Uber is wrapping up its widely discussed tokenmaxxing phase.

In a post shared on X this Wednesday Uber CTO Praveen Neppalli Naga pointed to notable patterns in AI spending and described them as a clear sign the company is moving past the tokenmaxxing period.

Tokenmaxxing refers to a corporate AI practice that gained traction in early 2026 where firms push staff to integrate AI tools into daily work as extensively as possible with some organizations even tying usage levels to performance reviews.

Naga observed that the number of Uber employees using advanced AI tools has grown four times since January while the cost per token has dropped during the same stretch.

The firm achieved these savings through refined prompt caching stronger default model selections improved visibility for engineers on their AI consumption and trials with open-weight models as Naga detailed in the post.

He added that the coming stage will be defined not by token volume but by how effectively those tokens are applied.

Uber finance chief Balaji Krishnamurthy echoed these points during the companys second-quarter earnings discussion on Wednesday.

He stated that while the company remains in early stages of AI integration it has already achieved cost-effective productivity improvements among developers.

Krishnamurthy also reported that current metrics show engineers producing twice as much code output.

Uber drew attention earlier this year as an early driver of the tokenmaxxing movement after Naga revealed in April that the firm had exceeded its full 2026 allocation for Anthropics Claude Code and noted in a March post that its internal coding agent was generating 1800 code changes each week.

By May however Uber COO Andrew Macdonald indicated in an interview that justifying continued AI spending was becoming more difficult due to uneven productivity returns relative to rising costs.

The challenge of balancing AI investment returns extends beyond Uber as other technology firms examine ways to improve outcomes from elevated AI budgets with some exploring model switching strategies that route complex tasks to premium models and simpler work to lower-cost options.

Enterprise concerns over AI costs have also spurred new service providers offering budget allocation guidance or inference tools designed to help organizations scale AI usage more economically.