In the early days of a new technology, it's hard to know where the enduring value lies.
Right now, many startups and incumbents are rushing to build with large language models. It feels like a gold rush, and understandably so — the capabilities are extraordinary, and improving fast. The instinct is to carve out a defensible position by training or fine-tuning a model on proprietary data. The assumption is that this vertical tuning creates a moat. But that assumption may not hold...
As foundation models get better — and they are getting better at an astonishing pace — the gap between a fine-tuned version and the best available base model may shrink. Tasks that used to require specialized training are now well within the reach of general-purpose models. What used to be "proprietary capability" may increasingly look like configuration.
This is a natural pattern in platform shifts. The things that were once scarce become abundant. The things that were once hard become easy. And when that happens, differentiation moves elsewhere. We don't really know what will happen with LLMs - but it is certainly possible that the models themselves become increasingly commoditized (just ask DeepSeek).
If it turns out that the model isn't the moat, what is?
Arguably, the old answers still apply. Customer experience matters. So does distribution and access to proprietary workflows, data and users. These aren't second-best strategies — they are the core ingredients of enduring companies. In a world where everyone has access to similar foundational capabilities, the question becomes: Who applies them better?
The companies that succeed won't necessarily be the ones with the most sophisticated prompt pipelines or the largest embeddings table. They'll be the ones that integrate LLMs in ways that are meaningful to customers, where the model is a core component of the value proposition — not the whole of it.
They'll be the ones that:
- Understand the customer's job to be done, and make that outcome faster or easier;
- Are deeply & intuitively embedded in the workflows that matter;
- Build interfaces and feedback loops that get better with use;
- Use AI not to dazzle, but to disappear — making the experience feel seamless and smart.
So what should businesses building natively with AI do? Focus on design.
AI isn't a silver bullet — it's a raw capability. The real leverage comes from how that capability is applied. That means being intentional about the experience: Where does AI show up? How does it interact with the user? What does it automate, what does it assist, and when does it stay out of the way?
This is where the opportunity gets exciting. We're not just building new features — we're exploring a new frontier in product design. Most current interfaces are still stuck in the metaphor of chat. But the future of AI-native experiences won't be limited to boxes and buttons. It will be rich, contextual, anticipatory. We'll need new patterns, new paradigms, and new instincts.
In that world, the winners won't just be good at machine learning. They'll be good at designing how people live and work alongside AI - and making that feel effortless.
That's not just a challenge. It's a generational opportunity.