Two Half-Finished PRDs for the Same Product Is Its Own Kind of Debt

I had two half-finished PRD drafts for Build AI with AI sitting around. That's its own kind of technical debt before a single feature ships: two different versions of the same plan, neither one complete enough to build from, both quietly assumed to be "close enough" to work with whenever I next opened either one.

The problem with two half-finished drafts isn't that either one is wrong. It's that neither one is a decision. Every session spent reading one instead of the other was a session spent not building, and the longer both sat there unresolved, the easier it got to keep treating the choice between them as something to figure out later.

I spent a day combining both into one 16-section, build-ready document. The biggest addition was a dedicated Agentic Layer section mapping the product's automatic behaviors, plus folding governance requirements directly into the features they constrain instead of listing them separately: an approval gate on the digest, an audit log, a kill switch on anything unattended. Folding governance into the feature list instead of keeping it as a separate document mattered more than it sounds like it should. A separate governance document is easy to write once and never open again. A requirement sitting directly next to the feature it constrains gets read every time that feature does.

Two new database tables came out of the merge: generation logs and automation controls. A new email dependency. The build order grew from 10 steps to 13 once webhooks, cron jobs, and governance documentation got their own explicit steps instead of being assumed.

I deliberately settled one open question by not fully resolving it: whether the product needs a day and night theme toggle at launch. Decided dark-only for MVP, pushed the toggle to a later phase, and wrote that decision into the document instead of leaving it as an unstated assumption. Answering every open question before moving forward is its own way of never shipping. Some questions are worth resolving on purpose by naming them and setting them aside, rather than either forcing an answer or letting them sit unaddressed.

If you're stuck choosing between two versions of the same plan, merging them has been worth the afternoon it takes, even when neither draft is wrong on its own. And if a question doesn't need to be answered to start building, that's permission to defer it on purpose, written down as a deferral, not a gap to feel bad about or quietly forget you haven't resolved.