Uber's CTO says the company is coming to the end of its tokenmaxxing era

Uber is shutting the doorway connected its infamous tokenmaxxing era.

In a Wednesday X post, Uber CTO Praveen Neppalli Naga said the institution is seeing immoderate "very absorbing trends connected AI costs," and that this was "another awesome that we're coming to the extremity of the alleged 'tokenmaxxing' era."

Tokenmaxxing is an endeavor AI inclination that emerged successful the first half of 2026, successful which companies impulse their labor to adopt AI arsenic overmuch arsenic imaginable successful their workflows. Some companies made AI usage a capacity metric that unit would beryllium evaluated on.

Naga said that since the opening of the year, the number of group astatine Uber utilizing frontier AI devices has quadrupled, but this has coincided pinch a diminution successful per-AI-token costs.

The institution managed to little costs by improving its punctual caching process, utilizing amended default models, giving engineers amended visibility into their AI usage, and experimenting pinch open-weight models, Naga wrote successful his post.

"The adjacent phase, immoderate we telephone it, will not beryllium characterized by who spends the astir tokens, but astir really group usage them arsenic efficiently arsenic possible," he added.

Uber's finance chief, Balaji Krishnamurthy, shared akin updates during the company's second-quarter net telephone connected Wednesday.

"On AI, we are very early, but what we are seeing is that we are capable to cost-efficiently present immoderate productivity lifts pinch developers," he said.

"And for the measurement that we are looking astatine correct now, we are seeing doubling successful the codification output for engineers," Krishnamurthy added.

Uber made headlines earlier this twelvemonth for igniting the tokenmaxxing trend, pinch Naga saying successful April that the institution had already blown done its 2026 fund for Anthropic's Claude Code. He said successful a March LinkedIn station that 1,800 codification changes play were wholly written by its internal coding agent.

But successful May, Uber COO Andrew Macdonald said successful an question and reply that it was getting harder to warrant the trade-offs of AI investments successful the company. He said he wasn't seeing proportional productivity gains from the accrued AI costs.

This is not only an Uber problem; the remainder of the tech manufacture has been grappling pinch really to get amended returns connected finance from their highly inflated AI spending. Some, for illustration Coinbase, person said they're experimenting pinch exemplary switching, which involves assigning the most challenging tasks to frontier models and offloading easier, repetitive tasks to cheaper ones.

The problem of endeavor AI spending has besides driven a new activity of businesses geared toward helping companies trim their costs. Some are consultancies that springiness executives proposal connected really to allocate their AI budgets; others are building products for illustration conclusion infrastructure to thief companies standard their AI products cost-effectively.