--- title: Software factories description: Build systems of agents that investigate issues, prepare code changes, and hand reviewers the evidence they need. This stack builds on the Agent stack. Learn from working factories, start from a template, then choose how to run, review, and improve the work. url: "https://vercel.com/kb/architecture/software-factories" last_updated: 2026-11-02 install_vercel_plugin: npx plugins add vercel/vercel-plugin --- # Software factories Build systems of agents that investigate issues, prepare code changes, and hand reviewers the evidence they need. This stack builds on the Agent stack. Learn from working factories, start from a template, then choose how to run, review, and improve the work. ## Start from a working factory See how Vercel teams split repository work between agents and people, then start from a complete factory or one focused agent. ### Decide what to automate Compare a factory that ships code changes with one that researches a backlog, then judge agent-written changes yourself. - [Building a software factory for AI SDK](https://vercel.com/blog/building-a-software-factory-for-ai-sdk): See how the AI SDK team's agents classify, implement, review, and backport changes in sandboxes, with a human merging every PR. - [How we closed 1,500 GitHub issues in one month](https://nextjs.org/blog/how-we-closed-1500-github-issues): Learn how the Next.js team ran 200 parallel read-only eve investigations and used the evidence to decide what to close. - [Agent responsibly](https://vercel.com/blog/agent-responsibly): Understand why passing CI isn't proof that agent-written code is safe, and what owning a change you didn't write looks like. - [Would You Merge This?](https://vercel.com/academy/creating-a-software-factory/would-you-merge-this): Review four agent-generated changes, decide which are safe to move forward, and trace where each request ends up. ### Start with a complete factory Deploy a pipeline that turns work items into reviewed draft pull requests, or build one step by step. - [Software factory](https://vercel.com/templates/eve/eve-software-factory): Deploy a four-station eve factory where agents classify, analyze, implement, and review while people decide what ships. - [Build a software factory with eve](/kb/guide/eve-software-factory): Foreman, a software factory built on eve. It turns GitHub and Linear work items into reviewed draft pull requests through four agent stations, triaging issues and fixing its own red CI along the way. - [Creating a Software Factory](https://vercel.com/academy/creating-a-software-factory): Build a factory from scratch that routes work by risk, proves each request before writing code, and keeps merge authority with people. - [Software Factories on Vercel](https://vercel.com/docs/agent-resources/software-factory): Compare AI SDK, eve, Mastra, and Flue implementations of a software factory on Vercel. ### Start with a focused agent Begin with one agent that owns a single repository task, then add stations as you learn what it can handle. - [eve Coding Agent on Slack](https://vercel.com/templates/eve/eve-coding-agent-template): Deploy a Slack and GitHub agent that investigates bugs, reviews changes, and opens PRs from a sandbox with repository-scoped credentials. - [eve GitHub Maintainer](https://vercel.com/templates/eve/eve-github-maintainer): Start with a research-only maintainer that sends weekly issue digests, summarizes PRs without approving them, and hands work to Linear. - [Build a GitHub agent with Vercel Connect](/kb/guide/github-agent-vercel-connect): Build a GitHub agent that helps your team work through issues and PRs. Chat SDK handles the interactivity and AI SDK runs the tool-calling loop. ## Run the work safely Decide where each station runs repository code, what it can reach, and whether you build stations or run an existing coding agent. ### Isolate each station's workspace Give every investigation its own sandbox, limit its network access, and decide how long its state lives. - [Reproduce the Claim](https://vercel.com/academy/creating-a-software-factory/reproduce-the-claim): Build a read-only Investigator that runs in its own sandbox and uses command output to tell a real bug from a false report. - [A sandbox without a network boundary is only half a sandbox](https://vercel.com/blog/a-sandbox-without-a-network-boundary-is-only-half-a-sandbox): Learn why agents running untrusted repository code need deny-by-default egress, with credentials kept out of the sandbox. - [How Vercel Sandbox duration and persistence work](/kb/guide/vercel-sandbox-duration-and-persistence): Session duration and persistence are two separate controls in Vercel Sandbox. The timeout option keeps a single run alive for up to 24 hours, while snapshots and persistent sandboxes carry state across runs for long-running agents and pipelines. - [Give your software factory a browser](/kb/guide/software-factory-browser): Add a KERNEL cloud browser to the eve software factory template so Foreman can reproduce flow bugs, sign in through managed auth, and verify fixes on preview deployments in attended sessions. ### Run an existing coding agent Run Cursor, Claude Code, Codex, or OpenCode as a station instead of building every agent yourself. - [Making every harness durable](https://vercel.com/blog/making-every-harness-durable): Learn how