From PRD to deployed app in a weekend
How a tightly scoped Product Requirements Document and a well-tuned CLAUDE.md collapses the gap between idea and live software.
The classic gap between "I have an idea" and "the idea is alive on the internet" used to be measured in months. With the right scaffolding, it is now a weekend. The two artefacts that make that possible are a rigorous Product Requirements Document and a tuned CLAUDE.md planning file — and together they do most of the heavy lifting before a single line of code gets written.
Why the PRD matters more, not less
It is tempting to think that with AI in the loop you can skip the planning. The opposite is true. AI coders are extraordinary at executing — and ruthless at executing the wrong thing if you let them. A short, sharp PRD is the contract that keeps the work pointed at the actual goal.
The PRDs we generate include:
- A one-line product north star
- Goals and explicit non-goals
- The minimum data model
- The smallest possible API surface
- A milestone list ordered by user value, not by stack layer
CLAUDE.md is the operational manual
If the PRD is the what, CLAUDE.md is the how. It tells the agent which files matter, which patterns to follow, what to leave alone, and what good looks like. A well-tuned CLAUDE.md is the difference between a coding agent that produces a coherent codebase and one that produces a quilt of pull requests.
The weekend
With both files in place, the rest is mostly waiting and reviewing. We typically run two or three loops: scaffold, fill in, polish. Each loop ends with a manual review pass — because no amount of agentic excellence removes the need for taste.
The result is a real deployed app, with real users-of-one, by Sunday evening. From there the question becomes: was the idea worth it? And that — the only question that ever mattered — is now answerable in days instead of months.