The AI software factory
By Sandeep PandaCo-founder, Bug0 · Building FactoryKitUpdated
Anyone can wire a coding agent to a repo now. That alone is not a factory; it is unsupervised output. What makes a factory is the inspection: fixed quality stations that every change passes before a human sees it. Those stations are the difference between a production line and a demo.
Definition
An AI software factory is a production line for code changes. Tasks go in; coding agents implement them in isolated sandboxes; every change runs the repo’s checks and browser QA; verified pull requests come out for human review. Output is measured in merged PRs, not suggestions.
What makes it a factory and not a script
Three properties separate an AI software factory from an agent hooked to a webhook. First, the unit of work is a task with acceptance criteria, not a chat prompt. Second, production is isolated: each task runs in its own sandbox with scoped credentials, so a bad run cannot touch anything beyond its own branch. Third, the stations are fixed. Checks, QA, and packaging run the same way on every change, whichever agent did the implementation.
FactoryKit’s line runs Claude Code, Codex, or Grok Build per task (Claude Code vs Codex covers how we route), clones your repos into a sandbox as siblings so tasks can span services, and ships one pull request per changed repo.
Quality control is the product
Factories earned their reputation from inspection, not speed. A line without QA produces defects at scale, which is worse than producing them slowly.
So every FactoryKit change runs the repo’s own checks, with up to 3 fix attempts, then a self-review, then QA in a real browser. The browser session is recorded and the video is attached to the pull request: your reviewer watches the feature work before reading a line of the diff. We learned this doing automated QA at Bug0: a passing test suite is not the same thing as working software. Reviewers deserve to watch the change run.
Yours, on your infrastructure
Most hosted agents ask you to send your code to their cloud and trust their tenancy story. For teams past a certain size, that conversation ends at security review.
FactoryKit inverts it. Our forward-deployed engineers build your factory inside your own infrastructure: your code stays in your accounts, you can bring your own model keys, and AI usage is billed at provider cost. No task gating; self-serve is $99 per user per month plus pay-as-you-go model usage (commercials).
How to evaluate an AI software factory
Five questions any vendor should answer in writing:
- Can you choose the coding agent per task, or are you locked to one vendor’s model?
- Where do credentials live during a run? The agent should never hold real tokens; FactoryKit injects them in transit, scoped per repo.
- What evidence ships with each change? Ask for recorded QA, not adjectives.
- Who runs git? In FactoryKit the agent only edits the working tree; the host commits, pushes, and opens the PRs.
- Can one task span repos? Real changes cross service boundaries and should produce one PR per affected repo.
Ours are answered by the architecture (coding agent security walks through it).
FAQs
What is an AI software factory?
A production line for code: tasks go in, coding agents implement them in isolated sandboxes, every change passes the repo’s checks and recorded browser QA, and verified pull requests come out for human review.
Does an AI software factory replace engineers?
No. It moves them up the line: engineers set direction, write acceptance criteria, and review output. The pull request is the approval gate; nothing merges without a human.
Can an AI software factory run on our own cloud?
Yes. That is FactoryKit’s default motion: forward-deployed engineers install the factory in your infrastructure, run it on your backlog for the first 1 to 3 months, and hand it over to your team.
Which coding agents does FactoryKit run?
Claude Code, Codex, and Grok Build, selectable per task. The factory around them, meaning the sandboxes, checks, QA, and credential brokering, stays the same whichever agent runs.
How is an AI software factory priced?
FactoryKit self-serve is a $99 per user per month platform fee plus pay-as-you-go AI usage, billed at provider cost with no markup, or bring your own keys. Enterprise deployments on your own infrastructure are usage-based, with rates quoted in writing during scoping.
About the author
Co-founder, Bug0 · Building FactoryKit
At Bug0, Sandeep architects Passmark, the open-source AI regression testing framework. He previously co-founded Hashnode and scaled it to millions of developers as its CTO. He wrote AngularJS: Novice to Ninja, co-wrote Jump Start HTML5, and thinks mostly about agentic testing and the infrastructure that makes AI-written code safe to ship.
See a software factory run on your repos
A demo is a working session: your repos, a real task from your backlog, a finished pull request.
Book a demo