Uber COO Andrew Macdonald said on the Rapid Response podcast that the company's roughly 5,000 engineers exhausted their 2026 Anthropic Claude Code token budget by mid-March, and that the link between AI coding-assistant spend and shipped product value "is not there yet." The dollar figure was not disclosed. The data point under the soundbite is the 2.5-month burn through an annual enterprise budget โ concrete evidence that Claude Code consumption at engineering-org scale exceeds prevailing planning models, regardless of whether the productivity case eventually closes.
The framing matters because the public conversation has settled into binary modes โ AI coding assistants are either transformative or overhyped. Macdonald's actual quote is more useful for builders: "if you're not actually able to draw a direct line to how much useful features and functionality you're shipping to your users that trade becomes harder to justify because it's not free." That is a productivity-attribution problem, not a tool-quality problem. Engineers used Claude Code heavily enough to blow past the annual budget at the 21% mark of the year. The bottleneck is connecting that usage signal to shipped-feature throughput in a way finance and engineering both accept, and no major enterprise has published a methodology that works at 5,000-engineer scale.
The ecosystem read for builders shipping AI coding tools: heavy enterprise adoption does not translate automatically into renewal economics. Anthropic, GitHub Copilot, Cursor, and the rest are competing not just on model quality but on the question their customers cannot yet answer โ what is the unit economics of AI-augmented engineering. Engineering finance teams will increasingly demand per-team attribution data (which features shipped while which engineers were using which tools), and the vendor that ships the audit/attribution layer first wins the enterprise renewal pricing conversation. The honest signal in Uber's data is also visible at smaller scale: any team running a heavy coding-assistant deployment is going to over-shoot its first budget estimate. Plan accordingly.
If you run engineering finance Monday morning: budget for 2-3x your initial coding-assistant estimate at enterprise scale and start instrumenting attribution from day one. If you sell AI coding tools: the renewal conversation is shifting from "use this tool" to "prove the productivity link," and the customers who cannot prove it will pull back on seat counts. The Uber story is not "AI is not worth it" โ it is "we cannot yet measure whether it is," which is a different and more actionable read.