AI agents such as Claude Code write the code and tests for every ArcLayer product. A designer and an engineer own every decision and review every change before it ships. This note describes where that line sits during a build.
What the agents do
- Write feature code against the design system and the agreed scope.
- Write tests alongside that code, so each change arrives with a way to check it.
- Run the tests and fix failures before a person sees the change.
What people do
People decide what gets built and how it should behave: scope, trade-offs, interface decisions and anything a user will notice. A person also reviews each change. They read the diff, run the product and check the result against the design system and the decision log. Nothing merges or ships without that review.
A change, step by step
1. Scope a person writes the task and how to check it
2. Build the agent writes the code and its tests
3. Test the agent tests the change and fixes failures
4. Review a person reads the diff and runs the product
5. Release the change merges and shipsAn agent can go round steps 2 and 3 many times while a person works through one review, so review is the step that sets the pace of a build.
Keep changes small enough to read
A reviewer can check a change to a few files line by line. A change that touches forty files gets skimmed. Ask the agent for small, single-purpose changes, each with its own tests, so the person reviewing can read every line.
Who is accountable
When something breaks, the answer to "who owns this?" is a named person. The agent wrote the code. The designer and engineer who reviewed it own the result.