The moat is not the model.

I've been living inside these tools for months now, the way anyone running marketing for a software studio does. And the pattern I keep coming back to is uncomfortable for the labs but obvious to anyone who has shipped a consumer product: the model is not where the value gets captured.

Here's the observation that started this for me. The agentic consumer products coming directly out of the frontier labs keep feeling heavier, not lighter, over time. More rough edges. Fewer of the integrations I actually need to get work done. A lot of raw capability, wrapped in an experience that feels like it was built by people who are brilliant at training models and bored by the boring work of product.

That's not an accident. It's structural.

Two games, two muscles

Training a frontier model and building a consumer interface require almost opposite organizational muscles.

One is a research org: long horizons, enormous compute budgets, a culture that rewards being right about hard problems eighteen months from now. The other is a product org: short loops, obsessive attention to the first ten seconds of a new user's experience, a culture that rewards shaving friction off a workflow today.

Asking one company to be world-class at both is like asking a Formula 1 engine manufacturer to also run the best chain of neighborhood car dealerships. Both are real businesses. Both are hard. They are not the same business, and the things that make you great at one quietly make you worse at the other.

The lab optimizes for the benchmark. The product company optimizes for the Tuesday afternoon when a real person is trying to get one real thing done.

When the same org tries to do both, the product almost always loses the internal fight. The model team sets the roadmap. The interface becomes a thin demo surface for the latest capability instead of a thing designed around what users actually do all day.

What actually wins the end user

The frontier model is becoming a commodity input. That's not a knock — commodities are essential. Electricity is a commodity. But you don't pay your loyalty to the power plant. You pay it to the products plugged into the wall that make your life work.

The companies I'd bet on to win the end consumer are the ones treating the model that way: as an interchangeable input behind a great product. The wrapper companies. The ones who are model-agnostic by design — Perplexity, Manus, and a growing set of others who route to whatever model is best for the job and spend their actual energy on the experience.

Being model-agnostic turns out to be a feature, not a compromise:

• You always run the best model for the task — instead of being locked to one lab's weakest week.
• You're insulated from any single lab's roadmap, pricing, or outages. The model layer can churn beneath you and the product stays stable.
• Your effort compounds where the user actually lives — connectors, memory, speed, trust, the thousand small decisions that make a tool feel like it was built for you.

None of that is glamorous. None of it shows up on a benchmark leaderboard. All of it is what makes someone open your product instead of the other tab tomorrow morning.

Distribution and design are the durable advantages

If the model is a commodity, the moat has to live somewhere else. I think it lives in two places that the labs are structurally bad at defending.

Distribution. Being the default. Being already open. Being the thing your team already trusts and already has wired into their workflow. Models get leapfrogged every few months. Habits and integrations don't.

Deeply user-oriented product design. The unsexy obsession with the actual job a person is trying to do — and removing every gram of friction between them and the outcome. This is a craft, and it's a different craft than research.

I'd put it bluntly: a slightly-less-capable model wrapped in a product someone genuinely loves to use beats a frontier model wrapped in something clunky. Every time. Consumers don't buy capability. They buy the experience of the capability.

Why this matters for how we build

I'm writing this from a marketing seat at a studio that builds software with AI-native engineers, so take the bias as stated. But it's exactly why we build the way we do. We don't bet the product on any single lab. We treat the models as inputs, stay agnostic about which one we reach for, and pour the effort into the part that's actually durable: the product and the people using it.

I think Anthropic is going to be one of the leading model companies for a long time. That's a great business. It's just a different business than winning the end consumer — and I don't think the same company gets to own both.

The model is the engine. It matters enormously, and it's getting cheaper and more interchangeable every quarter. The thing that lasts is what you build around it, and who you build it for.

The moat is not the model. It never was.

These are my own views, not a knock on the engineers building any of these tools — model or wrapper. I just think the structure of the industry is clearer than the hype lets on.