AI in the build, with a person on the end of it
We use AI where it is faster and where the result can be checked (writing code, drafting interfaces, moving data) and a person reads every line of it before it ships.
The tools, named.
Claude Code
Anthropic’s terminal agent, for repo-wide changes run on a branch and reviewed like any other pull request.
Claude Design
Anthropic’s design tool, for putting a first visual on the table in minutes instead of days.
OpenAI Codex
A second coding agent, for the tasks that suit a different model better.
Gemini CLI and Cursor
Google’s models at the command line, and the editor where the smaller edits happen.
Why this, and when not.
Why we use them
They remove the typing, not the thinking. The work is still designed, reviewed and tested the way it always was.
When we would not
Anywhere a wrong answer is expensive and hard to verify. Nothing generated reaches production unread.
Services built with these.
Building with AI?
A 45-minute call with the engineer who would lead the build. You leave with a scope sketch either way.