You walk out of the strategy meeting feeling great. Everyone loved the new feature idea. Engineering said it was feasible. Marketing is already brainstorming launch campaigns. The roadmap slot is secured. Not a single person raised a serious objection.
This is the most dangerous moment for a product manager.
Unchallenged ideas are fragile. Groupthink is a comfortable path to a failed launch. You need conflict. You need someone to poke holes, to ask the awkward questions, to find the fatal flaw before your customers do. Your team, incentivized to build and ship, often isn't the right group for this.
Your AI is.
Stop asking your AI assistant to summarize your meeting notes or draft your status reports. That’s the co-pilot model—a helpful but passive assistant that polishes the work you’ve already done. It’s a low-leverage use of a powerful tool. You need to flip the script. Your AI shouldn't be your scribe; it should be your sharpest, most ruthless critic. A sparring partner that never gets tired, never worries about your feelings, and exists only to make your ideas stronger by trying to break them.
The Co-Pilot Trap
We’ve all seen the demos. Paste in raw notes, get a clean summary. Describe a feature, get a first-draft PRD. This is fine. It saves a few minutes. But it also reinforces your existing perspective. An AI co-pilot, trained to be agreeable and helpful, will gladly organize your biased inputs into a tidy, coherent, and equally biased output.
It will smooth out the messy parts, fill in the gaps with plausible-sounding text, and hand you back a polished version of your own thinking. This isn't strategy; it's confirmation bias with an API key. A true sparring partner doesn’t just help you write faster. It forces you to think better.
Train Your Sparring Partner: Three Critical Drills
Treat your AI like a new hire on the red team. Give it a role, a personality, and a clear objective: to challenge you. Here are three practical exercises to turn your helpful assistant into a strategic weapon.
Drill 1: The Pre-Mortem Interrogation
Before you write a single line of code, you need to imagine the feature has already launched and failed spectacularly. Your AI can simulate this better than anyone.
The Scenario: You're about to kick off a project to add social "sharing" features to your B2B analytics dashboard.
The Prompt:
"You are a deeply skeptical, data-driven Head of Product. I am proposing we add social sharing features to our B2B analytics tool. This project has just launched and failed catastrophically, losing us our top 3 enterprise accounts. Write me a brutal but specific five-point pre-mortem explaining exactly why it failed. Do not be gentle. Focus on unstated assumptions, resource misallocation, and customer value mismatch."
This prompt forces the AI out of its "helpful" mode. You’re not asking if it will fail, but why it did. It will mercilessly attack your idea, pointing out that B2B users don't want to "share" sensitive data on social media, that the engineering effort cannibalized a critical security update, and that the success metric was a vanity one.
Drill 2: The "Steelman" Debate
A "strawman" argument is a weak version of your opponent's position that's easy to knock down. A "steelman" is the opposite: the strongest, most compelling version of the counter-argument. Making your AI build a steelman against your own strategy is a powerful way to pressure-test it.
The Scenario: You believe the best way to increase user engagement is to build an in-app community forum.
The Prompt:
"My strategic hypothesis is that building an in-app community forum will increase user retention by 15%. I want you to argue against me. Construct the strongest possible 'steelman' argument for why this is a terrible idea and we should instead invest every available resource into improving core product performance and reliability. Use metrics, user psychology, and opportunity cost in your argument."
The AI will now become your fiercest internal competitor. It will argue that forums require heavy moderation (a new cost center), that they often devolve into support queues, and that the perceived value of community is dwarfed by the daily frustration of a slow-loading core feature. It will quantify the opportunity cost. Now you have a real debate.
Drill 3: The Unstated Assumptions Excavator
Every product spec is built on a mountain of invisible assumptions. Your job is to find them before the market does.
The Scenario: You've just finished a one-page spec for a new notification system.
The Prompt:
"Read the following one-page project brief. Analyze it and list every single unstated assumption I am making about our users, our technical capabilities, and the market. For each assumption, state the potential risk if it proves to be false."
The AI will act as an archaeologist for your own logic. It will point out that you're assuming
Generated by Reportify AI — Automate your team's status reports, standups, and weekly updates. Try free →