Uber’s tokenmaxxing era is drawing to a close, with the company’s chief technology officer declaring that the next phase of enterprise AI will be defined by efficiency rather than raw consumption. In a post on X on Wednesday, CTO Praveen Neppalli Naga wrote that Uber is seeing ‘very interesting trends on AI costs’ and called it ‘another signal that we’re coming to the end of the so-called “tokenmaxxing” era.’
What Ended the Tokenmaxxing Era
Tokenmaxxing emerged in the first half of 2026 as companies began urging employees to use AI tools as heavily as possible, sometimes tying adoption to performance reviews. Uber was among the loudest proponents. According to Fortune, Naga admitted he went ‘back to the drawing board’ on AI spending after the company encouraged staff to use Anthropic’s Claude Code without restraint, and that Uber built leaderboards to rank software engineers on their AI usage.
The pivot to restraint was not cost-free to learn. In April, Naga said the company had already blown through its entire 2026 budget for Claude Code. A March LinkedIn post from Naga, reported by Business Insider, revealed that Uber’s internal coding agent had grown from producing less than 1% of all code changes to 8%, with 1,800 weekly changes written with zero human authoring, engineers only review and approve.
But the cost curve arrived with the adoption curve. Since the start of 2026, the number of Uber employees using frontier AI tools has quadrupled. Per-token costs fell over the same period. Naga attributed the decline to four levers: improved prompt caching, better default models, experimenting with open-weight models, and giving engineers visibility into their AI usage. Fortune reports that last item was more specific than it sounds: engineers were shown their costs per hour, a granular feedback mechanism that the company credits as a genuine check on spending.
Naga’s conclusion from those trends was pointed. ‘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 wrote.
AI Efficiency Gains Show Up in the Numbers
The shift in philosophy landed alongside Uber’s second-quarter earnings. During the Q2 2026 call, CFO Balaji Krishnamurthy confirmed that near 100% of engineers now use AI-based coding tools, according to the company’s Q2 investor relations press release. Krishnamurthy added: ‘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. And for the measurement that we are looking at right now, we are seeing doubling in the code output for engineers.’
AI’s reach at Uber extends beyond the engineering function. The Q2 earnings call transcript shows the company’s Cart Builder feature is increasing average order sizes by up to 2x for its delivery business, and that 75% of rides now use personalised destination suggestions powered by AI.
The broader financial results give context to those operational claims. Gross Bookings grew 22% in the quarter, Non-GAAP EPS grew 35%, and trailing twelve-month free cash flow exceeded $10 billion for the first time in the company’s history, according to the same press release.
The doubts about AI returns were not absent from Uber’s own leadership, even before Naga’s Wednesday post. In May, COO Andrew Macdonald said in an interview that it was becoming harder to justify the trade-offs of AI investment, with productivity gains not running proportionate to the costs incurred. The challenge is not unique to Uber; across the technology sector, companies are seeking better returns on AI spending, with some, like Coinbase, experimenting with model switching, routing complex tasks to frontier models and cheaper repetitive work to lower-cost alternatives.
Fortune notes that Naga did not specify whether Uber is consuming more or fewer total tokens than earlier in 2026. The declaration of the tokenmaxxing era’s end is a shift in philosophy, not necessarily a confirmed drop in volume. Whether peers follow Uber’s public pivot, or whether the token counts quietly keep climbing, will become clearer when the next round of earnings calls arrives.