Skip the listicle. The five prompts that transformed my workflow do not exist. What changed for me was smaller and slower to see, and it has held up.
I was skeptical of AI tools for years. Not the technology. I have a CS background and spent years building integrations that had to work in production. I was skeptical because most PMs used AI to make faster versions of work that was thin to begin with. PRDs full of filler. User stories polished to a shine that still missed the actual user.
Speed on the wrong work is still the wrong work.
"Forget whether AI can write your PRD. Ask whether it makes you clearer about what belongs in it."
What Changed My Mind
The first time I used Claude seriously, I did not use it to write. I used it to argue.
I was stuck on a call at Kabba. Prioritize the corporate B2B tier or double down on commuter retention. Both had a case. Both had data. I was going in circles, so I opened a session and briefed Claude like a sharp colleague who knew none of the context.
What came back was not a recommendation. It was three questions I had not asked myself. One of them, about the difference in purchase timelines between a commuter and a corporate HR director, reframed the whole analysis. The corporate buyer ran a long sales cycle and made one decision that committed a whole company. The commuter decided again every morning. Those are not two versions of one product decision. They are two different businesses.
That is when it clicked. I had been using the thing to think, and the thinking was the entire value.
The Workflows That Earn Their Keep
I have run Claude across product work for about a year. Four uses have earned a permanent place.
1. Pre-Meeting Preparation
Before a stakeholder conversation that matters, an executive review or a partner negotiation, I brief Claude and ask it to steel-man the hardest objections coming my way. I am not after scripted answers. I want to have met the counterargument before I walk in. The number of gaps this surfaces is embarrassing in hindsight.
2. Discovery Synthesis
After a round of interviews I paste the raw notes and ask Claude to find the patterns, the contradictions, the questions I forgot to ask. The model is not doing the analysis. It is a sounding board that forces me to externalize what I heard. Explaining findings clearly enough for an outsider exposes the holes in my own understanding every time.
3. PRD Pressure Testing
I write the PRD. Then I tell Claude to read it as a senior engineer who has been burned by vague specs before. The feedback is usually obvious in hindsight, which is exactly the kind of gap that sends engineers back to my desk mid-sprint. I would rather find it the night before.
4. Decision Documentation
When I make a real product call, I have Claude help me write a clean decision record. What we chose, what we weighed, what data we had, what we traded away. The record is for my own clarity. If I cannot explain a decision well enough for a smart outsider to follow the reasoning, I have not earned the right to make it yet.
What It Can't Do (And Shouldn't)
Some of the job does not move.
Judgment about what matters. Claude can lay out the options. It cannot tell me which one matters more. That answer lives in organizational context, in relationships, in strategy that exists only in my head. Hand prioritization to a model and you have given away the core of the job.
User insight from lived experience. The insight behind Kabba's commitment architecture, that commuters needed predictability and that a booking fee read as a trust signal, did not come from analysis. It came from standing on a curb at 7am when the minibus never showed. No model gives you that.
Stakeholder relationships. Trust gets built in conversations, over time, through judgment people watched you exercise. Claude can prep you for the room. It cannot sit in your chair.
The Bigger Win Is Team Time
The most underused move with Claude in PM work has nothing to do with personal productivity. It gives time back to the whole team.
Count the hours your team burns on low-leverage writing. Meeting summaries, status updates, briefings, records of decisions everyone already remembers. Every one of those hours is an hour stolen from discovery and the thinking that actually moves the product.
A few ways I claw it back:
- Sprint retrospective synthesis. Feed in the raw retro notes and ask for the patterns across the board. A 45-minute synthesis turns into a 10-minute review.
- Async decision alignment. When a call needs input from several stakeholders, have Claude draft the brief. It lands around 70% right. The point is that everyone reacts to the same framing, which kills the interpretation drift that wrecks async alignment.
- Onboarding. Building a context doc for a new engineer or designer, I have Claude structure what I already know. I still supply the substance. The scaffolding that used to eat an afternoon is done in minutes.
- Spec review before handoff. Claude reads the spec as a hostile engineer before the real ones do. It catches the obvious gaps faster than I will, fresh off writing the thing.
The pattern holds throughout. Let the model carry the structured work so people spend their hours on judgment. I am not trying to automate product management. I am trying to buy the people doing it more room to do it well.
The Real Leverage
What makes a PM hard to replace is rarely writing or raw analysis. It is holding a messy product and a messy org in your head at once, calling it under uncertainty, and pulling a room of people who want different things toward one direction.
AI makes you faster at everything else. Used well, that hands you back the time and the head space for the part only you can do.
That is the leverage. A quicker PRD was never the prize. The prize is room to think clearly about the problem underneath it.
"Give me six hours to chop down a tree and I will spend the first four sharpening the axe." The tool does not swing the axe. It sharpens it.
Start with one workflow. Find the place your process wastes the most and put it there. The PMs who compound with these tools raise the floor on their own thinking. The ones chasing fastest output plateau.