You're Using AI, but You Haven't Adopted It ... Yet.

I’m going to tell you about one of the worst two weeks of my PMM career, and how AI could have saved me from them.

Early in my PMM career I was handed an MPF (that’s what we used in that first role, today we use MSDs, but you know this by now), the messaging and positioning blueprint that everything else in the product asset portfolio hung off of. The L1 deck, website copy, datasheet (remember them), even the language to be used in an upcoming keynote. Getting it right wasn’t optional. Getting it wrong means everyone downstream is building on a sandcastle with an incoming tide.

So I did what you do. I locked myself in a room (literally, this was pre hybrid-work) and I wrote. And wrote. And rewrote. I picked one narrative angle, fell in love with it the way you fall in love with the first draft of anything, and spent two weeks polishing it before I showed it to a single other human being. Like a dumbass.

Then I brought it to a review meeting. You can guess what happened.

It got torn apart. Not because the writing was bad, but because I’d bet the entire document on one interpretation and perspective, and most of the room saw it differently. My PMM leadership thought I’d buried the lede. Product thought I’d overpromised on a feature that was still six months out. My GM thought it was “fine,” in the tone of voice that means it is not in fact fine. Two weeks of work, and I was defending a single bet instead of comparing options.

That was a meeting to remember, for all the wrong reasons.

The fallacy of using AI

Fast forward to today. Many PMMs I talk to have found a way to speed up that exact same broken process. They still write one MSD draft, alone, then bring it to review. The only difference is that the robots helped them type faster. Same linear path: draft, defend, revise, repeat. Same sunk cost by the time anyone else sees it. Same bruised ego in the room when it doesn’t survive first contact in review.

That’s “using” AI. A faster horse and cart. Nothing about the shape of the work changed, just the speed of one step in it.

Adopting AI asks a different question

Here’s what I’d do differently now, and what I coach the PMMs I work with to do: don’t write one MSD. Write five fast versions, each one making a different bet. One leans into category creation. One plays it safe and competitive. One leans on customer voice and pain. One is unapologetically technical for a PLG (product-led growth) audience. AI can spin up rough but indicative versions of all five in the time it used to take to grab that morning coffee.

Then, instead of a big review meeting where you defend your one bet, you get quick reactions on all of them from Sales, Product, and your GM, maybe even a few trusted customers if you can swing it, before you’ve spent a single minute polishing prose. You’re not asking “is this good?” anymore. You’re asking “which of these is even worth being good at?” That’s a completely different question, and it’s one you couldn’t afford to ask when every draft cost days of time.

That’s the part that gets missed in most “AI for product marketers” content. The real shift isn’t speed. It’s that AI makes exploration cheap enough to run in parallel instead of committing in sequence. And once exploration is cheap, you start asking questions you never would have bothered asking before. Could this be a category of one? What if the ICP (ideal customer profile) isn’t who we think it is? What if the real story is the one Sales keeps telling on calls that never made it into any deck? Those aren’t faster versions of old questions. They’re new destinations you couldn’t see from the old road.

The part AI still can’t do

Note: none of this means the human gets to leave the room.

AI is a vibe engine. It’s brilliant at simulating five different tones, five different narrative bets, five different personas reacting to your positioning. What it can’t do is know which one is actually true, which one holds up under a hard question from a skeptical VP, which one has the emotional resonance that makes a buyer lean forward instead of nod politely. That’s still you. That’s the 20% that was always the actual job, and it’s the part AI hands back to you once it’s done the 80% of execution grunt work. Zero-Trust fact-checking still applies here too. Every one of those AI-generated drafts is a hypothesis, not a fact, and you’re the one who has to verify it against real customers before it goes anywhere near the board deck.

The PMMs who just use AI get a faster version of a process that was already limiting them. The ones who adopt it get something better: permission to ask “what else is possible” before spending two weeks finding out the hard way.

I’d be disappointed if you read this and just started asking the robots to type your next MSD faster.

Adam