Online · ’26
Four practices, one operating thesis: every business function can be modelled as a graph of inputs, decisions, and outputs — and almost all of it can be safely automated, augmented, or replaced.
We refuse open-ended discovery. Each engagement is a fixed cadence of five focused passes — every one ends with something running, measured, and reviewable.
A small selection of what we’ve built recently. We optimise for operators on the floor and CFOs in the boardroom — the same number, viewed from two sides.
Replaced a 12-person dispatch desk with a hybrid agent that reads shipping orders, plans routes, negotiates rates over email, and escalates only when margin or SLA risk crosses thresholds.
Reads incoming KYC documents, pre-fills regulatory filings, flags inconsistencies, and ships an audit trail every regulator has so far been happy to read.
Replaced clipboards and three legacy systems with a single web + mobile flow. Voice notes get structured, insurance gets verified in-line, and physicians read a one-page brief before walking in.
Daily reprices 18,000 SKUs against competitor signal, weather, and shelf-life. Buyers approve the model’s suggestions in a three-pane web app instead of a thirty-tab spreadsheet.
We pick technology on a ten-year horizon, not a hype cycle. Everything below has earned its spot through repeated, scarred, in-production use.
Names below have given permission to be cited. Metrics behind the quotes are independently verified in the corresponding case briefs.
Altegena landed in week one with a working agent in our staging environment. Every other vendor we’d met arrived with slides. The difference was felt immediately across the dispatch floor.
They don’t treat AI as a feature you sprinkle on. They treat it as a load-bearing wall and engineer around the fact that it can fail. That distinction is why the system is still running, two years in.
Short, specific essays from inside our engagements — the architectural choices, the failure modes, and the numbers we actually watch in production.
Why the most lucrative AI deployments of the last 18 months are still small agents wrapped around email, an LLM, and a job queue.
A practical pattern for shipping agents that keep working after you’ve stopped staring at the dashboard.
Notes from a slow, careful migration that let us layer modern automation onto a system nobody wanted to rewrite.
Two paragraphs is plenty. We reply within 24 hours, on weekdays, from a human inbox. Existing engagements get answered first.