◆ About
Twenty-five years of building systems that have to be right.
All of it inside extremely regulated environments. Places where being wrong is expensive, everything is audited, and nobody is impressed by a demo. I now spend my time getting AI into organisations like those, and keeping it working once it is there.
◆ Where I come from
Software that was never allowed to be sloppy.
My career has been spent in extremely regulated environments, where the work was never a demo. It shipped, it was audited, and people relied on it. That is a different school from a startup that can roll back on a Friday.
I am used to places where the data cannot leave the boundary and the governance is not optional. That background is why I am careful about the boring, load-bearing parts of an AI system. Where the data goes, who can see it, what happens when it is wrong. Those are not afterthoughts. They are the job.
◆ What I do now
Making AI land where the work happens.
I work inside organisations that already have a business to run, and my job is to make AI land there without breaking anything. That covers working out what is worth building, getting the data fit to be read, engineering the system around the model, governing what it is allowed to see, and teaching the people who will live with it. All five are laid out on work with me.
The model is the easy part now. Anyone can call one. The value is in the reliable system you build around it: framing the real problem, feeding it the right context, checking its work, and keeping it honest in production. That is the discipline I write about in The AI Engineering Pipeline.
◆ What actually stops it
It is almost never the model.
A pilot that answered well in a room meets an estate nobody curated, permissions nobody audited, and approval processes nobody wrote down. That is where technically sound projects stall. Most of my work is in that gap, and most of it is unglamorous.
Engineering & rigour
Leak-tested pipelines, secure deployment, governance you can audit. Built to keep working after I leave.
Data & governance
What your AI can see, what it should not, and who is allowed to build with it. Turning a vague worry into a checklist a team can actually clear.
◆ How I work
I would rather tell you it will not work.
I say early which ideas will not survive contact with your data or your people. That is usually the cheapest thing I do all engagement. If a piece of work needs a specialist I am not, I will say so rather than learn it on your budget.
I write in public, including the parts that did not work, because you should be able to check how I think before you pay me anything. What I hand over is meant to keep running once I am gone. If your team cannot maintain it without me, I have built the wrong thing.
◆ The through-line
Don’t hire a smarter genius. Build a better harness.
If you are stuck between “the tech works” and “it is still not in production”, that gap is the work.