$2.5 Trillion Is Being Spent on AI This Year. Almost None of It on Making It Usable.
$2.5 trillion is being spent on AI this year. Almost none of it is going into making any of it usable.
Hyperscalers are pouring half a trillion into infrastructure. Data centres. Chips. Cooling systems. Models are smarter every quarter. Demos are mind-blowing.
Then a real person opens the product.
The gap between the demo and the user
They don't know what to type. They don't trust the answer. They can't tell what the agent just did on their behalf. They close the tab and go back to doing it manually.
This happens millions of times a day, across every AI product on the market. The technology works. The product doesn't.
The gap isn't technical. It's experiential. And almost nobody is spending money to close it.
Why the model was never the hard part
The AI industry has a blind spot the size of the entire user experience. It assumes that if the model is good enough, the product will work. This has never been true for any technology in history, and it isn't true now.
The best search engine in 1998 was useless without a text box and a button. The most powerful smartphone processor means nothing if the interface is confusing. Raw capability has never been the bottleneck for adoption. The bottleneck is always the moment a real person tries to use it.
For AI products, that moment is brutal. The user faces a blank prompt with no guidance on what's possible. The output arrives with no indication of confidence or reliability. The agent takes action with no visibility into what it did or why.
Every one of these is a design problem. Not a model problem. It's the same gap that killed Sora: impressive technology with no design for daily use.
The trust gap
Trust is the invisible currency of every AI interaction. Users need to trust the output before they'll act on it. They need to trust the agent before they'll delegate to it. They need to trust the product before they'll come back tomorrow.
Right now, most AI products do almost nothing to earn that trust. The output appears fully formed with no provenance, no confidence signal, no way to verify. The user is expected to either accept it wholesale or do their own verification, which defeats the purpose of using AI in the first place.
Designing for trust means showing your work. It means giving users the right level of visibility into what happened and why. It means making it easy to correct mistakes and hard to miss important ones.
None of this requires a better model. It requires better design. The SaaSpocalypse is already exposing what happens when products skip this work.
The compute race ends in commodities
Here's the strategic reality. Models are converging. The gap between the best and second-best model shrinks every quarter. Infrastructure is being commoditised. The compute advantage that costs billions to build today will be table stakes in three years.
The compute race ends in commodities. The design race ends in trust.
The companies that win the AI era won't be the ones with the biggest data centres. They'll be the ones whose products people actually trust enough to use every day. Figma's falling stock price is an early signal of what happens when the market starts asking whether AI features create real value.
That's a design problem. And right now, almost nobody is treating it like one.
---
