AI Engineering
Tell us the outcome. We'll build it.
Product builds, retainers, internal tools and automations, shipped by senior engineers who have run what they built. We start with the smallest version that works and build out from how people use it.
What we build / 02
We ship the AI products themselves.
We build these for ourselves first. If it isn't good enough for us, it doesn't ship to you.
AI agents & internal tools
Copilots, chiefs of staff, and automations that take hours off your team's week.
AI workloads
RAG pipelines, model routing, and the plumbing that makes AI useful and affordable at scale.
Prototypes that test the offer
Cheap and fast, real enough to put in front of buyers before you commit.
Products you take to market
Full v1 products, designed, built and shipped, ready to put in front of customers.
Shipped / 03
Some of what we've shipped.
Full-stack engineering, with AI agents built in where they help. Everything here shipped agentically, in weekly slices.
- Property · live
A national property consultancy
Several internal tools and client-facing apps, including a carbon index that is live today.
Read the case study - Billing · live
SIM billing and reconciliation
Built for a data and electrical contractor. It checks carrier bills against every SIM in the fleet and raises the monthly invoices.
- Trading · live
Real-time trading
A real-time trading platform that acts on live market signals in under 300 milliseconds.
- Accounting · in use
AI skills for an accounting firm
Day one, discovery. Overnight, we built the AI skills. Day two, they were running on the firm's own files.
Read the story - Our own · in use
Sidepress
Our newsletter curation platform. It runs CtrlAltDebrief, our own AI newsletter, every weekday, and opens to other newsletters soon.
See the bet
How it ships
Every build runs through our own agentic engineering setup: parallel coding agents, and multiple independent AI reviewers on every change before it ships.
How we build / 04
Working software, fast.
Whether it's a product for your customers or an internal tool for your team, we build the same way: a small working version first, in front of real users early.
Start from what your team already does
We build from the actual workflow or decision the tool needs to serve, with your team in the room.
Ship a thin slice first
The smallest working version goes in front of real users early, and we build out from how it actually gets used.
AI-native build
We build with the same AI tools we train teams on, so a small senior team ships quickly.
Built into your stack
Products and internal tools plug into the systems your team already uses, so they get adopted and stay in use.
Automations / 05
The work that only happens because you're there.
Most businesses have a person who is the system. We turn that into software that runs on its own and hands back the exceptions. You will recognise at least one of these.
- Quotes are priced from memory, and only one person can do it.
- Availability lives in someone's head, and the week starts with phone calls.
- The same answer gets written out by hand, many times a day.
- A rate change means asking a developer, or it never happens.
- Nobody can say what was quoted last month, or on what basis.
01
Your rules, as data
The pricing, the thresholds and the exceptions come out of people's heads and become something you can see, edit and audit. A rate change is an admin page.
02
The standard work, automatic
The jobs that follow the rules get handled without anyone touching them, and the customer gets an answer immediately instead of tomorrow.
03
The rest, queued
Anything unusual goes to a person, with the context they need to decide. You see the exceptions rather than everything.
04
A record of what happened
Every decision stores what was decided, which rule applied, and who touched it. That is worth having on an ordinary Tuesday and worth a great deal if you ever sell.
How we scope / 06
In phases, with a fixed fee.
Each phase has a fixed fee, so you can stop after the first one. Once it's live, a retainer keeps a senior team on it: improving it, running it and building what comes next.
Safe to deploy / 07
AI you can deploy without betting the business on one frontier lab.
We help you answer what your board is already asking: who controls the model, and what happens to your data. We're not tied to one frontier model, so we choose the right tool for your problem.
The right model for the job
We're not tied to one lab. We pick the model that does your task best, at the right cost.
Your data, your control
Capable models, including open and self-hosted ones, without handing your data to a single overseas provider.
Built in the open
We fine-tune and post-train models for your real task, and show the work.
Ready when you are
Bring us the job that only you can do.
Book a 15-minute call. We'll tell you whether it's worth building, and roughly what it would take, before you spend anything.