Uber just dismissed the two leaders who spent the past two years building its AI data labeling business, and the company won't say why.
Naga Kasu, senior director of engineering, and Pankaj Kamat, director of product at Uber AI Solutions, were let go this week, Bloomberg reported. Both had spent more than a decade at Uber before moving into the AI unit, running operations across the core ride-hail and delivery businesses. An Uber spokesperson confirmed the exits to Bloomberg and described them as part of a "broader leadership transition," adding that the division "is seeing strong momentum." Uber did not name replacements or explain what prompted the shake-up.
You don't dismiss the two people who built something from scratch and call it momentum without inviting the obvious question. What actually went wrong?
Uber AI Solutions launched in November 2024 under the name Scaled Solutions, according to TechCrunch's original report on the unit. The pitch was straightforward: take the same gig-labor playbook Uber built for driving and delivery, and apply it to the unglamorous work AI companies need done by humans. Workers validate code outputs, evaluate audio clips in different languages, and label video footage for driverless vehicles. Uber has counted Alphabet, Aurora Innovation and Niantic among its customers, and the service is now live in more than 30 countries, up from five at launch, according to reporting from Forbes.
The timing of the original push was not an accident. Uber accelerated its data-labeling ambitions after Meta's deal with Scale AI last June, which sent Scale founder Alex Wang to run Meta's Superintelligence Lab and threw the rest of the labeling industry into scramble mode. Rivals including Mercor, Turing and Invisible Technologies moved to pick up business Scale left behind. Surge AI, founded by Edwin Chen, has reportedly outpaced Scale in revenue despite raising far less venture money. Uber's argument for entering a market already crowded with well-funded startups was size: a company worth roughly $175 billion, with $43.9 billion in revenue last year, doesn't need a Series C to compete on price or reliability. The data-labeling market itself is projected to top $17 billion by 2030.
The bet only works if Uber's core insight holds: that the infrastructure built to route drivers and dispatch delivery couriers can be repointed at routing human judgment for AI training. Uber has already piloted using idle drivers as data labelers during downtime, treating annotation work as another task type inside the same app that handles rides and food orders. It's a clean idea on paper. Whether it survives contact with enterprise AI buyers, who care about consistency and quality control in ways a five-star rating system doesn't capture, is a different question.
Losing the two executives who ran engineering and product for the unit this early doesn't kill that bet, but it doesn't help it either. Kasu and Kamat weren't outside hires brought in to add AI credibility. They were internal operators who understood how Uber's dispatch and marketplace logic worked before they applied it to labeling, which is exactly the kind of institutional knowledge that's hardest to replace on short notice.
Uber has not disclosed revenue figures for the AI Solutions division, so there's no public number to say whether
Also read: Together AI Raises $800 Million From Aramco at an $8.3 Billion Valuation • Venice AI became a unicorn by promising to forget everything you tell it • Cisco is betting its whole workforce on AI agents while it cuts jobs
