What Is an AI Dark Factory?
An AI dark factory is an automated software system that works alongside people. It takes approved requirements as input and produces software ready to ship. It also watches production, investigates issues, proposes or applies fixes, and works on user feedback someone has already approved. People set direction, approve the work, and handle exceptions.
Key Takeaways
- The idea comes from lights-out manufacturing, where a plant runs for long periods with nobody on the floor.
- Inputs are approved requirements, not open-ended prompts.
- Outputs are changes ready to ship, with tests and evidence attached.
- The system watches production and investigates before anyone asks.
- It acts on user feedback that has already been approved.
- People own direction, approval gates, and exceptions.
How a Dark Factory Runs
Build From Requirements
Approved requirements go in. Agents implement, run the tests, and return a change a person can approve.
1Approved requirements
2Implement
3Verify
4Ready-to-ship change
Watch Production
An alert or a failed deploy starts an investigation. The system proposes a fix, or applies one inside limits people set.
1Alert or failed deploy
2Investigate cause
3Propose or apply fix
Act on Approved Feedback
User feedback that a team has approved becomes a scoped change, with the same review as any other work.
1Approved feedback
2Scoped change
3Review and merge
Levels of Autonomy
- 1
Assisted
A human drives the workflow while the agent suggests or performs bounded actions.
- 2
Supervised autonomy
The agent drives a task while a human monitors progress and steps in when needed.
- 3
Gated autonomy
The agent runs on its own until a review gate, usually a pull request, and people judge the result.
- 4
Narrow lights-out
A bounded class of work runs end to end without a routine approval gate, because its checks are strong enough.
- 5
Broad lights-out
Large parts of the software lifecycle run without routine gates. People set goals, policy, and exception handling.
Dark Factory vs. AI Assistant
Input
AI assistant: A prompt typed into a chat
Dark factory: Approved requirements with a clear definition of done
Output
AI assistant: Code for a developer to finish
Dark factory: A change ready to ship, with tests and evidence
Production
AI assistant: Engineers notice incidents and start the investigation
Dark factory: Watches production and investigates first
Human role
AI assistant: Supervises each step
Dark factory: Sets direction, approves gates, handles exceptions
What People Still Own
Direction
People decide what the system should build, which problems matter, and which constraints hold.
Approval gates
Every change that reaches production passes a person's review before it merges.
Exceptions
Unclear requirements, risky changes, and unexpected failures go to people.
Who Is Building Them
Uber · Software Factory
Managed agents handle code review, self-healing CI failures, on-call alert triage, and bug debugging. Humans review and escalate.
- Code review
- Self-healing CI
- On-call alert triage
- Bug debugging
StrongDM · Software Factory
Humans define intent: the behavior the system needs, the scenarios it must handle, and the constraints that matter. Agents generate code and validate it against those scenarios.
- Intent
- Scenarios
- Constraints
- Validation
Where to Start
Pick One Loop
Start with work that repeats and has a result someone can check, such as failed build triage.
- Failed build triage
- Flaky test cleanup
- Dependency updates
Define Done
Decide what a ready change looks like before the first run: tests pass, scope stays bounded, evidence is attached.
- Tests pass
- Scope stays bounded
- Evidence attached
Keep the Gate
A person approves every change until the results hold for several weeks. Then widen the scope.
A person approves each change
Common Questions
What is an AI dark factory?
An AI dark factory is an automated software system that works alongside people. It takes approved requirements as input and produces software ready to ship. It also watches production, investigates issues, proposes or applies fixes, and works on user feedback someone has already approved. People set direction, approve the work, and handle exceptions.
Where does the term come from?
It comes from lights-out manufacturing, a highly automated production facility designed to run for long periods without people on the factory floor. An AI dark factory applies the same idea to a defined class of software work: implementation, verification, and delivery run autonomously, while people set goals and handle exceptions.
Does an AI dark factory replace engineers?
No. People set direction, define requirements, approve the gates that matter, and handle exceptions. The system takes over repeatable steps so engineers spend their attention on the hard problems.
Is an AI dark factory the same as an AI coding assistant?
No. A coding assistant helps a developer while they drive the work. A dark factory takes approved requirements, runs the work with a defined boundary, and returns a result that is ready to ship.
What does a dark factory do after release?
It watches production, investigates alerts and failed deploys, and proposes or applies fixes within limits people set. Anything outside those limits goes to a person.
What can a team automate first?
Start with a narrow, repeatable loop that has a result someone can check, such as failed build triage, dependency updates, or flaky test cleanup. Expand once the results hold for several weeks.
How is a dark factory different from background agents?
A background agent is one unattended run started by an event, a schedule, or a request. A dark factory is a system of such runs that covers a whole class of work, from requirements through production follow-up.
Keep Your Attention for the Hard Problems
Start with one repo and one recurring task.