I’ve been using AI for a year as a PM. Here’s what actually changed.

A year ago I started using AI seriously in my day-to-day work as a PM. I’m not here to tell you it changed everything — I want to reflect honestly on what actually shifted and what didn’t.

Because here’s something nobody really talks about: a big chunk of being a PM isn’t building product. It’s maintaining the infrastructure around it. Documents that need updating. Statuses that need syncing. Alignment that slowly drifts the moment you stop pushing it. That overhead is real, it’s part of the job, and it keeps growing.

So I broke it down into four areas — Writing & Documentation, Alignment & Communication, Prioritization & Decision-making, and Tracking & Maintenance — and looked honestly at where AI made a difference, where it didn’t, and one area where I’m quietly glad it hasn’t solved anything yet.

1. Writing & Documentation

Writing is at the core of what PMs do. PRDs, epics, stories, Slack messages, stakeholder updates — the list is long, and the bar for consistency is high. Before AI entered my workflow, I knew there was room to improve. Too many formats, too many places, and no matter how much effort you put in, information starts to mutate the moment it touches multiple teams.

The real frustration wasn’t writing itself — it was the feedback loop. I’d share a doc with other PMs or my lead, wait for comments, incorporate them, then wait again. Good process, but slow. Every round of feedback was a pause in momentum.

That changed when we got access to Gemini at work. Suddenly I could debate, challenge, and refine a document in real time — no waiting, no scheduling. The quality improved and so did the speed. I took it further by setting up GEMs, which let me work through PRDs, epics, and hypotheses with a consistent structure every time, without rebuilding the process from scratch on each document.

This is the clearest win AI has given me so far. I write faster, more consistently, and the documents that come out are stronger than what I was producing before.

2. Alignment & Communication

Alignment is probably where PMs spend most of their energy. You’re constantly switching context, talking to engineers in APIs, latency, and edge cases, then turning around and translating that into timelines and business impact for stakeholders. The same information, two completely different languages.

AI can help you prepare for that. Better documents, clearer summaries, faster context-switching between modes. But the actual negotiation, reading the room, pushing back at the right moment, finding the version of the plan everyone can commit to, that’s still entirely on you.

I haven’t found an AI solution that meaningfully changes this, and honestly, I’m fine with that. It’s the part of the job I enjoy most. Some things are better left unsolved.

3. Prioritization & Decision-making

This is where I think AI is about to change everything for PMs, and we’re not quite there yet.

Right now, prioritization relies heavily on tribal knowledge. You ask around, you dig through Jira tickets from two years ago, you try to reconstruct the context of an A/B test that ran before half the team joined, and you dig deeper in your memory. It takes time, and even when you find the information, you’re never sure you have the full picture. Decisions get made on incomplete history, and sometimes that means repeating mistakes that were already made and already documented, just somewhere nobody thought to look.

What’s coming could be huge. Some tools are already starting to build toward this, the ability to cross your current initiative with everything that came before it. Not just searching old docs, but actually connecting the dots between past decisions, results, and what you’re trying to do now. For a PM, that’s not a productivity improvement. That’s a different way of working entirely.

I don’t have this at my company yet. But I think about it a lot and cannot wait until I can use this across my current backlog.

4. Tracking & Maintenance

Tracking and measuring results is an area where AI has started to show up in my workflow, but only partially.

In most companies, making sense of experiment results requires real statistical knowledge. We’re lucky to have data experts who help interpret outcomes, but as a PM you still need to add your own layer on top, the product context, the edge cases, the “yes but” that the numbers alone don’t tell you. AI can help bridge that gap, adding a third perspective that combines the data with the broader context of what you’re building.

We’ve had some early wins and created some interesting analysis based on AB results. But where it keeps falling short isn’t the analysis itself, it’s the context. Getting AI to truly understand the environment around the data, the history of the feature, the nuances of how the experiment was set up, that’s still the hard part.

The promise here is real though. As we get better at feeding the right context into these tools, I think this becomes one of the most exciting capabilities for PMs, not just understanding what happened, but understanding why, faster and more completely than before.

One year in

A year in, this is where I land: AI has genuinely made me a faster and more consistent PM. The writing problem is largely solved. Alignment remains human. Prioritization and tracking are works in progress, exciting ones.

But the most honest thing I can say is this: AI hasn’t changed what good product work looks like. It’s just removed some of the friction that was getting in the way of doing it. The overhead that used to eat into my week is lighter. That means more time for the parts that actually matter, and those parts are still entirely up to me.