The operating philosophy
How I work with AI
Faster to produce, sharper for having been argued with, still answerable to the work.
AI does two things here: It speeds the work up, and it pressure-tests the thinking behind it before a client ever sees the result.
The second use is the one that matters, and it only holds up because of the controls sitting under it. So the work lands faster, sharper for having been argued with, and still answerable to a real standard of accuracy and voice.
01
As an accelerator.
AI closes the distance between raw material and a finished read. Transcripts, data exports, research — structured into decision-ready output in hours. A client meeting goes in as a transcript; a stakeholder-ready recap or a strategic brief comes out the same day, not the same week.
02
As a challenger.
I use AI to argue against my own recommendation before a client hears it — pointed at the weak assumption, the risk I haven't named, the place the logic quietly stops holding. Its job is to find where the argument breaks, not to agree with it or make it sound finished.
A point of view doesn't leave here until it has survived being told it's wrong.
The controls.
The challenge is only worth anything if it's honest. So every claim, number, and quote is checked against the actual source before it ships — the AI isn't allowed to assert something it can't point to. It's held to disagree rather than flatter. And the standards carry across a long project instead of eroding halfway through.
3,000+
Working sessions running this method against real client work.
4 yrs
Of published thinking, written in the open and on the record.
Every
Claim, number and quote checked against source before anything ships.
The point of view.
AI is infrastructure for better thinking, not a way around it. The point was never that the work gets done faster — plenty of tools do that. It's that by the time a client sees it, the thinking has already been contested, checked, and rewritten. Speed without that is just faster mediocrity. This practice is built to not trade one for the other.
Where it shows up
The method is only visible in the work it produces.
Thirty minutes is usually enough to tell whether the problem in front of you is one I should take.