Uber is shutting the door on its infamous tokenmaxxing era.
In a Wednesday X post, Uber CTO Praveen Neppalli Naga said the company is seeing some “very interesting trends on AI costs,” and that this was “another signal that we’re coming to the end of the so-called ‘tokenmaxxing’ era.”
Tokenmaxxing is an enterprise AI trend that emerged in the first half of 2026, in which companies urge their employees to adopt AI as much as possible in their workflows. Some companies made AI usage a performance metric that staff would be evaluated on.
Naga said that since the beginning of the year, the number of people at Uber using frontier AI tools has quadrupled, but this has coincided with a decline in per-AI-token costs.
The company managed to lower costs by improving its prompt caching process, using better default models, giving engineers better visibility into their AI usage, and experimenting with open-weight models, Naga wrote in his post.
“The next phase, whatever we call it, will not be characterized by who spends the most tokens, but about how people use them as efficiently as possible,” he added.
Uber’s finance chief, Balaji Krishnamurthy, shared similar updates during the company’s second-quarter earnings call on Wednesday.
“On AI, we are very early, but what we are seeing is that we are able to cost-efficiently deliver some productivity lifts with developers,” he said.
“And for the measurement that we are looking at right now, we are seeing doubling in the code output for engineers,” Krishnamurthy added.
Uber made headlines earlier this year for igniting the tokenmaxxing trend, with Naga saying in April that the company had already blown through its 2026 budget for Anthropic’s Claude Code. He said in a March LinkedIn post that 1,800 code changes weekly were entirely written by its internal coding agent.
But in May, Uber COO Andrew Macdonald said in an interview that it was getting harder to justify the trade-offs of AI investments in the company. He said he wasn’t seeing proportional productivity gains from the increased AI costs.
This is not only an Uber problem; the rest of the tech industry has been grappling with how to get better returns on investment from their highly inflated AI spending. Some, like Coinbase, have said they’re experimenting with model switching, which involves assigning the most challenging tasks to frontier models and offloading easier, repetitive tasks to cheaper ones.
The problem of enterprise AI spending has also driven a new wave of businesses geared toward helping companies reduce their costs. Some are consultancies that give executives advice on how to allocate their AI budgets; others are building products like inference infrastructure to help companies scale their AI products cost-effectively.
