The company that spent a decade convincing investors that self-driving cars were just around the corner now can't figure out if its AI coding tools are worth the monthly bill.
The Summary
- Uber blew through its entire 2026 Claude Code budget in just four months, according to CTO Praveen Neppalli Naga's April admission that went viral.
- COO Andrew Macdonald says there's no visible connection between rising token costs and actual feature delivery: "It's very hard to draw a line between one of those stats and 'now we're actually producing 25 percent more useful consumer features.'"
- The gap between AI spend and measurable output has created what Macdonald calls a "head-exploding moment" inside the company, forcing senior leadership to justify token consumption against traditional costs like headcount.
- This is the first major tech company to publicly question whether AI coding assistants deliver returns that justify their rising costs.
The Signal
Four months. That's how long it took Uber to burn through what they thought was a full year of AI budget. When CTO Praveen Neppalli Naga mentioned this in April, it wasn't meant to be a confession. It went viral anyway. Now COO Andrew Macdonald is doing damage control, and his explanation is more revealing than the original slip.
The problem isn't that Uber is spending money on AI. Every tech company is. The problem is that Uber's senior engineering leaders looked at their token usage graphs, looked at their feature velocity metrics, and couldn't find the correlation. More Claude Code consumption isn't translating into proportionally more shipped features. Macdonald's exact words: "That link is not there yet."
"It's very hard to draw a line between one of those stats and 'now we're actually producing 25 percent more useful consumer features.'"
This matters because Uber isn't a startup experimenting with AI on venture money. They're a $150+ billion public company with actual P&L accountability. When Macdonald talks about trade-offs against headcount, he's asking a question every CFO will ask in 2026: if we're spending $X million on tokens and can't measure the output gains, why aren't we just hiring more engineers?
The timing is brutal. AI coding assistants were supposed to be the safe bet, the clearly valuable AI application that justified enterprise budgets while everyone figured out whether chatbots and image generators would ever make money. Claude Code, Copilot, Cursor, they all promised the same thing: your developers ship faster, you need fewer of them, the ROI is obvious.
Except Uber's COO just said the quiet part loud. The ROI isn't obvious. Maybe there's "implicitly" more getting shipped, but implicit doesn't survive budget review season. Macdonald needs a line from token spend to feature output, and he doesn't have one.
Key questions this raises:
- Are developers using AI assistants to ship incrementally more features, or to ship the same features with less mental load?
- If velocity gains are real but unmeasurable, is the problem with AI tools or with how engineering orgs measure productivity?
- How many other companies are in Uber's position but haven't admitted it yet?
The Hacker News thread got 237 comments and 178 upvotes for a reason. Engineers know this story. They've watched token budgets balloon while management struggles to quantify the value. Some teams swear by AI assistants. Others think they're expensive autocomplete. Nobody has clean data proving which side is right.
What makes Uber's admission significant is that they're big enough and technical enough that if AI coding tools had obvious, measurable ROI, they'd have found it by now. They have the engineering bench, the data infrastructure, the budget. If they can't draw the line between spend and output, it might be because the line is blurry for everyone.
The Implication
Watch for two things. First, watch whether other public tech companies start asking harder questions about AI tooling budgets. Uber just gave every CFO permission to demand proof. Second, watch whether AI assistant vendors start selling value differently, moving from "your devs will ship faster" to metrics they can actually tie to revenue or cost savings.
For companies still scaling up their AI assistant deployments, Uber's experience is a yellow flag, not a red one. The tools might work. But if you can't measure how they work, you can't justify scaling them when budgets tighten. Figure out your measurement strategy before you blow through your annual budget in Q1.