POST 02 · BUILDING WITH AI

The Process of Building Products with AI

BY SHYAM GOLE · 7 MIN READ

AI didn't change what a PM does. It compressed how long each step takes — some steps from days to hours, some not at all. After a year of building QuickHub features with AI in the loop (including Quinn, our own AI assistant), here's my honest map of the process: where AI is a rocket, and where it's just a rubber duck.

Stage by stage

• 1 · Scope — AI as research analyst. Interview notes, competitor teardowns, market scans: what used to take me a week of synthesis now takes an afternoon. I feed raw notes in and get themes, quotes, and contradictions out. The catch: AI synthesizes what you collected. If you talked to the wrong users, it summarizes the wrong users beautifully.
• 2 · Spec — AI as first-drafter and red-teamer. I draft the PRD skeleton with AI, then flip it: "you're a skeptical senior engineer — list every ambiguity that would block sprint planning." It catches edge cases before my engineers have to. The decisions inside the PRD — what's in, what's out, what "done" means — are still mine.
• 3 · Prototype — the biggest unlock. A clickable prototype used to mean days in Figma or a favor from an engineer. Now I get a working mock in hours and put it in front of users the same week. For Quinn, we tested conversation flows as prompts before writing a single line of product code.
• 4 · Validate — stubbornly human. AI can draft the interview script and synthesize the notes afterward, but the interview itself doesn't compress. Watching a restaurant owner hesitate over a button teaches you things no summary can. This is the stage where I've learned to slow down on purpose.
• 5 · Ship & decide — judgment doesn't compress. What to build next, what to kill, when "good enough" ships — AI will happily argue any side of these. The call is the job. That's a whole post of its own: the judgment that makes the difference.

What this did to my week

The honest summary: AI moved my time from producing documents to talking to users and making decisions. The parts of the job that felt like typing got cheap. The parts that felt like thinking got more valuable — because now everyone can generate options, and the differentiator is choosing well.