An AI implementation firm for mid-market companies
We build and run AI systems inside firms of 200 to 1,500 people, most often where documents, reviews and approvals move in volume. Our clients are financial services, professional services and legal teams with no internal AI capability.
Mission
Frontier AI should be as ordinary as the rest of your infrastructure
The models are already good enough. What separates the companies with AI in production from the ones with a folder of pilots is engineering, and that engineering is the whole of what we do.
How we operate
The three things that do not flex
The culture is the operating model. It is visible in how an engagement runs, not in a values poster.
We build and run systems, not strategy documents.
The deliverable is working software in your environment, monitored after it goes live. No roadmaps, no maturity models, no opinion decks.
Small by choice.
Few engagements at a time, start dates that queue, and no bench to keep busy. Growth is not the goal. Shipped systems are.
The builders are the room.
No account layer, no handoff between the people who sell and the people who deliver. Every conversation includes someone who writes the code.
The controls
What goes into every system we build
These five are not options, and they are not added at the end. They come from years of watching AI projects fail an audit rather than a benchmark.
We ground every answer in your documents, with citations.
Every response points back to the passage it came from, so an examiner can be shown a source instead of an assurance.
We respect the permissions you already have, rather than flattening them.
The system reads only what the person asking is already entitled to read. AI should never quietly widen access.
We write an audit record of what the system did.
Inputs, sources, outputs and actions are written to a record you can query.
Consequential actions pass through a human gate.
The system prepares the action. A named person approves it, and the approval is stored with the decision it authorised.
We measure a baseline before launch.
We record how the process performs before the system exists. Without that number, no improvement can be proven afterwards.
Who runs it
Why Chris Ik built the firm this way
Most AI projects in regulated firms fail on access and evidence, not on the model.
Founder
An assistant reads a repository it should never have reached. An answer arrives that nobody can trace to a source. An action is taken with no record of who approved it. By the time an auditor asks, the project is finished.
Those are security, risk and audit problems before they are AI problems. This firm came out of that work, in regulated environments where you learn quickly that these systems break somewhere other than the model.
So the controls are designed in the first week rather than retrofitted after the demo goes well.
Built and run by us
Four products we built and still run
These are ours, not client logos. Each one is live, and you can go and use it before you talk to us.
Start with one workflow
Tell us which review or approval process costs your team the most time. We will tell you if it is a fit and what a first build would involve.