Skip to content

Using AI

image-20260618-090831.png

We actively encourage using AI — especially Claude Code for automation tasks. It automates manual work, and speed is our edge. A few common-sense rules so we use it well, without turning into a bureaucracy.

  1. Use it. Default to reaching for Claude — code, drafts, analysis, research, grunt work. Being good with AI is part of the job here.
  2. No confidential customer data in prompts. Don't paste a customer's confidential designs, models, files, or data into AI. (Our Team plan doesn't train on our data — but that's not the point; NDAs and customer trust are.)
  3. Recreate an equivalent case instead. The key trick: when you want AI's help on a customer's problem, build a made-up example that reproduces the same technical problem with none of the confidential parts, and work on that. It's exactly what we ask customers to do when they can't share under NDA — and it works very well.
  4. You own the output. AI drafts; you check. Verify facts, code, and numbers before they reach a customer or go into the product. Don't ship what you didn't review. Decisions stay yours.