Recoleta
GitHub
Source note

Uber blows through its AI budget in 1 quarter

Authors
ChuckMcM
Source
Hacker News
Published
Collection
Software Intelligence
AI Coding ToolsEnterprise AI SpendingClaude CodeSoftware EngineeringAgentic AI
Open originalMarkdown source

Summary

Uber reportedly used its full 2026 AI coding tools budget in four months, and its COO says the company cannot yet link Claude Code usage to more useful consumer features. The piece frames AI coding agents as a cost-control problem for enterprises, especially as token-metered usage grows.

Problem

  • Uber increased Claude Code use, but Andrew Macdonald said the company cannot draw a clear line from usage stats to shipping more useful consumer features.
  • The cost issue matters because agentic coding tools can consume many tokens per task, so higher adoption can raise total spend even when model inference gets cheaper.
  • Internal incentives can push tool usage without proving product impact, as shown by Uber’s leaderboard ranking teams by AI tool usage.

Approach

  • This is a business news article, not a research paper, so it does not introduce a new technical method.
  • The core mechanism described is simple: employees use Claude Code and similar agentic tools; those tools spend tokens while helping with coding or work tasks; usage-based pricing turns more tool activity into higher bills.
  • Uber encouraged adoption through an internal leaderboard based on total AI tool usage, which may have increased consumption before value metrics were clear.
  • The article compares Uber’s situation with Microsoft reportedly shifting engineers away from direct Claude Code licenses toward GitHub Copilot CLI.

Results

  • Uber reportedly spent its entire 2026 AI coding tools budget in four months.
  • Uber CEO Dara Khosrowshahi said about 10% of the company’s committed code is built by autonomous agents.
  • Macdonald said Uber cannot yet show that Claude Code usage produces something like “25% more useful consumer features.”
  • Gartner projects inference on advanced AI models will cost AI firms 90% less by 2030 than in 2025, but also says enterprise AI may not get cheaper because agentic models use more tokens and providers may keep part of the savings.
  • Gartner forecasts AI agent software spending will reach nearly $207 billion in 2026, up more than 139% from $86.4 billion in 2025.
  • Uber spent $951 million on research and development in Q1 2026, nearly 17% more than the same quarter a year earlier; its 2025 R&D spend was 3.4% higher than 2024, according to the article.

Link