Our Thinking
Written by the founders from production work: what actually happens between a working demo and a system an enterprise can trust.
We publish because the decisions that determine whether an AI programme works are made before anyone writes code, and they are usually made on the wrong evidence. What the system is allowed to do. How the business will know it is right. Which part of the estate has to change first. Whether the capability should be bought rather than built. Our writing takes a position on those questions instead of surveying them.
The test before anything is published here is whether we have had to make the call ourselves, on an engagement or inside a system the partners built and still operate. Nothing is published to sound current, and nothing is written to flatter the market. We would rather be argued with than agreed with, so we say plainly where the standard answer is wrong and what we do instead. If a piece changes how you scope the next programme, it has earned its place.
The subjects we keep returning to
Each one carries the firm's position on the subject and the articles filed under it.
- Production AI The distance between a working demo and a system a business can put weight on, and the engineering that closes it.
- Evaluation and assurance How a business knows an AI system is right, and keeps knowing it after the model underneath has changed.
- Technology strategy Deciding what to build, what to buy and what to leave alone, before the budget is committed.
The articles
-
The seven controls an AI system needs before it goes live
The seven controls a reviewer asks to see before an AI system goes live, from access control to change control, what each proves and the evidence to ask for.
-
How to choose your first AI project
The first AI project decides whether there is a second. Score every idea on what it is worth and how easy it is to deliver, and start where both are high.
-
What an AI agent costs to run once it is in production
What an AI agent costs to run turns on steps per task, model retirements and upkeep more than on token prices. How to read the bill and budget it per task.
-
Getting an AI pilot into production in financial services
What it takes to move an AI pilot to production at a mid-market bank, lender, insurer or wealth firm now that SR 26-2 leaves generative AI to your own review.
-
Why most enterprise AI pilots never reach production
The demo was never the hard part. Pilots die in the gap between a working prototype and a system the enterprise can trust, and that gap is engineering.
-
Evaluation is the contract between an AI system and the business
An AI system without an evaluation suite is a system with no acceptance criteria. Evals are where the business writes down what it is actually buying.
-
Build versus buy for enterprise AI in 2026
The models are commodities and the vendor market is crowded. The build-versus-buy question now turns on where your differentiation actually lives.
Talk to a partner
If one of these arguments describes the decision in front of you, write to us and a founding partner will come back with a view on it.