How I use AI

Machines bring the reach, people make the calls.

I use artificial intelligence every day: to read contracts no person could get through, to build prototypes and to organize my own notes. I also investigate how governments use and buy it. So it seems fair to explain how I use it myself.

AI searches, sorts and suggests. A person decides.I use it to find patterns in millions of public records, classify documents, cross-reference sources and write drafts. What gets investigated, what gets published and how it is interpreted is always decided by a person.
No figure is published without checking it against its source.Every data point coming out of an AI system must be traceable to the row or document that supports it. If it cannot be checked, it is not published. I would rather report a smaller, defensible number than a bigger, wrong one.
Sensitive material stays out of commercial models.Leaks, source identities and non-public personal data are kept out of third-party AI services. When the material is delicate, I work from public sources or with tools running on infrastructure we control.
I say when AI was involved.If an investigation, a newsletter or a product used AI, I explain it in the work itself, as in State Algorithms: AI helped find and sort the purchases, while verification and interpretation were human. I also disclose it when I use it to prepare proposals.
I design so errors show.AI makes mistakes. I do not count on it to stop: I build workflows so every error has a place to be caught before it reaches a reader.
Government AI gets audited too.Governments in the region buy facial recognition, video surveillance and automated systems with little transparency. Investigating those purchases with the same open data is part of the job.

This website was also built with AI. The code, the design and much of the writing were done with Claude, by Anthropic. I decided what to say and how, and every fact and link was checked against its source before publishing.

Last reviewed: October 2026