Uber’s operations chief just said the quiet part out loud: the company is finding it harder to justify what it spends on AI, and the spending-to-results math isn’t adding up yet. The story, surfaced on Hacker News, comes from a Rapid Response interview released Saturday with Uber COO Andrew Macdonald. His core message? More AI tokens burned does not automatically mean more useful products shipped.
What he actually said
Macdonald pointed to a moment that rattled leadership inside Uber. CTO Praveen Neppalli Naga told The Information in April that Uber had already blown through its Claude Code budget for all of 2026. That comment went viral and triggered what Macdonald called a “head-exploding moment” internally, kicking off real debate about token consumption and its trade-offs, including head count.
The key admission: after talking with senior engineering leaders, Macdonald realized higher token usage didn’t translate into a proportional jump in useful consumer features.
“That link is not there yet, right?” he said. “I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25% more useful consumer features.'”
He also named the trap plainly. AI feels free when you’re “just a user sitting there coming up with interesting use cases” without paying for it. But the company foots the bill.
Why this matters
This is significant because it’s a rare public crack in the tokenmaxxing consensus. Most of Big Tech has been pushing the opposite line: use AI as much as possible, and grade employees on how much they use it. Hearing a top exec at a company Uber’s size question the return on that spend is a shift in tone.
What stands out is the framing. Macdonald isn’t saying AI is useless. He’s saying the measurement is broken. Token counts are easy to track. Actual shipped value is not. When those two numbers stop moving together, the spend gets hard to defend to a CFO.
This isn’t happening in a vacuum. Earlier this month, CEO Dara Khosrowshahi said on an earnings call that Uber was slowing hiring to offset its AI investments. So the trade-off is already concrete: dollars going to tokens are dollars not going to head count.
The bigger trend
Uber isn’t alone in second-guessing the “use AI for the sake of it” mandate. Hacker News notes that Duolingo walked back its plan to fold AI usage into performance reviews after employees pushed back. They asked a fair question: did they have to use AI just to be seen using it?
Duolingo CEO Luis von Ahn summed up the discomfort in an April podcast: “It felt like, rather than being held accountable for the actual outcome, we were trying to just push something that in some cases did not fit.”
That’s the same tension Macdonald is describing, just from the people side. Mandating usage measures activity. It doesn’t measure outcomes.
What to watch next
A few things worth tracking if you build with or buy AI:
- Budget controls tighten. Expect more companies to put hard caps and approval gates on coding-agent spend, the way Uber clearly hit a ceiling with its Claude Code budget.
- New metrics emerge. The pressure now is to connect token spend to shipped, useful features. Whoever cracks that measurement problem changes how AI tools get sold.
- The usage mandate softens. If two visible companies are backing off “use AI or else,” others watching their own bills may follow.
My take: this is a healthy correction, not a retreat. The first wave was about proving AI could do the work. The next wave is about proving it’s worth the money. Those are different questions, and Macdonald is one of the first big-company executives to say so on the record.
The full interview and additional context are available at the original source.