Domain Experience vs. Functional Expertise? Neither.

Hey there,

Happy New Year. I'm starting 2026 by revisiting an argument I've been making for twenty years—and wondering if AI just broke it.

My entire career has been built on my strong belief that functional product management expertise matters more than domain expertise. A great PM can product-manage anything. They don't need to be the target customer or have spent a decade in the industry. Coming up to speed on domain knowledge takes months. Building genuine PM capability takes years—if someone has the raw material to develop it at all.

I still believe this. But a quote I encountered recently from Build What Matters by Ben Foster and Rajesh Nerlikar reminded me of this age-old debate—and made me want to reframe it for our new AI-dominated world:

"Hiring someone with too much domain expertise carries its own risk. Remember, as a technology company, you want to hire product managers who are going to innovate. If they are steeped in domain expertise, they might be unmotivated or unable to think about revolutionary approaches that are needed to deliver a 10x outcome for your customers."

The authors are making the same argument I've made. But, while thinking on this, I realized that we're not really debating domain expertise versus functional expertise. We're debating proxies—and we've been doing it for decades without acknowledging that's what's happening.

The Real Problem We've Never Solved

Judgment is what actually makes product managers effective. The ability to make good decisions with incomplete information. Knowing when to push back on stakeholders and when to align. Recognizing which user feedback represents signal versus noise. Understanding when a framework applies and when it misleads.

But judgment is maddeningly difficult to evaluate. So we use proxies.

Functional expertise is a proxy: "They've shipped products before, so they probably have good judgment about shipping products."

Domain expertise is a proxy: "They know this industry deeply, so they'll probably make contextually-appropriate decisions."

Behavioral interviews are a proxy: "They can narrate a coherent story about exercising judgment, so they probably have it."

We've been arguing about which proxy is better when the real issue is that all of them are imperfect attempts to see something we can't measure directly.

"We've been arguing about which proxy is better when the real issue is that all of them are imperfect attempts to see something we can't measure directly."

What Judgment Actually Looks Like

Consider three scenarios where judgment—not expertise—determines outcomes:

A PM inherits a product in an industry where every competitor offers the same "table stakes" feature. Domain expertise says "we need this feature because everyone has it." Judgment asks: "Why does everyone offer this, and is the underlying customer need actually being served—or are we all copying each other's homework?"

A PM uses AI to generate a competitive analysis. The output is fluent, comprehensive, well-structured. A PM without judgment ships it to leadership. A PM with judgment notices the analysis completely missed the regulatory constraint that makes half the recommendations irrelevant—because they knew to look for what AI didn't know to include.

A PM evaluates a backlog using a prioritization framework. RICE, ICE, whatever the flavor of the month is. A PM without judgment applies the framework and reports the results. A PM with judgment applies the framework and recognizes it's optimizing for the wrong variable given where the company is strategically—then has the difficult conversation about it.

None of these scenarios are about domain expertise or functional expertise. They're about the meta-awareness to know what kind of knowledge a situation requires and to recognize when you're missing it.

AI Compresses the Gap

Enter AI...and things start to get uncomfortable.

Knowledge work has always been characterized by large differences in ability among workers. The gap between a mediocre PM and a strong PM is enormous—and historically visible in their artifacts. The PRDs. The strategic analyses. The stakeholder communications.

But research from Ethan Mollick and others consistently shows that AI provides the biggest boost to workers with the lowest initial ability. It turns poor performers into adequate performers. The floor rises.

This means the artifacts that used to signal PM capability—the outputs we evaluated in portfolios and work samples—are becoming less reliable indicators. A mediocre PM with strong AI skills can now produce documents that look like a strong PM's work. The proxy is getting noisier.

Simultaneously, surface-level domain knowledge is easier to acquire or simulate. Ask AI for an overview of fintech regulatory constraints or healthcare reimbursement models and you'll get something plausible in minutes. The "I can learn any industry" argument that generalist PMs have always made is now trivially true—which means it's no longer differentiating.

And behavioral interviews? Candidates can now rehearse and refine their STAR-format stories until they're perfectly structured, flawlessly delivered, and indistinguishable from genuine reflection on actual experience. The proxy degrades further.

The Retreat to Worse Proxies

Over the course of 2025, I noticed a trend—instead of solving the judgment problem directly, the industry is retreating to more specific proxies.

"Must have 5+ years in B2B fintech with PLG motion."

"Looking for someone who's built AI features at a Series C startup in the developer tools space."

"Need experience specifically with enterprise healthcare platforms serving payer organizations."

These requirements feel more verifiable than "has good judgment." But they're actually weaker proxies—they select for luck of career path rather than actual capability. They assume context doesn't transfer across industries, which contradicts everything we know about how expertise actually develops. And they create a hiring paradox: the more specific your requirements, the smaller your candidate pool, and the more likely you are to hire someone who matches the pattern but lacks the judgment you actually need.

The unicorn hunt isn't a sign that companies have gotten better at identifying what they need. It's a sign they've given up on evaluating judgment and are hoping that hyper-specificity will substitute for discernment.

"The unicorn hunt isn't a sign that companies have gotten better at identifying what they need. It's a sign they've given up on evaluating judgment."

The Question AI Forces Us to Ask

If all our proxies are degrading, we're left with a question we've been avoiding: What is PM judgment, and how would we recognize it if we saw it?

Here's one answer: judgment is meta-awareness—knowing what kind of knowledge a situation requires and recognizing when you're missing it. It's understanding when your instincts are well-calibrated versus when you're in unfamiliar territory. When a framework applies and when it misleads. When AI output is reliable and when it's confidently wrong.

That sounds abstract—maybe even unteachable. But here's what gives me hope: psychologists have studied a related capacity called Theory of Mind—the ability to recognize that others have mental states, beliefs, and goals different from yours. And evidence suggests it's not fixed. It can be developed through frameworks, deliberate practice, and structured feedback. If the underlying cognitive machinery for understanding other minds is trainable, there's good reason to believe the meta-awareness we're calling judgment is trainable too.

This shifts the question entirely. Instead of "who already has judgment" (which we can't measure) or "who has the right experience" (which is a degrading proxy), we could ask: "Who has the capacity and willingness to develop judgment—and what would we look for to evaluate that?"

I don't have a clean answer. But I'm increasingly convinced that's the right question. The proxy game is ending. AI is compressing the gap on everything except the judgment that determines whether AI outputs are actually useful in context. Product leaders who figure out how to evaluate that directly—not through proxies—will build stronger teams than those still hunting for unicorns.

The rest will keep arguing about domain expertise versus functional expertise, missing the point entirely.

Break a Pencil,

Michael

P.S. If you're a product leader wrestling with how to build AI-capable teams—not just teams that use AI tools, but teams with the judgment to use them well—that's exactly what I work on with organizations. Reach out if you'd like to explore what systematic capability building looks like.

P.P.S. Forward this to someone in your network who's either hiring PMs or job hunting right now. The proxy problem affects both sides of the table, and the leaders who recognize it first will have a significant advantage.

Sources: Ethan Mollick's "Co-Intelligence" and Riedl & Weidmann's "Quantifying Human-AI Synergy" informed several ideas in this piece.