The world of Product Management appears to be in a state of immense flux. At least that’s what LinkedIn wants me to think. Despite LinkedIn, I do feel that there is some level of it. While there is a lot of talk about AI and its impact, and most of what I see for PMs is about using agents to scale yourself - so your “impact” can be bigger across the discovery, writing, planning, and stakeholder work that fills a PM’s week.

The clearest example of this for me has been losing our Product Marketing Manager.

She helped me scale in a way I didn’t fully appreciate until she was gone. I could trust her with GTM end to end, without having to do all the thinking myself. What she actually did was harder than it looked. She would challenge me to articulate what the user should see in a feature for it to work for them - and because she worked at FareHarbor too, she could ground that conversation in the reality of the environment we operated in. Contextualising was easy, because the context was shared. From there she’d come up with two or three messaging options, get my take, share her own view on what would and wouldn’t land, and then make a judgment call on the investment - is this a blog, a dashboard notification, both? She’d schedule the comms with the client engagement team and get the material into the monthly product digest if it warranted that.

With her gone, I started rebuilding the workflow with AI. The first milestone was the execution layer - filling templates with the core information, the problem we’re solving, the measures of success, the target audience. That works fine.

The second milestone is harder. It’s the framing. How we communicate the value to the target audience. I’ve been leaning on my own notes and hers to train the agent toward the language she used, and for small to medium impact releases, it’s working reasonably well.

For the larger releases, it isn’t. I’m still going to the Director of Product Comms and the enablement specialists to nail the narrative. That gap doesn’t surprise me. The thing my PMM brought to large releases wasn’t messaging output - it was judgment built on shared context that I can’t easily transfer to a model.

There’s also a layer I hadn’t fully registered until recently. She wasn’t just helping me with each release - she was measuring what worked after we shipped, and we’d have check-ins to discuss the results and the next steps. That feedback fed into the next release. With the agent, every release effectively starts from scratch. I’m carrying that loop in my own head from my time working with her, which is part of why it’s holding up for small and medium releases. My judgment is the loop. For larger ones, the Director of Product Comms is, but his time is limited.

The uncomfortable thought is that the loop itself was the thing getting smarter, release over release. I won’t notice the cost of losing it in any one release. I’ll notice it a year in, when the messaging starts to feel generic in a way I can’t quite pin down.

What made any of this possible is that I already had a way of working with her. I knew what good looked like. I had a baseline. I could tell when the agent’s output was off, because I’d seen better.

I wonder if AI would have been useful here if I didn’t have that baseline. If I was a new PM, or if I’d never worked with a strong PMM, what would I be calibrating against? Most likely the AI’s own output, which is a thin reference point.

This is the same thing I’ve been circling in the other posts. The tool amplifies what you bring to it, and quietly underperforms when there’s nothing there to amplify. The AI didn’t replace my PMM. It let me hold the line on the parts of her work I already understood, and made it clearer than ever which parts I didn’t.