AI Policy

Last updated: October 7, 2026

I use AI tools in my work. This page lays out which ones, how I use them, and where I draw the line.

Principles

AI should be used with reliable datasets, not as a replacement for one. It's an input, not a source of truth.

I do use AI to support decision-making, but only inside a defined playbook: when a decision follows a repeatable structure with known variables, I'll run it through AI to pressure-test the logic or surface options I might miss. I don't use it to make judgment calls that depend on context only a human on the ground would have.

I am the final checkmark on all work. Whatever a tool produces, I've reviewed it, and I'm responsible for its accuracy.

What I use

Claude. Ideation, project management, invoicing tracking, and general small business tasks.

Claude Code. Software and tooling work.

Gemini. Editing and refinement.

Local models. I also experiment with models that run on my own hardware. Client information never goes into them.

Client confidentiality and data sovereignty

Client information belongs to the client, and where it lives is part of the work.

For general client work, I use Claude normally. For sensitive client information, I use a separate, ring-fenced system built on Gemini Pro, kept apart from my everyday tools, including the Gemini I use for editing.

Changes
I'll update this page as my tools and practices change. Check back here if you want the current picture.

Questions
Email me at dave@largefraction.com.