Stop Using AI Like an Employee, Start Using It Like a Cofounder
A year ago, I used AI the way most people do. "Write me an email." "Summarize this document." "Fix this bug." It worked, but I was leaving 90% of the value on the table.
The shift happened when I stopped giving AI tasks and started giving it context. Instead of "write a PRD," I started with "here's the problem we're seeing in our onboarding data, here are three possible solutions I'm considering, what am I missing?" The output went from generic to genuinely useful.
AI for Ideation: Your Brainstorming Partner
The hardest part of product work isn't execution. It's figuring out what to build. AI is exceptionally good at expanding your thinking when you give it enough context.
When we were redesigning the onboarding flow at Sonic Linker, I used Claude to pressure-test my assumptions. I described our user persona, the current drop-off data, and my hypothesis about why users were leaving. The AI came back with three alternative explanations I hadn't considered. One of them turned out to be right.
The key is treating it like a conversation with a smart colleague, not a search engine. Share your reasoning. Ask for pushback. Say "what's the strongest argument against this approach?" You'll get thinking you can actually use.
AI for Unbiased Validation
Every PM has confirmation bias. You fall in love with your solution and unconsciously filter evidence that supports it. AI doesn't have that problem.
I've started using AI as a validation layer before committing to a direction. I share the user research, the proposed solution, and ask: "If you were a skeptical VP of Product, what would you push back on?" The responses are surprisingly sharp. It catches logical gaps, questions assumptions, and forces you to tighten your thinking.
This isn't about AI making decisions for you. It's about having an always-available sparring partner that doesn't have political incentives or ego in the game.
Prompt Creation: The Underrated Skill
Most people write bad prompts and blame the AI. The difference between a useless AI output and a game-changing one is almost always the prompt.
Here's what I've learned about prompting effectively: - Give context before giving instructions. "You are helping a PM at a B2B SaaS company that sells to marketing teams" beats "write a feature spec." - Share your constraints. Budget, timeline, team size, tech stack. AI optimizes for what you tell it. - Ask for multiple options, not one answer. "Give me three approaches with tradeoffs" produces better thinking than "what should I do?" - Iterate. The first output is a draft. Push back, refine, and the second or third round is where the real value lives.
Execution: Vibecoding and Beyond
This is where it gets interesting. If you haven't tried vibecoding yet, start now. Tools like Claude Code let you assign AI a tech cofounder role. You describe what you want to build, and it writes the code, sets up the architecture, handles the deployment.
I'm not an engineer. But I've used Claude Code to build and ship this entire portfolio site, including the live Search Console dashboard, the blog system, and the deployment pipeline. That's not "AI assistance." That's a partnership where I bring the product thinking and AI brings the technical execution.
The mental model shift is crucial. Stop thinking of AI as an employee who performs tasks you assign. Start thinking of it as a cofounder who brings complementary skills. You bring domain expertise, user empathy, and strategic judgment. AI brings speed, breadth of knowledge, and tireless execution.
How to Start
If you're still using AI for basic tasks, try this tomorrow: take your hardest product decision this week and have a 20-minute conversation with Claude or ChatGPT about it. Share the full context. Ask for pushback. Explore alternatives. See what happens.
The PMs who figure out how to think with AI, not just delegate to it, will have an unfair advantage for the next decade. The tools are ready. The question is whether you're using them at 10% or 100%.