Uber’s head of operations has reignited the debate on AI spending, stating that the ride-hailing giant is finding it increasingly difficult to justify rising AI costs as gains in productivity and efficiency remain elusive.
Andrew Macdonald, the president and COO of Uber, said that he held talks with the company’s senior engineering leaders and realised that higher token usage of AI coding tools such as Anthropic’s Claude Code did not translate into a proportional increase in useful consumer features. A token is a unit of text processed by the AI model.
“That link is not there yet, right? I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features,’” Macdonald said in an appearance on the Rapid Response podcast released on Saturday, May 23.
“I think over the coming quarters and years, maybe that will become clearer, but I think today it’s hard, even if some of the underlying metrics are trending in a really astronomical direction,” the top Uber executive added.
Macdonald further said that AI tools can seem free if you are “just a user sitting there coming up with interesting use cases” without paying for it. But ultimately, the company foots the bill.
“We’re going to have to start talking about token consumption and the associated cost versus headcount. So if you’re not actually able to draw a direct line to how much useful features and functionality you’re shipping to your users, that trade becomes harder to justify,” he said.
Macdonald’s remarks come a month after Uber CTO Praveen Neppalli Naga, in another interview with The Information, said that the company had exhausted its annual AI just four months into 2026. Naga’s comment led to a “head-exploding moment” and sparked intense discussions about AI token consumption within Uber and the trade-offs it creates, such as on head count, according to Macdonald.
For context, Uber spent $3.4 billion – nine percent more than it had spent the previous year – on research and development efforts in 2025. Earlier this month, Uber CEO Dara Khosrowshahi said in an earnings call that the company was making up for its increasing AI investments by hiring fewer human employees.
Macdonald’s remarks have grabbed headlines because it comes at a time when several big tech companies are urging employees to use AI tools as extensively as possible, a trend increasingly referred to as ‘tokenmaxxing’. Meta, for instance, is even evaluating employees’ performance based on their usage of AI tools.
For the past few years, companies such as OpenAI and Anthropic offered low-cost or free access to advanced AI systems, fueling a surge in demand as investors poured billions to help them scale and build out their compute infrastructure. Now, however, these investors are expecting returns on their investments.
While OpenAI has rolled out in-platform advertisements, Anthropic’s response has been to move away from flat-rate pricing for enterprises toward per-token billing. This is now having downstream effects, with even major tech companies starting to feel the pinch.
For instance, the Experiences + Devices unit within Microsoft, which includes engineers responsible for Windows, Microsoft 365, Outlook, Microsoft Teams, and Surface, is looking to wind down their usage of Claude Code by the end of June, according to a report by The Verge. While part of the reason is to steer developers toward Microsoft’s own Copilot CLI, the decision is also reportedly tied to financial considerations.
Several medium and large organisations have incorporated AI tools into their workflows to boost productivity, but whether using AI tools translates to creating real value is an open question.
Research on the subject remains mixed. In January, developer analytics platform GitClear published a report which found that regular AI users averaged 9.4 times higher code churn than their non-AI counterparts, nearly double the productivity gains provided by AI coding tools.
However, other studies suggest that by the limitations and unreliability of AI coding tools. Data from Waydev, another developer analytics firm, shows that while initial AI code acceptance rates are between 80-90 per cent, the real-world acceptance rate falls to 10-30 per cent after subsequent revisions of the AI-generated code by engineers.
The disconnect between rising AI adoption and stagnant revenue growth may stem from the fact that tools like Claude Code have dramatically lowered the cost of producing code, but not necessarily the cost of producing meaningful outcomes.
As a result, companies are increasingly looking to . “This has to happen otherwise the balance sheet of the company goes in disarray. Your entire unit economics of the SDLC (software development lifecycle) goes for a toss – if you spend 50% more in input, with no or little change in outcome,” Arnav Gupta, an engineering manager at Meta, wrote in a recent essay published on X.
— Arnav Gupta (@championswimmer)
Gupta further argued that layoffs will continue to happen to make room for AI spending till companies figure out how to use AI, specifically how to “convert AI-tokens into outcomes and not just input.”
