Uber President Admits "Tokenmaxxing" Is Just Burning Cash for No Result

Between January and April 2026, Uber saw its engineering team's usage of Claude Code spike from 32% to 84%. To fuel this obsession, the company effectively vaporized an entire year’s worth of AI budget in just four months. While management initially justified this as a smarter alternative to hiring more humans, the reality check has finally arrived. Andrew Macdonald, the president of Uber, recently admitted that he was stunned when the company's CTO, Praveen Neppalli Naga, revealed the extent of the financial carnage.

The root of this mess is a trend known as "tokenmaxxing," where engineers are judged by how heavily they rely on AI tools. Uber even baked this into internal performance metrics, effectively gamifying the depletion of the corporate treasury. The company currently employs around 5,000 engineers, each burning between $500 and $2,000 monthly on AI subscriptions, yet Macdonald noted that drawing a straight line between this exorbitant spending and a measurable increase in user-facing features remains practically impossible.

This isn't an isolated incident, as Microsoft recently pulled Claude Code licenses from internal teams to pivot toward GitHub Copilot CLI in a desperate attempt to stop the bleeding before the new fiscal year. Similarly, Duolingo had to walk back its own AI-based employee evaluation system after staff complained that the tool was being prioritized over actual product output. Uber’s leadership is now vocalizing what was once only whispered in dark corners of the tech industry: AI costs are anything but free when the bill finally arrives.

When the people writing the checks start questioning the utility of their own expensive toys, the bubble isn't just leaking—it's whistling. This pivot from "AI will save us" to "AI is a black hole for our margin" signals that the industry is finally waking up from a year-long fever dream where burning capital was mistaken for innovation.

Source: Business Insider