Uber president says AI spending is getting ‘harder to justify’

Background

Uber has reportedly exhausted its annual AI budget just four months into 2026, creating pressure to evaluate whether its AI investments are justified. The specific focus in the excerpt is Claude Code usage, measured through rising token consumption. Tokens function as a proxy for AI tool usage and cost, but Uber is questioning whether this metric reflects actual product value.

Key Points

• AI usage is rising, but value is unclear.
  Andrew Macdonald says Uber is not yet seeing a clear connection between increased Claude Code token consumption and the delivery of useful features to consumers.

• The company lacks a reliable measurement link.
  Macdonald’s central point is that it is “hard to draw a line” between AI spending metrics and concrete product outcomes. More tokens consumed may indicate more activity, but not necessarily more meaningful output.

• Shipping more is not the same as shipping better.
  Macdonald acknowledges that AI may implicitly be helping the company ship more work. However, Uber cannot yet say that this translates into “25 percent more useful consumer” features or any similarly measurable improvement.

• The concern is return on investment, not AI adoption alone.
  Uber is not described as rejecting AI tools. Instead, it is asking whether heavy spending is creating measurable returns that justify the cost.

Significance

The article highlights a growing tension in corporate AI adoption: usage metrics are easier to measure than business impact. Token consumption can show how much employees rely on AI systems, but it does not prove that those systems improve the customer experience, accelerate meaningful development, or increase product quality.

Uber’s situation suggests that companies investing heavily in AI may face a measurement problem. If AI budgets are consumed quickly without a clear link to valuable outcomes, executives may become more skeptical of unlimited spending. The key challenge is moving from enthusiasm and adoption to evidence: AI tools must eventually be judged by deliverable improvements, not simply by how often they are used.