How I use AI
I use AI assistants every day, mostly Claude Code, as a fast pair programmer and editor. I decide what gets built and why, and nothing an assistant produces ships until it has been checked.
Last reviewed Oct 2026
Where it helps
- Drafting code and tests from a design I have already written down
- Reviewing my changes for mistakes before they are committed
- First drafts of documentation, which I then edit
- Finding my way around an unfamiliar library or codebase
- Checking my writing against my own style and confidentiality rules
Where I do not use it
- Choosing an architecture or a trade-off: those decisions, and the reasons for them, are mine
- Deciding whether data is correct: tests, reconciliation and alerts do that, not a model's opinion
- Employer or client code, data and documents: they stay out of my personal AI tools
- Claims about my experience: every figure on this site comes from my own records, not from a model
How the output is checked
- Every figure on this site sits on an allowlist with its source, and the build fails on any figure that does not
- Before anything deploys, the site runs content checks, a confidentiality audit and hundreds of browser tests
- Code changes come with tests, and the public pipelines show their checks in their repositories
This site
An AI assistant helped build this site and draft some of its text. I review every page before it goes live, and every figure on it links to its source.