Tune into any podcast about product management or tech these days and the topic will surely revolve, in one way or another, around AI and LLMs. So here’s another piece of work to that effect.

A lot of people get AI wrong. I’ve watched first-hand how product teams and management get swept up in the excitement, throwing ChatGPT at everything from PRDs to customer interviews to writing emails. And yes, this also includes all the LinkedIn posts these days. 99% of it is polished-sounding nonsense that nobody wants to read. AI slop, as the kids call it.

But here’s what I’ve learned after months of experimentation: AI is genuinely useful for product development. I have literally 10xed my work output, and I’ve even branched into other skills that would have taken me a lot longer without AI.

That’s why it’s so shocking to me that a lot of people still don’t know how to properly apply AI in their day-to-day operations.

The Prompting Problem

Let’s start with the obvious: people suck at prompting.

I’m guilty of this too. Early on, I’d type something vague like “write a PRD for a new feature” and wonder why the output was generic rubbish. The problem here isn’t the AI. It’s that most of us treat it like a magic box instead of a tool that requires clear input.

Good prompting takes practice. You need context, constraints, and examples. But even with perfect prompts, there’s a bigger issue: not everything should be AI-generated in the first place.

High Stakes Requires Human Hands

I have a simple principle: if the stakes are high, keep humans in the loop. If they’re low, let AI handle it.

What do I mean by this?

Well, I wouldn’t send an AI to a customer meeting. I wouldn’t let it write the strategy deck for my next board presentation. High-stakes work requires judgement, empathy, and the ability to read the room. AI can’t do that. (Yet!)

Think about a PRD. The point isn’t just to document features. It’s about communicating an idea to a specific audience. Engineers need technical clarity. Designers need user context. Stakeholders need business rationale. A good PRD reflects understanding of these different needs. AI doesn’t have that understanding. It can string together sentences that sound right, but it’s missing the nuance that comes from actually working with these people.

And to be honest, nobody wants to read AI slop. If you’re going to ship a document with your name on it, make sure it reflects your thinking, not a language model’s pattern matching.

Equally, there’s no substitute for human-led discovery where people actually feel like they’re moving the problem forward together. That’s the magic of a good workshop: everyone leaves feeling like they contributed something meaningful, like they helped crack the problem. To me, that’s the real beauty of product management. AI is nowhere close to be able to recreate that.

Where AI Is Actually Useful

Right, so where does AI fit?

It’s all the boring stuff that can be automated.

Competitive research. This is where I’ve found AI extremely useful. Need to understand how my competitors approach pricing? Want to synthesise what users are saying about a feature across Reddit, Twitter, and review sites? AI can pull this together faster than you can open ten browser tabs. It won’t give you the strategic insight (that’s still your job) but it’ll give you the raw material to work with. Simply amazing.

User feedback synthesis. If you’ve got hundreds of support tickets or survey responses, AI can help identify patterns. I want to add a disclaimer to this though because again, based on my experience, it’s not at a level where it can replace your qualitative research skills, but it’s a decent first pass for spotting themes. Just don’t trust it blindly. Always validate the patterns yourself.

Project management grunt work. Monthly release notes? Metrics reports for VPs who just want the headlines? Updates on sprint progress? This is low-stakes documentation where the format matters more than deep insight. I pretty much resort to AI to draft these and everyone saves time.

The key takeaway here is that AI is awesome for low-stakes grunt work, not for anything requiring judgement or insight.

Final Words

We live in exciting times, there’s no doubt about it. But the AI hype will settle (which may happen sooner rather than later?).

When it does, the PMs who succeed will be the ones who figured out where AI adds value and where it’s just noise. Spoilers: PMs who can do discovery work, draft cohesive product strategies, and maximise shareholder value will always be in demand!