What is a software factory?
A software factory is a repeatable system for turning ideas into shipped software. Today, AI agents do much of the work inside it, and people decide what ships. Here's how a modern one runs, and how to keep it under control.
A software factory, defined
A software factory is a set of standard tools, workflows and checks that lets a team produce software the same way every time. Work comes in, moves through defined steps, and comes out as tested, reviewed code ready to ship.
The idea borrows from manufacturing. On a well-designed production line, every station does a known job and quality checks happen along the way, so results stay consistent whoever is working that day.
In 2026, the term mostly describes an AI software factory, also called an agentic software factory, where AI agents do much of the hands-on work. Agents pick up tickets, write code, run tests and open pull requests. People set the rules, approve changes at key points, and stay accountable for what ships.
Where the term "software factory" comes from
The software factory is an old idea with a new engine. Hitachi was the first company to adopt the term, in 1969, aiming to make software production as standardized and repeatable as hardware production. In the late 2010s, the US Department of Defense took up the term for its own in-house software teams, starting with the Air Force’s Kessel Run.
Today’s software factory keeps the same goal and adds AI agents to the line. The new question is how to let agents do real work while people keep control of what ships.
What a modern software factory looks like
A modern AI software factory moves each piece of work through the same five stations. Agents handle the repetitive work at each one, and people step in where judgment matters.
1. Work comes in
A ticket, an issue or a pull request starts the process. The factory picks it up from the tools the team already uses, such as GitHub, GitLab, Jira or Linear.
2. Agents do the work
Each task runs in its own isolated environment, so agents can write code, install dependencies and run tests without touching anything else.
3. Checks run automatically
Tests, reviews and policy checks catch problems before a person ever looks at the change.
4. People approve
At set points, a team reviews what the agent did and decides whether it moves forward. Nothing important ships on an agent's say-so alone.
5. Changes ship, with a record.
The approved change goes out, and every step, decision and cost is logged so the team can trace what happened later.
How is a software factory different from an agent or a harness?
These three terms get mixed up. An agent is the worker, a harness is the workstation it runs in, and a software factory is the whole production line.
| Agent | Harness | Software factory | |
|---|---|---|---|
| What it is | An AI tool that writes and edits code for a developer | The software around an agent that gives it tools, context, permissions and a place to run | A repeatable system that takes work from request to shipped code |
| Who directs it | One developer, usually in real time | The developer or team that configures it: which tools, files and commands the agent can use | A team, through defined workflows and approval points |
| What it produces | Code changes, one task at a time | A working session where the agent's actions run in one environment | Reviewed, tested changes ready to ship, with a record of each step |
| Where control sits | In the prompt and the person watching | In its settings: permissions, sandboxing and tool access | In the workflow: isolation, checks, approvals and audit trail |
An AI software factory puts agents and their harnesses to work on the line, then adds what neither has on its own: shared workflows, approval points and a record of what happened.
Four things every software factory has to get right
Agents make it easy to produce more code. The hard part is producing it in a way the team can trust, afford and explain afterwards. These four problems decide whether a software factory helps or creates new risk.
Review that keeps up
Agents can open pull requests faster than people can read them carefully. Without clear approval points, reviews get rushed and problems slip through.
Costs you can see
Agents run on compute and model time. If nobody can see what each run costs, or stop a run that's going nowhere, spend becomes hard to predict.
Work that stays contained
Several agents working at once can collide or reach systems they shouldn't. Each task needs its own isolated space.
A record of what happened
When something breaks, the team needs to know which agent made the change, what it was asked to do, and who approved it.
How Upsun Dispatch™ runs a governed software factory
Upsun Dispatch™ is a governed software factory where your team and AI agents ship together. Agents do the work inside structured workflows, and your team approves what moves forward, with every step recorded and every run costed.
Workflows
Start with ready-made workflows for PR review and turning issues into code. They're triggered from GitHub, self-hosted GitLab, Jira or Linear.
Learn moreSandboxes
Each step of a run happens in its own isolated sandbox, so agents can't interfere with each other or reach beyond their task.
Learn moreHuman gates
Add approval points where your team reviews the agent's work before it continues. Sandboxes stop billing while a run waits for approval.
Learn moreCost transparency
See what every run costs. Pricing is $1.80 per agent-hour, billed by the second, with spend limits on your wallet. Cancelling a run stops billing straight away.
Learn moreFrequently asked questions (FAQ)
Is a software factory a tool or a way of working?
Both. It's a way of organizing software production into repeatable steps, usually supported by a platform that runs those steps, isolates the work and records what happened.
How is a software factory different from DevOps?
DevOps is a set of practices for building, testing and releasing software quickly and reliably. A software factory builds on those practices and packages them into a standard production line, which AI agents can now work inside.
What is a dark software factory?
The name comes from "lights-out" factories that run with no workers on the floor. A dark software factory is an agentic software factory that ships code with little or no human involvement. Most teams keep people at key approval points, because someone still has to answer for what ships.
What is a DoD software factory?
It's an in-house software team inside the US Department of Defense, such as the Air Force's Kessel Run, that delivers secure software through a standard DevSecOps pipeline.
Do software factories replace developers?
They change where developers spend their time. Agents take on more of the repetitive coding, while people define the work, set the rules, review changes and decide what ships.
Build your software factory with Upsun Dispatch
Let agents handle the repetitive work while your team keeps control of what ships, with every run isolated, approved and costed.
Stay updated
Upsun Cloud: compliant and validated