01Make the intent available.
Connect the request to the decisions, discussions, and constraints behind it. An agent should not have to reconstruct the brief from a ticket while the latest decision lives in Slack.
Connect work and contextScale human + AI craft
What is a lights-on
software factory?
A lights-on software factory combines autonomous AI execution with visible work, shared context, and human judgment throughout delivery.
Where a dark software factory emphasizes unattended completion, a lights-on factory keeps collaboration and judgment available as the work changes. Scale execution while preserving creativity, taste, and the ability to intervene.
Keep the whole delivery workflow in view.
A lights-on software factory combines AI execution with visible work, connected context, and accountable decisions. Agents can operate autonomously while people can shape intent, review outcomes, and intervene when their judgment matters.
The lights stay on across the whole delivery workflow: requirements, design, implementation, review, and the decisions that connect them. The factory succeeds when that work becomes useful software.
Lit Factory brings your tools, teams, and coding agents together on one live Floor. Shared knowledge, task autonomy, automations, and operational insights help the entire workflow keep pace with agent execution.
The lights belong wherever a decision changes the outcome.
An agent can implement the wrong decision perfectly. More code can leave reviews waiting, questions unanswered, and a release unfinished. Build a system that carries intent and responsibility alongside execution.
02Choose autonomy for the task.
Some work needs close collaboration. Other work can proceed independently with review when it is ready. Set the level of autonomy to suit the task, and stay able to step in when the situation changes.
Explore task autonomy03Keep judgment in the workflow.
Give clarifications, design decisions, and reviews an owner and a place in the work. Creativity and taste matter before implementation as much as they do when someone inspects the result.
Turn interruptions into action04Carry evidence through delivery.
Keep the result, connected activity, and review decisions in context. Passing a test answers one question. Whether the change solves the intended problem and belongs in the release requires a broader decision.
From ticket to trusted delivery05Improve the whole operation.
Look for waiting, repeated explanations, and rework across the delivery process. Use operational insights to improve how work moves, rather than optimizing only how much code an agent produces.
Observe and optimize the flowIllustrative workflow
A lights-on software factory in practice
Consider a team improving its signup experience. A coding agent can build a shorter flow and pass every test while leaving a product decision unresolved. A lights-on workflow makes that decision part of delivery.
01 · Product judgment
Define the outcome
Improve signup for a new customer segment. Agree what success means, the constraints, and who accepts the result before implementation starts.
02 · Design collaboration
Resolve the decision
The shorter flow removes a step that some customers need. Bring the question to the person who can decide, and carry that decision into the task.
03 · Agent execution
Execute with context
The coding agent works from the agreed intent in the developer's configured harness. Choose autonomy for the task and keep the work available for intervention.
04 · Review and evidence
Accept useful delivery
Check the implementation against the intended experience as well as its tests. Keep the review decision with the work, then verify the change reaches its users.
The agent can handle implementation independently. People contribute where intent, taste, or acceptance needs judgment. When the brief changes, the decision and its owner remain connected to the work.
Explore the delivery workflowKeep your harness
Your coding agents.
A shared way to get work done.
Developers are most productive in the agent environment they have tuned for their workflow. Keep your configured Claude Code or Codex harness, your skills, and your team instructions.
Lit Factory connects that work to the wider delivery process. The task's autonomy setting guides how the connected agent collaborates and proceeds, with people able to step in as needed.
Your shared Brain carries operational knowledge. Dynamic channels organize connected work. Automations help route the next action, while metrics and optimization reveal where the workflow needs attention.
Explore connected agents and toolsLights-on vs. dark software factories
These are approaches to organizing work, rather than fixed descriptions of every product. Independent execution is valuable when the intent and evaluation are clear. A lights-on approach also supports the collaboration needed when those decisions are still taking shape.
What you optimize
Lights-out emphasis
Unattended completion of specified work.
Lights-on emphasis
Useful delivery, including the decisions and reviews that shape it.
Where judgment enters
Lights-out emphasis
Primarily in the specification and automated evaluation.
Lights-on emphasis
In planning, collaboration, review, and intervention as the work develops.
How autonomy is chosen
Lights-out emphasis
Independent execution is the organizing ambition.
Lights-on emphasis
The task determines the appropriate level of collaboration and independence.
What success looks like
Lights-out emphasis
The agent completes the brief and passes its checks.
Lights-on emphasis
The result meets the intent, is accepted, and reaches the people who need it.
Real engineering work
More throughput.
Human judgment retained.
6.7×
Merged PRs per development week
Output rose from about 19 to 127 merged PRs per development week across the measured phases. Human review and QA remained part of the workflow. The team observed a substantial improvement in quality as output increased.
See the measurement and methodologyBuild your factory around one real workflow.
Start where agent execution meets a recurring bottleneck: a review queue, a design decision, or a handoff across teams. Connect the tools and context, choose the task's autonomy, and identify what an accepted result needs to prove. Measure waiting, rework, and useful delivery as you expand.
For enterprise teams, run Lit Factory on your infrastructure using Kubernetes. Bring your models and the tools your organization uses, while keeping the shared operating workflow under your control.
Questions about lights-on software factories
What is a lights-on software factory?
A lights-on software factory combines AI execution with visible work, connected context, and accountable decisions. Agents can operate autonomously while people retain the ability to shape intent, review outcomes, and intervene. The goal is to scale useful software delivery while preserving creativity, taste, and judgment.
What is the difference between a lights-on and a dark software factory?
Dark software factories emphasize unattended implementation and verification. A lights-on approach makes collaboration and judgment part of the delivery system. Both can use autonomous agents; the difference is how the wider workflow handles changing intent, review, and responsibility.
Does lights-on mean watching every agent action?
No. Continuous supervision would create another bottleneck. Keep work and decisions visible, choose autonomy for each task, and bring people into the moments where their contribution changes the outcome.
Can we keep Claude Code, Codex, and our existing coding harness?
Yes. Lit Factory works with connected coding agents such as Claude Code and Codex. Keep the harness, skills, and team instructions you have optimized for your workflow. Lit Factory coordinates task autonomy, shared context, and the work around that agent environment.
Can a lights-on software factory run on our infrastructure?
Yes. Lit Factory Enterprise runs on your infrastructure using Kubernetes, with your model choices and connected tools. Your organization can coordinate people and agents without moving its operating workflow into a separate vendor-hosted workspace.