We use it to do the work, not to describe it. Our own engineering tooling is AI-assisted, which is why a small team can carry projects across software and hardware at once. The important constraint is that AI output is treated as a claim to be verified, never as a result. Anything it produces goes through the same checks a person's work does, and where we cannot verify something, we say so rather than shipping it with confident wording. An unverified answer delivered quickly is worse than a slower one, because it manufactures confidence.
