How we use AI
We use AI every day. A person on our team reads, understands and owns everything that ships.
The short version
- We use Claude to draft, search our own documentation, summarize long threads and write first passes of code.
- Nothing ships until someone on the team has read it, understood it and changed it where it missed the goal.
- We don't ship anything we can't explain.
Code
Every line of code we ship is reviewed by a person who understands it.
AI writes a lot of code quickly, and most of it is more than the problem needs. We have over-engineered things before, so we know what that costs. We treat code we didn't write as code we don't understand yet. We read it, cut what the goal doesn't need and rewrite anything we can't explain.
We keep the simplest version that does the job. Simple code is easier to fix later.
Writing
AI helps us draft. A person decides what we say, checks every claim and number by hand and signs off before it reaches a client or this site.
We keep a written voice guide and a list of approved claims, and AI drafts from them. A person still checks every first pass.
Email and client work
AI can draft a reply. A person reads it, fixes it and sends it.
Passwords and secrets
Passwords, keys and other secrets never go into an AI tool. They live in our password vault.
Our AI tools reach our own documentation through connections we set up. Every time one of them reads or changes our knowledge base, it's logged under the person who connected it.
Questions
If you want to know whether AI touched something we made for you, ask us. We'll tell you what it did and who checked it.
