Practical AI capability through responsible, hands-on learning.

I used to fix cars. When I decided to retrain, I started where the government's own guidance pointed me: its official AI skills platform. I wanted what it promises: an accredited and recognised way in. Each course I picked handed me off to a technology company's own signup instead — another account, more hoops to jump through. After a few rounds of that, I stopped.

Grounded AI Practice is what I built next. Part of it is research: what the UK has promised on AI skills, and what its published numbers can and cannot show. Part of it is practice: the system I learn and work with now, kept in the open where anyone can check it.

The research so far comes down to one gap. Government documents put AI's value to the country in the hundreds of billions of pounds. Millions of workers have been promised training by 2030. Progress is reported in courses completed, not people trained, and the counting is done by the companies that sell the courses. The one number that would settle it — how many people have actually been trained — is not published.

I research that gap in the open: every claim on this site traces to a source anyone can check, and when I get something wrong the correction stays visible instead of being tidied away. The teaching this builds toward is aimed at the people the official numbers say least about: ordinary workers and small organisations.

What is here

Three ways into the work

The numbers

What was promised, what has been counted, and the missing number between them. Every source is named, along with who paid for it.

Method

The rules I work to, and the path a claim has to survive before I rely on it.

The system

The tools, logs and rules I work with, and the difference between how a large organisation deploys AI and what a small firm gets.

Current work

What is being built now

In draft

A public report on the UK's AI skills programme

A short report for a general reader: what was promised, what the numbers show, and whether anyone outside government can check them. The fully sourced technical draft sits behind it as the evidence companion.

In design

A pilot learning unit on effective prompting

One practical skill, taught properly: what actually happens when you hit send. Thirty to ninety minutes, moving from a worked example to doing it yourself. I'll test it with a small group of real learners before anything bigger is built.

In active use

The practice system itself

The rules, logs and custom tools this site describes, and the machines they run on — including an always-on home server I built and documented step by step, so the process itself becomes teaching material.