Next.js Clears Issue Backlog Using AI Research Agent and Human Review

X postNext.jsAnnouncement

The Next.js team closed nearly 1,500 GitHub issues in one month, bringing their repository backlog down from over 2,200 to under 1,000 open reports. The process was powered by 'closability', a research agent built with Vercel's open-source 'eve' framework and GPT-5.6 Luna.

Key points

  1. The vercel/next.js issue backlog peaked at 3,109 reports in early 2025 and still held 2,244 on 10 August 2026.
  2. Vercel built 'closability' on top of its open-source 'eve' framework, running investigations inside isolated sandboxes equipped with Node.js, Playwright, and Chromium.
  3. Powered by GPT-5.6 Luna with maximum reasoning effort, the workflow ran up to 200 concurrent sessions, averaging 30 minutes of investigation per issue.
  4. The agent operated read-only outside its sandbox; maintainers manually reviewed the evidence queue before closing any ticket.
  5. Primary reasons for closing were existing fixes (37%), duplicates (19%), and expected behaviour (16%); maintainer Marcos Hernanz noted only three issues have been reopened.

Why it matters

High-profile open-source repositories face an overwhelming volume of incoming issues, amplified by AI coding tools generating detailed bug reports. This implementation demonstrates how agentic workflows can shoulder the heavy lifting of reproduction and historical codebase research without stripping maintainers of oversight.

Analysis

Automated stale-bots that close tickets based solely on inactivity often discard valid bugs. Next.js resolved this by using an agent specifically tasked with gathering falsifiable evidence across canary, stable, and historical releases. With strict sandbox isolation, prompt-injection safeguards, human review queues, and a 14-day automatic reopening mechanism, the team established a balanced, responsible framework for large-scale issue triage.

Original post

Next.js

@nextjs · 4 September 2026

We closed 1,500 issues on 𝚟𝚎𝚛𝚌𝚎𝚕/𝚗𝚎𝚡𝚝.𝚓𝚜 in one month by building a 𝚌𝚕𝚘𝚜𝚊𝚋𝚒𝚕𝚒𝚝𝚢 agent with @eve. The workflow: 1. Search related work 2. Try to reproduce bugs across versions in a sandbox 3. Return context for review t.co/uuz2BYDHjf

502 likes22 replies503 bookmarks178.3k views

View on X
Selected replies (5)
  • @cbratschi @nextjs @eve Ignoring issues is also a way of treating them. Just saw a lot of issues I follow being closed as wontfix.
  • @MarcosHernanz @nextjs @eve The agent helped us find the right evidence, but every call was made by a human. If you believe we closed an issue that we shouldn't, feel free to re-open it or to shoot me a DM. Thanks @vercel for this amazing internship.
  • @philw_ @nextjs @eve I thought @MarcosHernanz had been busy! Got a bunch of closed issue emails recently 😄
  • @floydophone @jelanifuel @nextjs @eve luna max reasoning
  • @DevCalledFede @nextjs @eve 1,500 closed is 1,500 the sandbox could reproduce. The ones still open lean toward the bugs that only break on someone's real setup, and since a comment reopens an issue automatically, the reopen rate a month from now is the number I'd want sitting next to it.

Mentioned links

Links and tools from this entry

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Article:Marcos Hernanz

    How Next.js Closed 1,500 GitHub Issues Using AI Agents

    The Next.js team built autonomous research agents using Vercel's open-source framework eve to investigate over 2,200 GitHub issues. Within weeks, maintainers closed 1,462 tickets with high precision based on evidence collected by the agents.

    AI & LLMs· News

  • Video

    Video:ByteGrad

    Next.js 16.3: Runtime Debugging with next-dev-loop and agent-browser

    ByteGrad's Wesley demonstrates the next-dev-loop skill introduced in Next.js 16.3 alongside the agent-browser package. Rather than relying solely on compile or type checks, an AI agent takes control of a real browser, reproduces a visual bug in a chart component, and verifies the fix at runtime.

    AI & LLMs· Demo

  • Link:Vercel

    Eve: Open-Source Framework for Durable Backend AI Agents

    Eve is an Apache-2.0, filesystem-first framework designed for building durable backend AI agents. Currently in beta, it can be deployed on Vercel or self-hosted as a Node service on custom infrastructure.

    AI & LLMs· Tool

  • Repository:ShenSeanChen/launch-DeepResearch-Frontend

    Deep Research Agent Frontend: Next.js UI for Streaming Research Workflows

    launch-DeepResearch-Frontend is an open-source Next.js frontend designed for Deep Research Agents. It provides a chat-like interface similar to Perplexity, featuring live streaming of reasoning steps, multi-model support, and document export.

    37starsTypeScript

    AI & LLMs· Tool

  • X post:Marcel Kargul

    Marcel Kargul Open-Sources Next.js Workflow Editor

    Marcel Kargul has released the source code for his interactive workflow editor on GitHub. The application is built with Next.js 16, React 19, and Tailwind CSS 4, and was developed in a single day using AI tools for routine tasks.

    2kbookmarks87kviews

    Web development· Announcement

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.