Your company probably handed engineers AI coding tools last year. Have you checked the invoice yet?

Uber just did. Their entire 2026 AI budget was gone by April. Not because the tools failed — because they worked too well.

The Setup

Uber gave 5,000 engineers access to Claude Code. Usage jumped from 32% in December to 84% by March. They gamified it with internal leaderboards. The CTO spent $1,200 in two hours during a single demo session.

Individual engineer costs ran $500 to $2,000 per month. That's the range if you're curious what your own team might be spending.

What Actually Happened

Before: Annual AI budget set in late 2025, based on assumptions about token consumption. Per-seat pricing logic from traditional SaaS contracts.

After: 70% of committed code now comes from AI. Roughly 11% of live backend updates are written by AI agents with zero human intervention. 1,800 code changes per week happen automatically.

What broke: The budget model. Token-based billing doesn't work like per-seat SaaS. Autonomous agents running refactors, tests, and backend generation chew through tokens at a rate annual cycles can't absorb.

What Uber learned: The teams driving adoption weren't the teams managing the budget. Internal leaderboards incentivized usage without connecting it to cost.

The Irony

The same CEO who subsidized rides for years with VC money to gain market share is now learning that aggressive AI adoption has the same cost structure. Dara Khosrowshahi told investors he doesn't know what happens to the 9.5 million drivers and couriers on the platform.

When asked about replacing human work, he said: "I do not know."

What Reddit Noticed

"Fire almost all computer engineers thinking it will save costs, but ends up spending all the savings on tokens." — FreshestCremeFraiche

"When people talk about how we're in an AI bubble, this is exactly the kind of thing that validates it." — chi_guy8

"My company has a weekly quota. I don't understand how they did this." — Most-Bookkeeper-950

So What

This isn't about Uber specifically. It's about what happens when you treat AI coding tools like regular software subscriptions.

Traditional SaaS: per-seat, predictable, annual contract.

Agentic AI: token-based, consumption-driven, no ceiling.

Uber's $3.4 billion R&D spend rose 9% year-over-year with AI as the primary driver. They're now testing OpenAI's Codex and considering local models like Qwen3.6-27B to shift costs from unpredictable token billing to predictable hardware depreciation.

The uncomfortable part: Most companies don't have FinOps for AI. No dashboards for team-level spend. No caps. No allocation. Just monthly invoices that keep climbing.

Sources

https://startupfortune.com/uber-burned-its-entire-2026-ai-budget-in-four-months-and-claude-code-is-why-finance-teams-should-be-worried/ https://aimagazine.com/news/why-uber-has-already-burned-through-its-ai-budget https://finance.yahoo.com/sectors/technology/articles/ubers-anthropic-ai-push-hits-223109852.html https://www.reddit.com/r/artificial/comments/1t1mhx6/uber_burned_its_entire_2026_ai_coding_budget_in_4/ https://www.reddit.com/r/singularity/comments/1ssp3bw/uber_blows_through_its_it_budget_for_ai_for_2026/