Building with AI & BMAD

I work AI-augmented. Agentic coding tools and a structured BMAD workflow are a core part of how I deliver, not a novelty bolted on the side. This site was built that way end to end. It's worth being precise about what that does and doesn't mean.

The agent does the mechanical work: reading every artefact, building the inventory of decisions, reconciling indices, running the repetitive sweeps a human does slowly and badly. I do the judgement and the sign-off. That division is deliberate, and I hold a hard line on it. The trail you're reading was collated, not reconstructed. Every material call was written down at the moment it was made, on every piece of work, so that pulling the record together at the end meant confirming a trail that already existed rather than inventing a tidy story after the fact.

Decisions recorded as they were made, not reconstructed. An agent will happily write you a confident history of a decision that never happened, so the human signs off what was actually decided, and nothing gets a tick it didn't earn.

Each decision lives in a short, plain record: what was decided, and why, for whoever reads the code next. Nothing ceremonial, just enough that a future reader (quite possibly me, eighteen months on) doesn't have to reverse-engineer the reasoning from the diff.

That's the honest version of "built with AI". It lets one person deliver at the pace of a small team, but only because the human stays accountable for the calls that matter. The speed comes from handing the machine the work it's good at, not from letting it make the decisions.