Assisted engineering is now normal. Being honest about how it is governed is not. This page is what our engineers actually work inside, published so you can hold us to it.
What the kit contains
- Assisted editors and review tooling, configured per project rather than per developer
- A shared prompt and context library for this codebase, kept in the repository
- Generated-code review rules that are stricter than our hand-written ones
- Evaluation harnesses for any feature that calls a model
- Cost and usage reporting, per project, visible to the account team
The rules that matter to a client
Client code and client data do not go into tools that train on them. Assistance is allowed for writing, refactoring and test generation. It is not allowed to be the last reviewer of anything. Every change is read by a named engineer before it reaches your branch, and that has not changed.
Where it genuinely helps
Tests, migrations, boilerplate, documentation, the second implementation of a pattern already agreed. Work where the shape is known and the typing is the cost.
Where we keep it out
Architecture decisions, security-sensitive code, anything touching money or personal data, and any place where a plausible-looking wrong answer is expensive to detect. QA and testing does not get more relaxed because a tool wrote the first draft — it gets less.
Why we publish this
Because the alternative is a client discovering it from a commit message. If your procurement team has a question about tooling, ask it during discovery. We answer in writing.
What it is not
It is not a productivity claim. What the kit measurably saves is on productivity gains, with the caveats attached.
Back to the AI map.