A recent quote from Uber CEO Dara Khosrowshahi on Verge’s Decoder podcast struck me. Paraphrased:

If AI agents bring Uber new customers, Uber will work with them. If those agents take customers away from the Uber app, Uber will charge them heavily.

That one idea explains a lot about how Uber sees the future and why I think Uber’s biggest challenge isn’t competition, regulation, or even autonomous cars. It’s something deeper.

Uber Is an Incredible Marketplace... Today

First, credit where it’s due. Uber has built one of the most impressive local transportation marketplaces ever created. In cities without reliable public transit, Uber is often the default option. You open the app, tap a button, and within minutes a car shows up, sometimes a Camry, sometimes a Highlander, sometimes a Tesla like the one I drive.

That doesn’t happen by accident. It takes enormous coordination between technology, drivers, and logistics. Uber deserves recognition for making that work at scale. But Uber’s strength also reveals its weakness.

What Uber Actually Owns

Uber doesn’t own cars. Uber doesn’t manufacture vehicles. Uber doesn’t physically move people. What Uber owns is:

• the app

• the marketplace

• the customer habit

• the psychology of choice

Uber makes money by arbitraging information between passengers and drivers, which economists call asymmetric information. Uber sees both sides of the transaction and uses that visibility to influence behavior. And that influence shows up very clearly from the driver’s seat.

A Real Example: Same Ride, Same Pay, Different Story

On one shift, I accepted a ride paying $14.44 with a surge and priority pickup. The passenger canceled, and Uber sent me the exact same ride with different surge information, but the same total pay.

• same drop-off

• same distance

• same total pay

The only thing that changed was how the surge was presented. Uber increased the displayed surge amount, even though my total pay stayed the same. That’s not transportation optimization. That’s psychological framing.

It works because drivers are human. We’re tired. We’re driving. We make quick decisions. Uber knows that.

The Overlooked Cost: Declining Quality

As Uber has pushed toward profitability, driver pay has quietly eroded, often through subtle mechanisms rather than obvious cuts.

Drivers respond predictably:

• some quit

• Some downgrade vehicles

• Some stop caring as much.

• Some never start

Uber doesn’t seem overly concerned. The implicit assumption is that driver supply is infinite and interchangeable. That assumption holds until it doesn’t.

Why This Works on Humans (and Fails on Agents)

Uber’s system works because humans tolerate friction and manipulation for convenience. AI agents won’t. Agents don’t care about surge language, urgency cues, framing tricks, or brand loyalty.

• urgency cues

• framing tricks

• brand loyalty

They reduce everything to math:

• cost

• time

• reliability

• quality

An AI agent won’t just open Uber. It will instantly calculate who can get a passenger from A to B most efficiently, comparing Uber to all alternatives and making its choice without bias or loyalty, fundamentally changing how customers might be allocated.

The Assumption Hiding in Dara’s Quote

Dara’s quote assumes Uber always controls the door, that customers, or their AI assistants, must come through Uber’s marketplace. That’s true today. But as autonomous fleets from companies like Tesla, Waymo, and others expand, Uber’s leverage changes. Uber must partner. Manufacturers own the asset.

Marketplaces rely on persuasion.

Manufacturers rely on production.

AI agents prioritize the most efficient and reliable option, not the most persuasive marketplace. This shift in decision-making undermines strategies that rely on influencing people rather than delivering the best outcome based on measurable criteria.

Why This Matters to Investors, Drivers, and Passengers

For drivers, psychological nudges eventually stop working when pay no longer matches effort.

For passengers, declining quality becomes apparent when an agent compares outcomes rather than vibes.

For investors, current profitability doesn’t guarantee long-term durability if the company doesn’t control quality at the asset level.

Uber’s success today is real. But success in an app-driven world doesn’t automatically translate to success in an agent-driven one.

A Possible Path Forward

Uber’s real advantage is human diversity. Rather than treating every ride as identical, Uber could focus on customization: driver experience, safety, local knowledge, vehicle condition, and passenger preferences.

• safety profiles

• local knowledge

• vehicle condition

• passenger preferences

In a world of identical autonomous cars, uniqueness may actually matter.

Final Thought from the Driver’s Seat

Uber optimizes for psychology because psychology works on people. AI agents optimize for math, and math doesn’t care how a ride feels unless feeling improves outcomes. Uber isn’t broken. But assuming permanent control over the marketplace is a dangerous form of confidence. The future won’t ask who had the best app. It will ask who delivered the best ride: consistently, transparently, without persuasion.