Definition
Software factory
An AI software factory is a structured, automated production system where autonomous AI agents write, test, review, and ship code. Humans shift from writing code to designing and governing the workflow: what work enters the system, how it is verified, and what gets merged.
Why it matters
A local harness makes one person faster at the keyboard. A factory changes how a team produces software. Without structure, you still have people stitching steps together by hand.
How Islo helps
Islo gives you an easy way to build software factories: automations to route work and trigger agent runs, sandboxes where agents work on your real stack, governance so humans stay in control of what ships, and a gateway that scopes credentials and audits every outbound call.
Factory vs local harness
A local harness — Claude Code, Cursor, or Codex on your laptop — makes one person faster while they are at the keyboard. A software factory changes how work enters the system, how it is verified, and how output reaches production. Tickets arrive from Linear or Jira, agents pick them up on isolated computers in the background, run against your real services unattended, and return pull requests with evidence. Humans design the intake rules, verification gates, and merge policy. Agents handle the repetitive implementation and testing in between.
What you need before calling it a factory
A factory needs durable environments so agents do not rebuild the world on every run, scoped credentials so tokens never land in model context, and an audit trail so security can answer what an agent touched. It also needs orchestration: something that routes work, retries failures, and posts results back to Slack or the issue tracker. Without those pieces, you have a harness on a laptop, not a production system.
Keep reading
Common questions
Is a software factory the same as an agent?
No. The factory is the production system. Agents are the workers inside it; humans design the workflow and approve what ships.
What do humans do in a software factory?
Design intake, verification, and governance. Agents handle implementation, testing, and opening pull requests.
Do we need a new issue tracker?
No. Agents read from GitHub, Linear, Jira, and Slack as they are.
Who approves the output?
Your team. Agents open pull requests; humans merge.
Where should we start?
One workflow you can grade quickly, then expand when you trust the loop.