Uber Had Only 4 Rides Booked by AI Agents: Disintermediation Hits a Wall Right at the Place Order Button

Uber shared a surprisingly small figure during its latest earnings call: in Q3 2026, third-party AI agents arranged a grand total of just 4 rides. In the week of Halloween, around 500 Uber Eats orders were completed by AI agents. This marks the first time Uber has ever disclosed this type of data. For context, the total combined volume of Uber Rides trips and Uber Eats deliveries hit 320 million over the same 3-month window.

Image: Cath Virginia / The Verge, Getty Images

This set of numbers stands out starkly in ongoing industry conversations about disintermediation. The so-called disintermediation scenario describes a future where users no longer open the Uber app themselves, and instead task AI agents such as Meta's Muse and OpenAI's Dots to compare prices, hail rides, and place food orders on their behalf. Platforms would then be pushed entirely to the background, responsible for nothing but order fulfillment. User relationships would be fully claimed by the AI agents, reducing platforms to little more than neutral delivery pipelines. This narrative has gained massive traction over the past year, with many experts warning that platforms like Uber are at risk of losing their status as the primary user entry point.

Uber's new disclosure pours a bucket of cold water on that hype. The 4 total rides figure is functionally indistinguishable from statistical noise. The 500 food delivery orders, while larger than 4, is still negligible when measured against the 320 million total order baseline. For an AI agent to pull off a single completed ride booking in a real-world setting, it has to successfully navigate a long list of complex variables: location confirmation, payment processing, vehicle type selection, wait time validation, last-minute order adjustments, refunds, customer service queries, and clear liability attribution. There is a massive gap between an AI model that can hold a casual conversation, and one that can reliably walk through every required step to finally hit the confirm order button — a gap guarded by layered platform risk control systems and commercial terms of service.

My assessment: In the short term, the threat that AI agents pose to Uber has been wildly overestimated. The real determining variable is whether platforms will open official dedicated interfaces and clear revenue-sharing rules for AI agents. If platforms choose to grant formal access permissions to agents, AI-placed order volumes could actually rise to meaningful levels; if platforms withhold that access, agents will be forced to rely on simulated click automation, which will put hard caps on their scale, operational stability, and leave them exposed to steep compliance costs. Developers building AI agent products should not only focus on raw model capabilities, but also calculate the real costs of securing platform partnerships. Average users do not need to expect AI to hail their Uber reliably anytime soon — those 4 total rides make it clear the technology is still barely standing at the front door.

Source article: The Verge