Eleven Days Before a Deadline, I Almost Scrapped a Finished PRD

Eleven days before a hackathon deadline, I had a finished PRD and nine build steps already mapped out for Pigeon, an email-voice-matching tool for course creators. Then a shinier idea started pulling at my attention, the way a new idea always does right when the current one has entered the unglamorous middle stretch.

I know that pull well enough by now not to trust it alone in a room. Instead of spiraling over it by myself, I uploaded both PRDs to Claude, the finished one and the new shiny one, and asked for a real answer. Not encouragement. Not "both are great, follow your excitement." An actual answer.

What came back was blunt, and I mean that as a compliment. The new idea's flashiest feature, an app that improves itself based on usage, would take longer to demo convincingly than the days I had left. Not impossible to build. Impossible to build and make legible in a two-minute demo, which is the actual constraint a hackathon puts on you. The new idea's target user was also too broad, vague enough that it would be hard to make a judge feel like it was built for anyone specific. And pivoting meant throwing away architecture I'd already finished, restarting design decisions from zero, burning two to three of my eleven remaining days just getting back to where I already was.

Then it asked me the question that actually made the decision: which product will you pitch with more conviction?

Pigeon stayed. Not because the new idea was bad. Because the honest answer to that question was Pigeon, once I stopped looking at the shiny idea's potential and looked at what eleven days could actually produce, credibly, for each option.

The second-opinion prompt

I'm considering switching from [current plan] to [new idea]. Don't
encourage me. Give me the argument against my own excitement: what
would actually fail, what's the realistic time cost of switching this
close to the deadline, and which option would I pitch with more
conviction if I had to defend it in front of someone right now?

Here's the lesson, stated as a framework you can use directly, because this one is genuinely reusable outside of my specific situation: give the AI both options in full, not a summary of one and a vibe about the other. Ask it to argue against your excitement, specifically, rather than argue with you generally. Then look for a decision you can actually commit to, not validation of whatever you already wanted to hear when you opened the conversation.

The excitement of a new idea is real information. It's just not the only information, and it's rarely the information that should decide what you ship eleven days out from a deadline.