The Diffusion Gap Is Widening

A shout out to Pat Grady and the Sequoia team for sharing what was, in my view, the single most important slide from AI Ascent 2026. Watch it here.

If you only take one frame away from this year's keynote, make it this one: a green curve of AI capabilities rocketing nearly straight up on the left, a white curve of AI adoption crawling along the floor and only beginning its ascent on the right, and a giant arrow stretching across the void between them labeled "The Application Layer Opportunity."

That's the diffusion gap.

Why this is important ?

Capabilities are compounding on a timeline measured in weeks. Adoption (the messy, human, organizational work of actually getting those capabilities into the hands of a large enterprise employee on a regular weekday) is measured in years. Pat's point, made bluntly, is that the rate at which capability is being created is wildly outpacing the rate at which it's diffusing into the real economy. And every time the frontier moves, that gap gets wider, not narrower.

Most people read that and think it's a problem. It's not. It's the opportunity. The bigger the gap, the bigger the prize for anyone willing to stand inside it and build a bridge.

Why incumbents can't close it

Large enterprises are not built to absorb a step-change in capability every quarter. Their procurement cycles are longer than the half-life of the model they're evaluating. By the time a working group has formed, an RFP has gone out, and a pilot has been scoped, the underlying model has moved on, the primitives have changed, and the original use case probably needs to be rethought. The org chart is fundamentally mismatched with the rate of change at the foundation layer.

This is structural. You can't fix it with another committee.

Why this is a startup's moment

Startups close the diffusion gap. That’s the job. Not to “build an AI product.” Not to “wrap a model.” The job is to take a capability that exists in a lab or an API and turn it into a real outcome a specific customer can use today, seamlessly, without having to think about the underlying technology. The greater the distance between raw capability and practical adoption, the more interesting the company. That gap is where value is created, and it’s only getting wider. That requires a few things:

1. Speed. Not speed as a vibe, speed as a structural commitment. If you're not shipping faster than your customer can re-org around your last release, you're not closing the gap, you're just sitting in it.
2. Execution under uncertainty. The model under your product will get better. It will also occasionally collapse a feature you spent three months on into a single API call. You cannot let fear of disruption from below paralyze you. The companies that win in this gap are the ones who treat the foundation layer's progress as fuel, not threat.
3. Embedded customer obsession. You don't close the diffusion gap from the outside with a demo. You close it from inside the customer's workflow, their data, their constraints, their politics, their teams. The product is the "path of least resistance" to an outcome they couldn't reach before. That path is custom and hard, but also the actual moat.
4. A willingness to be disrupted weekly. The most counterintuitive part of building in the gap is that you have to be more willing to disrupt yourself than the foundation models are, with product sprints of a couple of days not weeks. If a new capability lands that makes half your product obsolete, the right move is to rebuild around it on Monday, not protect what you had on Sunday.

The takeaway

Capabilities are not the bottleneck. Diffusion is. And diffusion is a human problem: a workflow problem, a trust problem, an integration problem, a "make it simple for one specific customer with one specific job" problem.

That’s not to diminish the labs, the frontier should keep moving fast. But frontier breakthroughs don’t create value on their own. Someone still has to carry them across the gap into the real world.