StrategyFind the compelling path forward.
We look for where the money actually moves: where it leaks, and where more of it could come in. That means following it both ways: the backlog that delays revenue, the work that forces a hire, the errors that cost rework, and the products you haven't built yet. We'd rather fix one thing that shows up on the P&L than five that look good in a demo.
OperationsFit it to how the work runs.
A system nobody uses is a cost, not an asset. We design around the people who will use it every day: how they already work, what they won't give up, and where a change actually saves them effort. If it doesn't fit the Tuesday afternoon version of the job, it doesn't ship.
EngineeringBuild it to last.
We write custom software against your real data and your real edge cases, not a sample file. It's tested on the cases that matter to you before we call it done. AI goes in where it beats a simpler approach, and nowhere else.
The three are worked together by the whole team. A strategy call changes the build. A build finding changes the strategy.
- Sit with the work.Before proposing anything, we spend time inside your operation with the people who do the work: the inbox, the spreadsheets, the handoffs.
- Pick one problem.Choose the one where fixing it changes cash or capacity. You'll know what it is, why it was chosen, and what "fixed" means before anything is built.
- Build it in place.A working version on your real data within weeks. Your people use it early, and we change it based on what actually happens.
- Keep it running, or hand it off.We maintain it, or we hand it to your team with documentation written for the people who will run it. Your call.
- One person who owns the problem and can make decisions.
- Access to the systems and data the problem runs through.
- Time with the people who do the work. An hour here and there, not a workshop series.
We'll say so early if the problem isn't one we can fix well.
How it's built
Where AI fits
We use language models and agents for work that involves reading, judging, and writing: sorting incoming requests, drafting from records, matching across messy data. We use plain software for the rest. An agent is software that takes a task from start to finish, with a person checking the parts that matter.
How we build
Tooling is chosen per problem, for the domain's problems and goals. Examples of what that has meant:
- agent harnesses such as Y Combinator's QM or the OpenAI Agents SDK;
- Claude Cowork for knowledge work;
- Boring Semantic Layer, so agents query business data through consistent definitions;
- custom MCP tools for your systems.
Every build is evaluated against your real cases, which are kept as a test set so later changes can be checked.
Where it runs
In systems you control. Your data isn't used to train anything.
What you're left with
- Working software.
- The test cases that prove it works.
- Documentation for whoever maintains it.
How long does an engagement take?
First working version in weeks. The full length depends on the problem; we'll give you our estimate after sitting with the work.
What if it doesn't work?
You'll hear it from us early, while there's still time to change direction.
Do you work on site?
Both. Whatever the work needs.
Who owns what we build?
You get full rights to run and change the system we deploy for you. We keep ownership of the source it's built from.