The Human Premium

Part of my role involves helping roll out AI tools to product marketing teams, and over the past year I've noticed a pattern that I think explains why so many AI enablement efforts plateau. The conversation in most organizations has become entirely about consumption—token usage, cost per query, utilization dashboards with red and green zones—when it should be about capability. We're measuring how much AI people are using without asking whether the work is actually getting better.

What I keep seeing is that tool access is essentially universal at this point, with everyone having some combination of Copilot, ChatGPT, or Claude at their disposal, but actual capability follows a much more uneven distribution. A relatively small percentage of marketers are using AI to genuinely elevate their strategic work, while the majority are using it as a faster way to produce outputs that aren't meaningfully different from what they would have created before—just more of them, and quicker.

The marketers who break through this plateau share something in common, and it has almost nothing to do with their prompting technique or their comfort with the tools. What distinguishes them is that they have strong foundations in the underlying craft—they understand positioning frameworks deeply enough to know when the AI's suggestions are off-target, they can identify weak messaging without needing someone else to point it out, and they have a clear mental model for what good work looks like that allows them to evaluate AI outputs with real discernment. When these people collaborate with AI, the output sounds like them, carries their perspective, and reflects genuine strategic thinking rather than a kind of averaged synthesis of everything the model was trained on.

That observation has become a working thesis for me, and the uncomfortable implication is that AI amplifies whatever capability you bring to it. If you have strong foundations, AI becomes a genuine thought partner that helps you explore more alternatives, pressure-test your assumptions, and refine your thinking faster than you could alone. If your foundations are shaky, AI becomes a way to produce plausible-sounding work that you can't actually evaluate—and the scary part is that you may not even recognize the gap between what looks good and what actually is good.

Two Kinds of Assessment

This is what led us to build assessment into the center of the AI curriculum we've been developing, and to think carefully about what we're actually testing for. We use two distinct types of assessment, and each one reveals something different about where people are and what they need.

The first is a baseline assessment of core PMM competencies—not AI skills, but the underlying craft itself. Can you identify weak positioning when you see it? Do you understand the difference between describing a feature and articulating a value proposition? Can you construct a competitive narrative that goes beyond listing capabilities and actually tells a story about why your approach matters?

What we've found is that most participants come in with more confidence than competence in these areas. They've been doing the work for years, often successfully, but when you ask them to evaluate a positioning statement or explain why one messaging approach would perform better than another, the reasoning gets fuzzy. They know what they like, but they struggle to articulate why—which means they also struggle to guide an AI toward producing work that meets a standard they can't quite define. Our goal is to move people from intuition toward explicit frameworks they can apply consistently, because that's what allows you to have a real conversation with an AI about whether something is working or not.

The second assessment is what we call the cognitive assessment, and it focuses on how people actually think and collaborate with AI. This is where we explore the difference between delegation and co-authorship—between treating AI as a task-completion machine and treating it as a genuine thinking partner.

Most people start on the delegation side of that spectrum. They prompt AI to "write me a positioning statement" or "draft a competitive overview" and then accept or lightly edit whatever comes back. The co-author model is fundamentally different: you bring your own hypothesis to the conversation, you ask the AI to challenge your assumptions, you generate multiple alternatives and evaluate them against criteria you can articulate, and you treat the first draft as the beginning of a dialogue rather than the end of a task. The cognitive assessment helps people see where they fall on this spectrum, and in my experience that moment of recognition—realizing that there's a different way to work with these tools—is often what creates genuine interest in developing new capabilities. People get curious once they understand there's something more to learn.

Creating Safe Spaces to Learn

One thing we've learned is that the learning environment matters enormously. People are reluctant to experiment with AI when every prompt costs money, gets logged somewhere, and might be used in their deliverables before they've had a chance to develop real fluency. They play it safe, accept the first reasonable output, and never build the muscle memory for pushing back, iterating, and genuinely collaborating.

This is why we built what we call AI Labs—hands-on exercises where people can work through real marketing problems with AI as a collaborator, but in a designed environment where they're not burning through tokens, not producing anything that will be evaluated, and not afraid to make mistakes. The labs create a space where you can try prompting approaches that might not work, see what happens when you push back on the AI's first suggestion, and build intuition for the back-and-forth of real collaboration without any stakes attached.

This turns out to matter more than I expected. When people feel safe to experiment, they learn faster and develop more sophisticated mental models for how to work with AI effectively. They start to internalize the patterns—when to accept, when to push back, when to reframe the question entirely—in a way that transfers to their real work. By the time they're applying these skills to deliverables that matter, the fundamentals are already in muscle memory.

What's Actually Working

I want to be clear about something: this approach is working. The people who engage seriously with the curriculum come out different, and the difference shows up in their outputs. They produce work that has a point of view, that reflects genuine strategic thinking, and that sounds like it came from a human with something to say rather than from a sophisticated autocomplete. They also report feeling more confident—not because they've learned tricks for getting better outputs from AI, but because they've strengthened their own foundations and can now collaborate with AI from a position of expertise rather than dependence.

What strikes me most, though, is how much people want this kind of development. There's been remarkably little resistance to the idea that foundational skills matter, or that assessment is a valuable part of the learning process. If anything, people seem relieved to be offered something more substantial than another prompt-engineering cheat sheet or lunch-and-learn about the art of the possible. They want to get better at their craft. They want to be genuinely capable, not just more productive. They want to be valued for their judgment and strategic thinking, not just for their ability to generate more outputs with better tools.

That's the human premium, and I think it's what this moment actually calls for. The companies that invest in building real capability—not just distributing licenses and tracking consumption—will develop an advantage that compounds over time. Better people using AI become more effective, which frees them to learn more, which makes them better at using AI. But you have to kick-start that flywheel with genuine investment in human development, not just technology procurement.

Tokens are cheap and getting cheaper. The constraint isn't access to intelligence—it's the human capability to know what to ask, how to evaluate the response, and what to do with it. That's where the real leverage is, and right now almost nobody is investing in it.