Uber COO Questions ROI of AI Tokenmaxxing

According to Business Insider, Uber operations chief Andrew Macdonald said it is getting harder to justify the money spent on what reporters call "AI tokenmaxxing." Business Insider quotes Macdonald saying that, based on talks with senior engineering leaders, increased token usage has not produced a proportional rise in useful consumer features, asking, "That link is not there yet, right?" The article also reports that, per The Information and Business Insider, Uber CTO Praveen Neppalli Naga told The Information in April that Uber had already blown through its Claude Code budget for 2026, triggering internal debate.

Industry-pattern observations: Rapid increases in generative-AI token consumption commonly raise three technical cost-pressure points for large platforms: compute spend on large-context models, production-quality evaluation overhead for model outputs, and engineering effort to integrate models into robust user workflows. Those costs often scale nonlinearly with usage and can outpace short-term feature velocity unless teams hard-link model outputs to measurable user metrics.

Editorial analysis: Public reporting places Uber's comments alongside other firms that are reassessing how they measure AI value. Business Insider cites examples such as Duolingo, which faced employee pushback after tying AI usage to performance reviews, illustrating wider debate about usage-for-usage's-sake metrics versus outcome-based KPIs. For platform operators and ML engineers, this episode highlights an intensified focus on per-feature ROI, cost-accounting for third-party models, and the need for production instrumentation that ties token spend to user-facing impact.

Industry context: Observers should track three indicators that would show whether firms are moving from token-centric metrics to outcome metrics:

• •Adoption of ROI or lift-based A/B testing tied to AI-enabled features
• •More granular internal chargeback or cost-allocation for API/model calls
• •Public or investor commentary quantifying AI spend versus feature or revenue impact

For practitioners: expect increased emphasis on measurement frameworks that connect model calls to user engagement, retention, or revenue, and on engineering work to reduce unnecessary token usage through caching, prompt engineering, or smaller specialized models.

All direct quotes and the reporting on the Claude Code budget are attributed to Business Insider and its reporting of an April interview in The Information, per Business Insider's coverage. The article did not contain an official Uber statement of broader strategy or specific next steps.