I’ve been using AI tools heavily for about two years. Here’s an honest account of what’s changed, what hasn’t, and what I’ve learned.
What AI is actually good at (for PMs)
First drafts. The blank page problem is real. AI eliminates it. I can go from “I need to write a PRD for X” to a structured first draft in minutes. The draft is never good enough to ship, but it’s good enough to react to — and reacting is much faster than creating.
Structured thinking. When I’m stuck on a problem, I describe it to Link and ask it to help me think through it. Not to solve it — to structure it. What are the key variables? What are the tradeoffs? What am I missing? This is surprisingly useful.
Research synthesis. I can dump a pile of user research, competitive analysis, or data into a conversation and ask for synthesis. The AI is good at finding patterns and surfacing themes I might have missed.
Spec writing. Acceptance criteria, edge cases, error states — the mechanical parts of spec writing that are important but tedious. AI handles these well.
Communication drafts. Stakeholder updates, meeting summaries, status reports. I write the key points, AI writes the prose.
What AI is bad at (for PMs)
Strategy. AI can help you think through a strategy, but it can’t set one. Strategy requires judgment about what matters, what your organization can execute, and what the market will reward. That judgment comes from context AI doesn’t have.
Prioritization. AI will give you frameworks for prioritization. It won’t tell you what to actually prioritize. That requires knowing your users, your business, your team’s capacity, and your company’s bets. AI doesn’t know any of that.
Saying no. The hardest PM skill is saying no to good ideas. AI is constitutionally incapable of this — it will find merit in everything. You have to bring the no yourself.
Reading the room. Stakeholder management, organizational politics, knowing when to push and when to wait — none of this is AI-accessible. It requires human judgment about human dynamics.
My actual workflow
Morning: I start conversations with Link by loading context — what I’m working on, what decisions are pending, what I need to think through today. This takes about 5 minutes and sets up the rest of the day.
Writing: For any document longer than a page, I start with an AI-generated outline. I edit the outline heavily, then ask for a draft. I edit the draft heavily. The final document is mine, but the AI saved me 60% of the time.
Data analysis: I describe what I want to understand, Link writes the SQL, I review and run it. I interpret the results myself — AI is good at writing queries, not at knowing what the numbers mean for my business.
Spec review: Before sharing a spec, I ask Link to play devil’s advocate. What edge cases did I miss? What assumptions am I making? What would a skeptical engineer ask? This catches maybe 30% of the issues before review.
The thing I’ve learned most
AI tools are force multipliers, not replacements. They make good PMs faster. They don’t make bad PMs good.
The skills that matter most — judgment, strategy, communication, prioritization — are still human skills. AI can help you execute on those skills faster, but it can’t substitute for them.
The PMs who will be most valuable in an AI-augmented world are the ones who develop strong judgment and use AI to execute on that judgment faster. Not the ones who outsource their thinking to AI.
Use the tools. But keep thinking for yourself.