I'm Dave Holmes-Kinsella. I build AI-expert data teams.
For 25+ years I've turned messy data into decisions and products across healthcare, fintech, and telecom. Now I help data teams, and the organisations that want to get there, become AI-native: evaluation loops, standards, and habits that make AI dependable when the tools change faster than the org chart.
- tool Impact Advocate (beta) 2026-10-10
- tool work-ledger 2026-10-03
- project mcp-remote-sudo 2026-09-27
- tool Tricorder 2026-09-05
- study Terms & Conscience 2026-09-03
- study Seven Grids 2026-08-26
- essay Just say no to STE-100 2026-08-25
- study Five Years of Open Water 2026-08-25
- study Two Swims, One Route 2026-08-25
- study An Opinionated Guide to Writing Techniques 2026-08-25
Just say no to STE-100
Or: the right tool for aircraft maintenance is the wrong tool for your Tuesday.
Everyone wants to write with impact. Almost nobody knows the methods.
The techniques are a century old. Choosing between them is the part nobody does, so I built something that does. Testers wanted.
More on Mechanical Writing
Exploring Claude's Output Styles as communication mechanisms.
In Defense of Mechanical Writing
Writing has a job to do. Perhaps we should specify the job.
Impact Advocate (beta)
An agent skill that turns your own GitHub history into a self-review draft, linked to evidence and calibrated to your role. Runs in your own Claude session. Technical beta, testers wanted.
work-ledger
A local reader for your own Claude Code transcripts. It attributes cost to the sessions, prompts, skills and agents that caused it.
mcp-remote-sudo
Narrow, expiring, receipted authority over a remote Linux machine for an AI agent, instead of an open shell. Early design and prototype.