I Spent a Weekend With a Head of Product at an AI Company. I Haven’t Worked the Same Way Since.

Six months ago I spent a weekend with a good friend who runs the product department at an AI company. The plan was to do nothing. Long walks, good food, zero laptops.

That lasted until Saturday breakfast.

He asked how I was using AI in my work. I’m a product consultant for SaaS teams across different industries, so I gave him my proud answer. I write PRDs in a fraction of the time. Roadmap updates too. Release notes practically write themselves.

He nodded. Then he said: “So you write documents faster. Who reads them?”

I did not have a good answer.

He opened his laptop (so much for zero laptops) and showed me the doc he uses to hire PMs. It lays out what a good PM looks like in 2026. I read it over coffee and quietly realized I’d fail half of his interview.

By Sunday night I had a page of notes and a mild identity crisis. I’ve spent the six months since rebuilding how I work with my clients, their engineers and their stakeholders. Here’s what changed.

“If It’s a Format, Automate It”

His first rule. Specs, roadmap updates, check ins, release notes. If the work follows a format, AI handles it.

I was already doing that part. What I missed was the second half. Every hour AI gives back has to go somewhere specific: the customer, the business, the decision. If it goes into more documents, it was wasted.

His test: if AI made you produce more documents faster, it failed. If it moved you closer to the customer and the business, it worked.

By that test, I had been failing at record speed.

Now the formatted work runs in the background and the hours go to my clients’ customers. I’m on more customer calls. I query the data myself instead of waiting on an analyst. I bring answers to meetings, not questions.

“A Prototype Is an Argument”

Second thing. He doesn’t settle debates with documents. He builds the thing, wires it to real product data, and puts it on the table. A working prototype ends a discussion that three PRDs couldn’t.

So I started building. Vibe coded prototypes, connected to my clients’ real APIs. No more static mockups. I put competing versions in front of real customers in test sessions I run, and watch which one they actually use. The version that wins becomes the spec.

The first reaction from engineering is usually a polite panic. A PM with code? Who maintains this? Is this going to production?

His three rules fixed that, and I now set them on day one with every client.

• The customer rule. The moment a real customer relies on it, engineering owns it. A private beta with five design partners is still production. A customer can test a prototype with me in a session. They can’t start depending on it.
• The pager rule. Whoever gets woken up at two in the morning owns the code. That should never be the PM.
• The codebase rule. A prototype lives in its own repo. It enters the main codebase only if an engineer adopts it.

Once engineers know a prototype is input and not a mandate, and that they keep the right to throw it away, they stop worrying and start using it. And I get to build more, not less.

My short version, which I now repeat to every team: prototypes are for learning. Anything customers rely on, engineering owns.

“Treat Sales Like Users”

Third. He treats the relationship with sales and CS as a product surface. Every handoff, update and report that can be automated gets automated.

I used to be the human relay between R&D and go to market. Now I set up agents with my clients that keep GTM informed without anyone chasing anyone. What shipped. What’s coming. Which deals in the pipeline it unblocks.

Two of them do most of the work. One scans the CRM and emails the rep when a prospect or client hasn’t changed status in two weeks: worth checking in? The other pulls customer and stakeholder requests from Intercom and Slack every week, scores them against the prioritization rules the team agreed on, and sends a ranked list.

Fewer pings for the product team. A better informed sales team. Hours of triage meetings gone, and nobody misses the old status meeting.

And they pass the pager rule. They’re internal tools. If one breaks, a list shows up late. Nobody gets woken up.

“Execution Is Cheap Now. Judgment Isn’t.”

This one stuck.

When anyone can generate a backlog in ten minutes, the backlog isn’t the job. The job is reasoning from the business. Tying every feature to revenue, retention or cost. Making the call when the data is incomplete, and defending it with your name on it.

He listed what AI doesn’t do. It doesn’t choose the problem worth solving. It doesn’t carry a decision or its consequences. It doesn’t sit in front of a customer. And it doesn’t create conviction. It only makes conviction cheaper to test.

That last line changed how I run discovery with clients. More experiments, smaller and faster, aimed at closing product market fit gaps. Not at shipping more.

“Ask the Uncomfortable Questions”

The last one came from his hiring process, and I stole it for my consulting work.

His best candidates don’t wait to be briefed. They ask how the company adopted AI internally and what changed because of it. Which workflows got rebuilt, and which only got talked about. What leadership got wrong on AI last year. Who owns the context layer, and whether one even exists.

I now ask those questions in my first week with every client. The answers tell me more than any onboarding deck. Sometimes the answer is a long silence. That silence is usually where the real project is.

So, Six Months Later?

I’m a better PM than I was that Saturday. Not done. Some of what I tried didn’t stick, and I still learn something new every time my friend and I talk.

But the shift is real. Less time producing documents. More time building, deciding and talking to customers. The part of the job I signed up for in the first place.

Fewer documents. More prototypes. More time with customers.

Thanks to my friend for a weekend that cost me a few comfortable habits.

If you’re a PM or a product leader going through the same shift, I’d love to hear what’s working for you. Grab 30 minutes with me or find me on LinkedIn.