Uber burns 2026 AI budget in 4 months on Claude Code — $500–$2k per engineer monthly
Uber burned through its full 2026 AI budget in just four months, driven by adoption of Claude Code and Cursor. The company's CTO confirmed that the annual AI allocation was exhausted by April, with monthly API costs ranging from $500 to $2,000 per engineer. The tools became so valuable that limiting access felt counterproductive, forcing leadership to reassess budgeting strategy.
Adoption trajectory
Claude Code access rolled out to Uber's engineering team in December 2025. Usage doubled by February as developers leveraged its multi-step capabilities. By April, the bill consumed the entire year's budget. Currently, 95% of Uber engineers use AI tools at least monthly, and 70% of all committed code originates from AI. Cursor usage has plateaued while Claude Code dominates workflows.
Cost implications
With Uber's R&D spending at $3.4 billion annually, the AI coding tool expenditure now represents a meaningful and unexpected chunk. The CTO stated the company is "back to the drawing board" on AI budgeting, trying to determine if this level of productivity can be sustained at scale.
Broader signals
Uber's experience suggests that other companies may face similar cost overruns as developer adoption of autonomous coding agents accelerates. The incident highlights that existing AI budgets, set before the current adoption curve, may be outdated before the fiscal year ends.
📖 Read the full source: HN AI Agents
👀 See Also
Claude Code developer acknowledges adaptive thinking flaw, provides workaround
Boris Charny, creator of Claude Code, confirmed a flaw in the adaptive thinking feature that causes performance degradation. Users experiencing issues even with effort=high settings can use CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1 as an interim workaround.
Analysis of Claude Code's ~12K Token Forced System Prompt Reveals Priority Rules Overriding User Config
An analysis of Claude Code's injected ~12K token system prompt shows priority rules for song lyric bans, subagent delegation, and brevity that override user CLAUDE.md and memory files.
Agent Memory Is Not a Storage Problem: It's an Authority Problem
A developer argues that agent memory fails not because retrieval misses, but because all notes return with equal authority. The fix: a graph with roles, expiration, and activation fields.
Ex-OpenAI Researcher Quits Anthropic Over AI Safety Fears
A former OpenAI researcher has left Anthropic, citing fears about losing control of AI systems. The resignation underscores ongoing safety concerns in the AI industry.