For twenty years, product teams believed something that felt obviously true. Build a better feature, ship it faster than the competition, out-execute on the roadmap, and you win.

That belief is now wrong. Not directionally wrong. Fundamentally wrong.

I want to walk through why this happened, what it means for how we think about differentiation, and what the best product teams are doing instead. This is one of the more important shifts I have seen in my career, and most companies have not caught up to it yet.

The Old Moat

The old moat was built on a simple asymmetry. Building software was hard. It took skilled engineers, months of work, and real investment to ship a meaningful feature. If you had a strong team and a head start, competitors could not easily catch up. By the time they shipped their version, you had moved on to the next thing.

This asymmetry is what funded an entire generation of product strategy. Move fast. Ship first. Build a lead and defend it.

I have spent my career telling teams that speed and execution matter. Move fast. Ship first. Build a lead and defend it. I still believe that. But I no longer believe that shipping a feature, by itself, buys you a durable advantage. The economics changed under our feet.

What Changed

AI collapsed the cost of building software.

A feature that used to take a strong engineering team six weeks to design, build, and ship can now be prototyped in an afternoon. Production-ready in days. This is not a modest efficiency gain. It is an order-of-magnitude shift in production cost.

Here is the part that matters most for product leaders. The thing that used to protect you, the time and effort required to replicate your work, has largely disappeared. If a competitor sees your new feature, they don’t need to reverse-engineer your architecture or staff a project for a quarter. They point an AI coding assistant at the problem, describe what they saw, and have a working version days later.

I talk to founders who describe watching their most innovative feature get matched by three competitors within a month of shipping. Not a rough approximation. A close match, sometimes a better one, because the competitor got to see your design decisions play out in the market before they built their own.

Speed used to be a weapon you could wield against the field. Now speed is table stakes for everyone, including the competitors chasing you.

The Second, Quieter Problem: Reversion to the Average

There is a second dynamic at play, and it is more subtle than the first. It is also more dangerous, because most product leaders have not fully reckoned with it.

Large language models are probability engines. They generate the next statistically likely token given the provided input. This is a remarkable technology. It is also, by construction, a technology that pulls toward the average.

Think about what this means in practice. If your product team and your competitor's product team are both using equivalent AI tools, working from comparable prompts, and drawing on analogous training data and best practices, you should expect similar outputs. Not identical, but structurally similar. Same patterns. Same conventions. Same feature shapes.

I have started calling this the convergence problem. When every team has access to the same probability engine and feeds it similar inputs, the natural AI output is undifferentiated software features. AI does not have a point of view. It reflects the aggregate of what came before. If you ask it to design a dashboard, you get a dashboard that looks like every other AI-assisted dashboard. If you ask it to design an onboarding flow, you get the median onboarding flow, competently executed.

This is the opposite of what well-executed product work looked like. Good product work has always required a point of view that the average would not produce. A specific bet about what customers need that most people would not make. Which is precisely what a probability engine, left to its own devices, cannot generate.

So we have two forces working together. First, building has gotten dramatically cheaper, which erases the time advantage. Secondly, the tools used by everyone pull outputs toward the mean, which erases the differentiation advantage. Put those together, and you get a market where feature parity arrives faster than ever, and the features themselves are increasingly interchangeable.

Two Product Teams

Let me make this concrete by contrasting two representative teams that I have seen versions of many times in the past year.

The first team treats AI as a feature factory. They see a competitor ship something, they task their AI tools with building an equivalent, and they ship it within the week. They measure success by velocity. Features shipped per quarter. Time to parity when a competitor moves first. Their roadmap is a running list of what everyone else in the category already has.

This team is busy. Their release notes are long. Their product looks competitive on the surface. But ask their customers why they chose this product over the alternative, and you will get a shrug. The features are fine. The features are always fine. Nobody has a strong reason to stay.

The second team uses the same AI tools, but for a different purpose. They use AI to compress the time from having an idea to achieving a testable version of it. That is a genuine gift, and I don’t want to undersell it. But they spend the time they saved on something the first team skipped: figuring out what is actually true about their customers that their competitors have not yet figured out.

This team ships fewer raw features by count. But what they ship tends to come from a specific insight, one built from direct customer contact, from data nobody else has, from a judgment call that a probability engine would never have generated on its own, because that judgment call is not average. It is a deliberate departure from average.

Six months later, the first team is still shipping, still matching, still busy, and still struggling to explain why anyone should choose them. The second team has built something hard to copy, not because it was difficult to build, but because it required knowing something the competitors did not know.

This is the distinction that matters now. It is not build speed versus build speed. Everyone has build speed. It is insight versus imitation.

So Where Does the Moat Come From Now?

If features alone will not hold, where does durable advantage come from? I see four places, and the best companies are usually working on more than one at a time.

Proprietary insight. This is the foundation. Insight that comes from a depth of customer understanding that cannot be prompted into existence. This usually comes from founders and product leaders who have lived the problem, who talk to customers constantly, and who notice things that a generic AI tool trained on public data has no way of knowing. If your product decisions can be replicated by describing the problem to an AI, they were never truly differentiated to begin with.

Proprietary data. A model is only as sharp as what it has learned from. If your product accumulates a unique dataset over time, such as usage patterns, outcomes, edge cases specific to your customer base, you have something a competitor cannot simply prompt their way into. This is why the products that will hold advantage over the next five years are the ones designed from the start to compound a data advantage, not just to ship features.

System and workflow depth. Any single feature can be copied quickly. A deeply integrated system in which dozens of decisions reinforce each other across a customer workflow is much harder to replicate wholesale. Competitors can match individual pieces. Matching the coherence of the whole system is a different order of problem, and it is exactly the kind of problem that requires sustained product judgment rather than a well-crafted prompt.

Business model and pricing. This is the one most product teams underweight, and it deserves more attention than it has historically gotten. When products are close to parity on features, how you package and price becomes a primary lever of differentiation, not an afterthought handled by finance. Usage-based pricing that aligns cost with the value a customer actually receives. Packaging that removes friction competitors have not bothered to remove. A model that makes switching away expensive not through lock-in tactics but through genuine accumulated value. These are product decisions as much as they are commercial decisions, and they are much harder for a competitor to copy than a UI pattern or a workflow screen.

What This Means for How We Work

I am not arguing that speed no longer matters. Speed still matters enormously. What has changed is what speed buys you. It used to buy you a lead. Now it mostly buys you the right to compete. It is necessary and no longer sufficient.

The teams that will win in this environment are the ones that treat AI-driven build speed as a tool for faster learning, not faster shipping. Use the compressed build time to run more experiments, validate more assumptions, and get closer to customers faster. Then spend the differentiation you earn from that learning on the things that are genuinely hard to copy: insight, data, systemic coherence, and business model.

The teams that will struggle are the ones that treat AI as a way to ship the same roadmap faster. They will find that their competitors can ship the same roadmap just as fast, because the whole category is drawing from the same probability engine, prompted with the same average questions, producing the same average answers.

The moat was never really the feature. The feature was always downstream of something harder to build: a specific, well-earned understanding of the customer that most people lacked. That was true before AI. It is more true now. AI made the building part easy and accessible to everyone. What is left to compete on is the hard part. That has not gotten any easier. If anything, it has become the entire game.