How we use AI.
We teach teams to check what AI gives them. So here’s where we use it in our own work, and what a person does about it.
Where we use it.
Five places, each with what the AI does and what a person does about it.
AI in our own workAs of October 2026
- Our questionnaire
-
A language model that Cloudflare runs for us reads your last answer, in your own words, and matches it to pieces in our library. That picks the starting points you see at the end.
Only Caleb and Sasank read your answers. The note they read before the call is written by code from your answers, never by a model.
- The next-word game
-
The game on our home page replays real runs of two small models, Phi-3 mini and Gemma 4 E4B, recorded on our own computer in September and October 2026. Nothing is generated while you watch.
The odds you see are the ones the models gave on the day, saved as they ran.
- Research for the library
-
We use AI to help find sources and draft pieces. Every quote and figure is then matched to its source document in a research ledger.
Nothing goes live until one of our founders has said yes.
- Our free packs
-
Claude, Anthropic’s AI, wrote the practice passages in Series 3, on 21 September 2026. Spotting what a machine wrote is the point of that pack.
The pack says so where the passages appear.
- Building this site
-
We use AI coding tools to write the site and the tests it has to pass. Each page is checked in a real browser before it ships, keyboard and screen sizes included.
Each release goes live only when one of our founders says so.
Where it doesn’t go.
These are rules, so they don’t change from one piece to the next.
- Pictures from an image generator. We don’t publish anything one makes, though we help people explore these tools and have tried them ourselves.
- A generated voice. Caleb records every voice-over himself.
- Reviews or quotes we wrote ourselves. We have none yet, and we won’t make any up.
- A fact we haven’t matched to its source.
- Your answers in a training set. Cloudflare says it doesn’t use what we send it to train AI models.
Ask us how we checked something.
Every piece in the library names its sources. If one doesn’t hold up, we want to know.