Services

We work with small development teams as a design partner: our tooling on your repository, us with you every step of the way. Our aim is to build tools and processes that your team continues to use because the value is undeniable. Working with us starts with a Consulting Pilot.

You get AI agents whose changes you can review and trust, working to the same workflows we build our own tools with. You get release and CI workflows that are already built and run on the CI you have, and CI that costs less because it runs only the checks a change reaches. And you get a record of gated commits and written decisions that still holds when the team grows.

Development tooling

dev-tool, our development command-line tool, treats your repository as data. It discovers the modules in your codebase and the dependencies between them, records them in a manifest, and checks that what the code actually uses matches what is declared. It runs your verification suite (formatting, lint, type checks and tests) through one set of commands, and checks that generated code matches the schema it was generated from. The same commands run on a developer's machine, in CI and for an AI agent, so every change is held to the same gate. Python and TypeScript are supported today.

CI/CD

CI/CD and a unified release process come through ops-tool, our operations command-line tool. It defines the workflows around a change (the checks a merge is held to, publishing, version bumps) as commands, and the CI you already run invokes them. There is no pipeline service of ours to keep up and nothing of ours to go down: it complements your CI rather than replacing it, unless you want it to.

The workflow patterns

Alongside our tools come patterns for working with an AI coding agent without it making a mess of your repository. Each is a rule the tooling enforces rather than a convention people are asked to remember:

• Nothing is finished until it passes the gate. Formatting, lint, type checks and every generator's check run after each change, and the agent reports the output rather than its opinion of it.
• Every AI prompt is a commit. The agent's work lands in small, numbered increments on its own branch, so any step can be read, reverted or questioned on its own.
• Follow-ups and decisions are tracked automatically. As the agent works, it records what it deferred, what it found and left unfixed, and what it chose not to do, and each item lands on the pull request. Before the branch merges, every item is ruled on: accepted into the backlog, resolved by a named commit, or rejected with the reason written down. Nothing deferred is quietly forgotten, and every decision keeps its reasoning next to the change it was made about.
• Every AI model run is measured. Each run records how long it took and whether its result was correct, so the data is there to improve the workflow, and the choice of model for each task, over time.
• The agent works bounded. Agent work runs in a container with credentials scoped to its own branches, and starts by proposing changes rather than pushing them. You turn that off when you trust it.

Taken together, these rules give a small team confidence in the changes its agents make, and give a larger organisation a history of gated commits and recorded decisions that its compliance process can check.

What we support

Today we work best with teams of five developers or fewer who are willing to adopt a stack we support:

• Languages: Python and TypeScript
• AI services: Claude Code
• Git hosts: GitHub and Bitbucket
• CI/CD frameworks: GitHub Actions and Jenkins
• Issue trackers: Jira

Support for more of each is on the way.

Consulting Pilot

Working with us starts with a pilot: a defined period in which our tooling and processes are brought to a repository you own. During this time we will work closely alongside your team to remove stumbling blocks and answer questions.

Before it starts

It starts with a conversation about your team, your codebase and the challenges you have had in adopting AI agents in development. If the fit is right, we agree to the pilot in writing before any work begins: the scope, how long it runs, what it costs, whose account model requests run on and under what terms, and what you keep when it ends. Nothing is left to be settled at the end.

Step by step

The pilot works through the steps on the Getting Started page. We work through them with your team rather than handing over instructions, and each one begins once the previous one is in use.

While it runs

We keep in regular contact through a mutually agreed channel. We like to hear about the parts you switch off as much as the parts you like. What you turn off tells us which rules are doing real work and which are only in the way.

When it ends

We review what worked and what did not, and the terms agreed at the start decide what happens next.

What we hope to achieve together

What your team gains is at the top of this page. For us:

• evidence from a codebase we did not shape, which is the only real test of tooling
• knowing which parts carry the value, and which only get in the way

Success is simple: your team keeps using the tooling after the pilot without being asked to.

Candidates for pilots

• five developers or fewer
• a stack we support

If that is not you, get in touch anyway. It tells us which of these requirements to widen first.

Starting a conversation

Email info@solidplanssoftware.com with a line about your team, your stack, and any challenges you have had in working with AI agents.