Bloomberg Tech

Tiffany Luck on why AI's bottleneck is adoption, not intelligence

Tiffany Luck· Partner at NEA
·~6 min·English·Bloomberg
AI CompanyBusiness StrategyAgentsLLM
TL;DR

NEA partner Tiffany Luck argues the real constraint on AI has moved from intelligence to adoption, and that the gap between what frontier models can do and what the world actually deploys is where the value, and the investment, now sit.

01Core Mental Model

The Adoption Gap

The frontier keeps getting smarter, but the binding constraint is how slowly the world actually adopts it, the same lag that kept Waymo off the road for years after the cars were ready.

Probably the cars were ready to be on the road far ahead of when they actually got on the road.

— Tiffany Luck, Bloomberg Tech
Key Insight
Luck reframes slow adoption as evidence of upside, not failure: the technology curve and the adoption curve are separate lines, and the distance between them is a market that is still unbuilt.
02The Diagnosis

Intelligence Is Not Enough

Raw intelligence is not enough on its own — it has to be converted into real outcomes — which is why OpenAI and Anthropic are now standing up dedicated deployment arms.

We need to keep, you know, kind of keep going up that intelligence curve. But we really need to figure out how to get up the deployment curve of the adoption curve.

— Tiffany Luck, Bloomberg Tech
Key Insight
When the two leading labs stand up deployment businesses, they are conceding that selling intelligence is not the same as selling results, the hard problem has moved downstream of the model itself.
03The Bridge

Forward-Deployed Engineers

Many companies still need help turning raw AI into working deployments, so the stopgap is embedding engineers inside the customer, and Anthropic is funding that bridge directly with a $100 million academy to train 10,000 engineers.

they're investing $100 million to train 10,000 engineers, kind of between now and the end of next year

— Tiffany Luck, Bloomberg Tech
Key Insight
The need for forward-deployed engineers is a tell about where the difficulty now sits: closing the gap takes people embedded in the customer to work out how the models should actually be used, which makes it an operational problem as much as a technical one.
04The New Metric

Rubber Meets the Road

The buyer's question has shifted from whether AI is powerful to what return it actually produces for every dollar spent.

what is that ROI based on? You know, what is the value I'm driving based on what I'm spending?

— Tiffany Luck, Bloomberg Tech
Key Insight
The shift from “is it powerful?” to “what’s the ROI?” raises the bar: capability becomes table stakes, and the harder question is whether the value produced actually exceeds what was spent to get it.
05The Enterprise Shift

Own Your Own Intelligence

Enterprises increasingly want to own part of their intelligence stack rather than rent all of it, because they see keeping data and workflows in-house as the way to keep compounding their moat.

enterprises want to keep their moat and they want to keep compounding their moat. That means data private, that means compounding workflows

— Tiffany Luck, Bloomberg Tech
Key Insight
The moat here is not the model, it is everything that accumulates around it, so every private dataset and every refined workflow that stays inside the company is an advantage a rented frontier model can never hand back to you.
06The Routing Decision

Not Every Query Needs the Frontier

Because not every request needs a frontier model, routing the everyday work — roughly 80% of tasks — to a cheaper in-house model can beat the base models on both performance and cost.

the 80% that gets done or the everyday things that get done could actually be tackled by an enterprise's own model.

— Tiffany Luck, Bloomberg Tech
Key Insight
This is the economic logic underneath owning your intelligence: if the routine work can run on a cheaper in-house model, you stop paying premium frontier prices for tasks that do not need them, and reserve the frontier for the hardest cases.
07What Comes Next

Physical AI Is the Next Wave

Luck's next frontier is physical AI: the breakthrough came through text, and she is now excited about world models and physical intelligence — AI that acts in the real world, not just on a screen.

the other area we're super excited about is this physical AI

— Tiffany Luck, Bloomberg Tech
Key Insight
Her ladder from text to world models to physical AI is a claim about modalities, not just raw capability: each step pushes intelligence further off the screen and into systems that must act in the physical world, where the bar for reliability is far higher.