Service 03
AI software factory
We set up coding agents across your delivery pipeline, from spec to merge. Your engineers own the specs, the review gates and the releases.
In one paragraph
An AI software factory is a delivery pipeline where coding agents do most of the implementation work. Engineers write and approve the specs, own the review gates and decide what ships.
Deliverables
What you get
- 01 A readiness review of your repositories, including tests, CI, documentation and AGENTS.md files
- 02 A spec-driven workflow from ticket to pull request
- 03 Background agents in CI for well-defined work such as upgrades, test coverage and small fixes
- 04 Review agents and eval gates that run before a person reviews the change
- 05 Shared Agent Skills, rules and plugins that encode your conventions
- 06 Delivery metrics, training and pairing sessions for your engineers
Where it fits
Typical use cases
-
Backlog of small changes
Bug fixes, copy changes and small features that are clear enough to spec in a few lines.
-
Test coverage
Agents write tests for code that has none, so later changes are safer for people and agents.
-
Dependency and framework upgrades
Upgrades across many repositories, with one pull request per repository and the test results attached.
-
Internal tools
Admin screens, scripts and integrations that your team needs but never has time to build.
-
Code review support
A first review on every pull request that checks it against your conventions, security rules and the linked spec.
-
Documentation
Architecture notes and onboarding guides generated from the code and kept current in CI.
How we work
What changes
In most teams, engineers still type most of the code. In a software factory, they write the spec, an agent writes the change, other agents check it, and an engineer reviews the result. The work moves from typing code to deciding what to build and checking that it is right.
This works only when the repository is ready. Agents need fast tests, clear conventions and an explicit definition of done. Much of our first two weeks goes into that groundwork.
The pipeline
- Ticket. A person describes the change and the acceptance criteria.
- Spec and plan. An agent drafts a spec and a plan from the ticket and the code. An engineer approves it.
- Implement. A coding agent makes the change in a sandbox and runs the tests.
- Review. A review agent checks the change against the spec, your conventions and your security rules.
- Verify. CI runs the tests, the evals and the static analysis.
- Merge and release. An engineer reviews and merges. Your release process takes it from there.
Where to start
We start with one team, one repository and one class of work, such as test coverage or upgrades. We measure for four weeks, then expand to the next class of work. Most teams see the first useful results in the first two weeks, and the larger changes in the following months.
Guardrails
- Agents never get write access to protected branches.
- Every agent run is logged with its prompt, tool calls and diff.
- Secrets stay in your vault. Agents use scoped, short-lived credentials.
- Your engineers can stop or override any agent at any stage.
FAQ
Questions about software factory
Will agents push code to production on their own?
No. Agents open pull requests. Your branch protection, your CI and your reviewers decide what merges, and your release process decides what ships.
Which coding agent should we use?
It depends on your stack, your repositories and your security rules. We usually test two or three on your own backlog for two weeks and compare the merged results.
What do our engineers do differently?
They spend more time on specs, architecture and review, and less on routine implementation. We train the team on writing specs that agents can follow and on reviewing agent output fast.
Is our code safe?
Agents run in sandboxed runners with short-lived, least-privilege tokens and no production secrets in their context. We use enterprise model agreements in which your code is not used for training.
Related
Often combined with
-
Service 05
Legacy code modernization
We use coding agents to document, test and migrate legacy systems in small, verified steps, from COBOL and old Java to aging .NET and monoliths.
-
Service 07
Evals, security and AI governance
We measure whether your AI systems work, test how they fail and document them for auditors, with evals, red-team tests, tracing and compliance mapping.
Tell us which process you want to hand to an agent
A 30-minute call with an engineer. We will tell you whether an AI system is the right tool for it, and what it would take to run it in production.