Two months after Claude Code burned through Uber's annual budget, the COO bluntly stated: There is no proportional relationship between Token consumption and useful output.

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1734 words

Uber COO Andrew Macdonald admitted in a recent interview that the company's AI spending is becoming increasingly difficult to justify internally. CTO Praveen Naga revealed two months ago that the Claude Code budget had been burned through ahead of schedule, but the more fundamental issue is: higher token consumption isn't yielding a proportional increase in consumer feature output. (Background: Not just ride-hailing — Uber partners with Expedia to add hotel bookings, moving toward a one-stop travel super App) (Context: Anthropic report: In the 2028 battle for AI supremacy, if the US fails to defend its computing power advantage, it could be overtaken by China) When every engineer at a company burns up to $2,000 a month on AI tools, 70% of committed code comes from AI generation, yet no one can answer "how many features did this actually deliver" — the issue is no longer a technical one, but a management crisis. Uber COO Andrew Macdonald recently sat down for a Rapid Response interview and said out loud what the tech industry has tacitly understood: the money spent on AI is getting harder and harder to justify. Earlier, when Uber CTO Praveen Neppalli Naga was interviewed by The Information this past April, he said something similar: "The budget I thought we had has already been burned through ahead of schedule." The context at the time: among Uber's 5,000 engineers, adoption of Claude Code surged from 32% to 84% within just a few months. Individual engineers were spending between $500 and $2,000 per month; Naga himself once consumed $1,200 worth of token quota in two hours during an internal demo. Macdonald described how this statement caused tremors among Uber's senior leadership, sparking a chain of discussions about AI token consumption — including whether such spending is worth it, and the trade-off pressure it places on headcount. CEO Dara Khosrowshahi made it explicit during this month's earnings call: Uber is slowing down hiring, partly to offset AI investment costs. In other words, the AI tooling bill is starting to affect real hiring decisions. In the interview, Macdonald described what he discovered after speaking with senior engineering leaders at Uber: higher token usage has not translated into a proportional increase in consumer feature output. "That link doesn't exist yet, right?" he said. "Maybe there's vaguely more stuff being delivered, but drawing a line between those numbers and 'we shipped 25% more useful consumer features' is very difficult." This question exposes the core contradiction in the current AI adoption wave: token consumption is measurable, but what it measures is "degree of use," not "output value." Salesforce has recently labeled such metrics "vanity metrics" and has explicitly opposed using token consumption as a yardstick for employee performance. Notably, Macdonald also pointed out a cognitive blind spot: for individual engineers who don't pay out of pocket, AI tools "feel free," and they can experiment with various use cases at will; but ultimately, it's the company footing the bill. This cost misalignment between the individual and the organization is one of the structural reasons token consumption is spiraling out of control. Uber's confusion is not an isolated case — it's just the first time a top executive has called it out directly. At I/O 2026, Google vigorously promoted "tokenmaxxing" — that is, using AI as heavily as possible, and using that as one of the metrics for measuring engineer engagement. The logic of this approach is: usage itself will drive capability evolution, and quantitative change will eventually trigger qualitative change. But some companies are starting to go in the opposite direction. Duolingo once included AI usage frequency in performance reviews, but after employees raised the question, "Are we supposed to use AI just for the sake of using AI?", quietly retracted the policy. CEO Luis von Ahn said in a podcast interview in April: "It felt like, rather than holding people accountable for actual outcomes, we were pushing something that in many cases simply didn't apply." A healthcare company's case is even more extreme: it consumed 1 trillion tokens in six months, generating over $6 million in unplanned costs, and the finance department didn't even know what was driving it at the time. This isn't a problem of using AI — it's that nobody knew who was using it, what for, or how much money was being burned. In the interview, Macdonald did not announce any specific cuts, nor did he say Uber would abandon AI tools. He simply put into words a problem that pervades the corporate world but is rarely stated bluntly by senior leadership. There's no industry-standard answer yet for measuring AI return on investment. But mounting signs suggest the gap between "how much you used" and "how much you got" is still enormous.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-05-24

Claude Code Cost-Saving Hack: Engineers Save 300 Million Tokens a Week with Caching—The Key is Not to Interrupt

2026-05-28

Anthropic launches its latest model, Claude Opus 4.8! Claude Code simultaneously adds "Dynamic Workflows" — one person can match a hundred-person development team

Similarity 170%關鍵字 code/claude同分類 zh

2026-05-24

My 30-person company went all-in on Claude Code, and the result is "humans have more work to do than ever."

Similarity 170%關鍵字 code/claude同分類 zh

2026-05-22

Andrej Karpathy's distillation of the "CLAUDE.md Four Principles" goes viral on GitHub, boosting AI coding accuracy to over 90%