AI is not your product. The experience is.

"We use AI to…" is how most AI pitches begin. It's also the tell. Because the AI — the model, the capability, the thing the pitch is proud of — is the part your users never actually touch. What they touch is the experience wrapped around it: the flow, the feedback, the moments of trust or doubt. That wrapper is the product. The model is just an ingredient.

The model is the commodity now

Not long ago, the intelligence was the hard, defensible thing. Today the frontier capability is a few lines and an API key away, and it's the same few lines your competitor is using. When everyone can summon roughly the same capability, the capability stops being the advantage. What you do with it — how you shape it into something a person understands and relies on — becomes the whole game.

This is why so many technically impressive AI demos never become products people keep using. The demo shows the capability. The product has to survive the second week, when the novelty is gone and the only thing left is whether it's genuinely useful and genuinely trusted.

AI products fail on experience, not on the model

In our work, the AI products that struggle almost never struggle because the model was too weak. They struggle because the experience was unclear, generic, or untrusted:

• Unclear — the user can't tell what the AI will do, when, or why, so they never build a mental model of it.
• Generic — it looks and behaves like every other AI wrapper, so there's no reason to choose it or remember it.
• Untrusted — it's confidently wrong at some point, and after that every answer feels suspect, including the good ones.

Fixing any of these is a product and design problem, not a model problem. A better model doesn't make an opaque flow clear or a suspicious answer trustworthy. Better judgment does.

Trust is the real feature

The strange thing about AI is that more capability can make a product worse if the experience can't hold it. Power without legibility feels like a black box, and people don't hand real work to black boxes. So the actual job is making the intelligence usable: showing where an answer came from, letting people check and correct it, keeping a human in the loop where the stakes are high, and drawing clear boundaries so the product is predictable.

That's not a constraint on the AI — it's what turns an impressive capability into something someone will trust with their work, their money, or their customers. Trust isn't a nice-to-have you add at the end. It's the feature everything else depends on. If you want a structured way to pressure-test yours, our AI Trust Audit walks your product through the questions that matter.

Design for the second week, not the demo

The reframe that changes how you build: stop optimising the demo and start optimising the second week. The demo rewards the flashiest capability. The second week rewards clarity, reliability, and a reason to come back — none of which the model gives you for free.

So when we scope an AI product, the first questions aren't about the model at all. What is the core job? Where does the intelligence genuinely help, and where would it just add risk? How does the user stay in control? What makes this feel trustworthy on day thirty, not just day one? Get those right and even a modest model becomes a product people love. Get them wrong and the best model on earth won't save you.

Your users will never meet your model. They'll only ever meet the experience. Build that. See how we approach it, or start a conversation about your product.