026, it still sat at 2,244 with fixed bugs, duplicates, and outdated reports burying current regressions. Simple inactivity-based auto-closing was not working; a timestamp cannot distinguish a real bug from a resolved one.
The Solution: An AI Research Agent
The team built the closability agent on top of eve, Vercel’s open-source agent framework. Running in isolated sandboxes with the full Next.js repo, Node.js, Playwright, and Chromium, the agent:
- Reads the full GitHub conversation and checks supported versions
- Searches related issues, pull requests, commits, releases, and documentation
- Attempts to reproduce the bug on the reported version, latest stable, and canary
- Looks for contradicting evidence before making a recommendation
- Returns a structured confidence score and primary reason for closure
Key Results
The agent ran against the full backlog on GPT-5.6 Luna with reasoning effort set to max. With up to 200 concurrent eve sessions, investigations averaged 30 minutes each. Here is how the 1,462 closed issues broke down:
- Already fixed: 543 issues (37%)
- Duplicate: 278 issues (19%)
- Expected behavior: 237 issues (16%)
- No longer reproducible: 89 issues (6%)
- Unsupported or obsolete: 66 issues (5%)
- Other: 249 issues (17%)
Of the 1,462 issues closed, only 3 were reopened – a 99.8% accuracy rate.
Safety First: Human Review Remained Central
Critically, the agent was kept read-only outside its sandbox. It could research but not act – no commenting, closing, pushing code, or deploying. Every closure decision was reviewed by a maintainer. The team also added a GitHub Action allowing anyone to request a reopening within 14 days if a closure was incorrect.
What’s Next: Ongoing Automation
The one-time backlog clear was just the beginning. Every Monday, closability now reviews up to 100 issues with 30+ days of inactivity. For issues scoring 80+ confidence, a second agent double-checks before auto-closing up to 25 issues per week – still with human oversight on all code changes.
Why This Matters for Developers
This story demonstrates a practical, responsible model for using AI agents in open source maintenance – one that keeps humans in the loop while dramatically scaling the team’s capacity. As AI coding tools generate more detailed bug reports, the volume of issues across all major open source projects will only grow. The Next.js approach could become a template for the industry.
Key Takeaway: The Next.js team reduced their GitHub backlog by 56% in under a month using a purpose-built AI research agent with a 99.8% accuracy rate on closure decisions.
Hashtags: #NextJS #AI #OpenSource #WebDevelopment #GitHub #Vercel #JavaScript #ReactJS #DeveloperTools
Source: Next.js Blog – How we closed 1,500 GitHub issues in one month