AI Tools Won't Fix Your PM Team's Speed Problem

Tool access is no longer the differentiator

Every PM roundup publishes roughly the same 25 tools. The same names keep appearing: Notion AI, Dovetail, Perplexity, GitHub Copilot, a handful of others depending on the week. If you and your competitors all have access to the same stack, the tool itself cannot be what makes you faster. It never could be.

This matters because most teams are still treating AI adoption as a procurement problem. Get the tools, pay the subscriptions, encourage people to try things. That is table stakes now, not strategy.

The PMs I have seen pull meaningfully ahead are not using more tools. They are using fewer, with a sharper and more consistent practice built around each one. That difference is entirely in the operating layer, not the tool layer.

What the salary data is actually measuring

There is a well-cited figure going around about AI-skilled PMs commanding a 28% salary premium. My view is that most people are reading that number wrong. They assume it rewards knowing the tools. I do not think that is what is being paid for.

What actually commands a premium is prompting depth, judgment about when to use AI versus when not to, and the ability to integrate outputs into decisions without introducing noise. Those are not things you get by subscribing to a tool. They are things you build through deliberate, repeated practice with a consistent method.

Shreyas Doshi put it well in a post on X: "The biggest AI risk for PMs isn't that AI will replace them. It's that PMs who don't develop genuine AI judgment will be replaced by PMs who do." Judgment is the operative word. Tools do not give you that.

The context-switching tax nobody is talking about

Here is the uncomfortable version of what is happening at most product teams. There is now an AI touchpoint at every layer of the PM workflow: capturing research, synthesising interviews, scoring prioritisation, writing specs, reviewing pull requests. But these are point solutions with nothing connecting them. No shared standard for prompting. No agreed trigger for when AI enters a step versus when a human does it unassisted. No feedback loop that captures what worked and improves the next output.

The result is a new kind of cognitive overhead. Every PM on the team has their own version of the workflow, their own prompts saved somewhere, their own judgment about when to reach for which tool. That is not a productivity gain. That is context-switching tax at the team level, dressed up as adoption.

Operating discipline means three specific things: a shared prompting standard that the team actually uses, a defined trigger for when AI enters each workflow step, and a feedback loop that makes outputs measurably better over time. Without all three, AI delivers one-off wins. It does not compound.

The European constraint that makes this non-optional

For teams based in Europe, there is an additional layer that makes discipline around AI tools a governance requirement, not just a performance question. Data sovereignty matters. Most of the default AI stack is hosted by US providers operating under US law. Which means every time a PM pastes customer interview data or roadmap context into a prompt, there is a data handling question sitting underneath it.

I have seen teams in regulated industries discover this late, after months of ad-hoc AI use across the workflow. The answer is not to stop using AI. The answer is a clear internal policy about which tools touch which categories of data, and where the boundaries are. That policy is part of operating discipline. It is not separate from it.

Start with one workflow step, not the whole stack

If your team is feeling the gap between AI access and actual speed, the fix is not another tool evaluation. Pick one repeatable workflow step where quality and consistency matter most. Customer interview synthesis is a good candidate. Write a shared prompting standard for that one step. Run it for four weeks. Capture what the outputs look like, what gets edited, what gets used as-is. Then improve the standard.

That is a compounding loop. Every other AI benefit your team eventually gets will be built on the same principle.