durable-harness lets interrupted Claude Code, Codex, OpenCode, and Pi turns resume without repeating finished tool calls. - [How Conductor moved parallel coding agents to the cloud](https://vercel.com/customers/how-conductor-moved-parallel-coding-agents-from-the-laptop-to-the-cloud-with-vercel-sandbox): See how Conductor runs Claude Code and Codex agents in parallel on Vercel Sandbox, each on its own branch, and keeps them working after the laptop closes. - [Run Cursor Cloud Agents on Vercel Sandbox](/kb/guide/cursor-vercel-sandbox): Learn how to run Cursor Cloud Agents on Vercel Sandbox with self-hosted worker pools, durable workflows, isolated microVMs, and scale-to-zero compute. - [Investigate GitHub issues with \`HarnessAgent\` and Vercel Sandbox](/kb/guide/sandboxed-coding-agent-with-harnessagent): Use HarnessAgent to run coding agent harnesses like Claude Code or Codex against untrusted code inside Vercel Sandbox. Reproduce GitHub issues with the agent harness and generate a report for the maintainer. ## Review and merge Decide which work needs a person, what evidence reviewers get, and who holds merge authority. ### Route work by risk Let deterministic rules, not the model, decide which work proceeds automatically and which waits for approval. - [Classify, Then Authorize](https://vercel.com/academy/creating-a-software-factory/classify-then-authorize): Validate a model's type and risk labels with structured output, then write a deterministic router that sets each request's lane. - [Human-in-the-Loop with Chat SDK and Workflow SDK](/kb/guide/human-in-the-loop-with-chat-sdk-and-workflow-sdk): Combine Chat SDK and Workflow SDK to suspend workflows on approval cards in a chat platform, then resume on click via createWebhook. - [How Vercel Agent accesses production safely](https://vercel.com/blog/how-vercel-agent-accesses-production-safely): See how Vercel Agent separates investigation from action, asks for one approval per plan, and scopes credentials to that plan. ### Hand reviewers evidence, not claims Verify each change independently and open draft pull requests that carry their proof. - [Verify in Fresh Context](https://vercel.com/academy/creating-a-software-factory/verify-in-fresh-context): Build a Verifier that reruns checks on the pushed branch in a separate sandbox and sends blocking findings back to the Builder. - [Publish a Draft](https://vercel.com/academy/creating-a-software-factory/publish-a-draft): Limit the factory to draft PRs, add tests that reject ready-for-review PRs, and confirm merge authority stays with a person. - [Using Vercel Agent to review pull requests](/kb/guide/vercel-agent-code-review): Set up Vercel Agent Code Review to automatically review pull requests, apply validated fixes, request reviews with @vercel, and tune feedback with coding guidelines. - [Migrate from GitHub Actions to Vercel CI](/kb/guide/github-actions-to-vercel-ci): Move Turborepo tasks from GitHub Actions workflows to Vercel CI, migrate secrets to CI environment variables, and switch your required status checks. ## Operate and improve Connect the factory to production signals, then turn its mistakes into tests and evals. ### Feed production context in Let agents read deployments, errors, and incidents so investigations start from real evidence. - [Manage Vercel projects with a software factory](/kb/guide/software-factory-vercel-mcp): Add Vercel's hosted MCP to the eve software factory template so Foreman can read build logs, runtime errors, and deployment history for the app your target repository deploys, brokered at app scope through Vercel Connect. - [eve Incident Response Agent](https://vercel.com/templates/eve/eve-incident-response-agent): Deploy an SRE agent that tests hypotheses about production incidents and returns evidence your team can check. - [Introducing Vercel for Slack](https://vercel.com/blog/introducing-vercel-for-slack): See how a team tags an agent in a Slack thread to investigate an incident, fix a failing build, or turn a decision into a tested PR. ### Learn from failed runs Find the earliest wrong decision in a run, and lock in the fix so the factory doesn't repeat it. - [Teaching agents product design at Vercel](https://vercel.com/blog/teaching-agents-product-design-at-vercel): Keep accepted decisions in the repo as an agent skill, enforce clear rules with linters, and run a review loop that turns feedback into guideline updates. - [Learn From Failure](https://vercel.com/academy/creating-a-software-factory/learn-from-failure): Inspect durable traces from a flawed run, find the first wrong decision, and save it as an eve eval. - [Test the Stopping Rules](https://vercel.com/academy/creating-a-software-factory/test-the-stopping-rules): Write deterministic tests that make sure unclear work stops for a question and public API changes wait for approval. ## Reference Architecture Choose a stack and work through the decisions behind it, with curated documentation, guides, templates, and courses to help you build. - [Agent stack](/kb/architecture/agents): Build and operate agents with models, tools, and durable workflows. - [Web applications](/kb/architecture/web): Choose rendering, data, authentication, and caching approaches. - [Backends](/kb/architecture/backends): Choose a runtime, connect services, and run background work. [Browse all stacks](/kb/architecture)