Uber is pouring more money into AI, but its top executives are starting to sound less like evangelists and more like accountants with a headache. The company says the technology is still not delivering a clean enough productivity boost to justify some of the staffing trade-offs attached to it, even as Uber pushes employees to use AI more aggressively across engineering and other workflows.
That tension is becoming familiar across Big Tech. Companies are spending heavily on models, tokens, and internal tooling while asking whether the promised speed-up is showing up anywhere customers can actually see. Uber’s answer so far: not really, at least not in a way that makes the math easy.
Uber’s AI bill is getting harder to defend
Last month, Uber CTO Praveen Neppalli Naga said the company had already blown through its 2026 AI budget in the first four months of the year. Soon after, executives said Uber would increase AI spending even more and slow hiring to help pay for it. That’s a pretty familiar Silicon Valley move: spend first, ask questions later, then call it efficiency.
But COO Andrew Macdonald drew a line between internal AI usage and actual output. In a podcast interview over the weekend, he said Uber still cannot tie AI-assisted code work to a clear rise in shipped features. He pointed to figures such as 25% of code commits coming via Claude Code last quarter, but said the company has not seen a direct link to more useful product work reaching users.
- Uber says AI is being used more inside engineering teams.
- Executives say the productivity case is still hard to prove.
- The company is still increasing AI spending while slowing hiring.
Claude Code is popular, but gains are still unclear
Uber’s comments cut against the industry’s favorite fantasy: that AI can replace enough human labor to pay for itself almost immediately. That idea has helped justify enormous infrastructure spending across the sector, from cloud capacity to chips to model training, and it has also fed a quieter corporate strategy of pushing employees to use AI more so the bill can be spread around.
Macdonald’s view is simpler and less glamorous. If AI saves time but does not clearly translate into better products, faster shipping, or more revenue, then ”somebody’s paying the bill” starts to matter. He said the company’s leaders are not yet able to draw a direct line from AI usage to the kind of output that would make headcount cuts feel justified.
Agentic commerce has not taken off yet
Uber is also watching the much-hyped world of commerce agents with a skeptical eye. Some analysts have argued that chatbot-style assistants could eventually route shopping and delivery demand away from apps like Uber and DoorDash, turning large-model providers into the new gatekeepers of commerce. That threat has been floating around for long enough to sound inevitable, which is usually the signal to check the calendar.
Macdonald said Uber has been working with major model companies on commerce rollouts, but nothing has taken off yet. A year ago, he said, Uber was worried that all commerce would soon flow through large model codes and chatbots; that has not happened. For now, the app economy still has a pulse, and the AI agents are still mostly living in slide decks and demos.
The bigger question is whether Uber is an outlier or an early warning. If even a company this exposed to automation cannot find a clean productivity payoff, the broader AI spending boom starts to look less like a miracle and more like a very expensive experiment. The next test is simple: can the industry show real output gains before the bill keeps climbing?