Dive Brief:
- The engineering team at Uber Technologies has overtaken finance as the company’s leading adopter of artificial intelligence, a senior finance executive said Thursday, amid reports the ride-hailing company is grappling with high costs from AI tools used to automate software coding.
- More than 90% of Uber finance professionals now routinely use AI tools, according to Tiho Nedkov, a director of finance at Uber, highlighting the team’s rapid integration of the technology into daily workflows even as engineering has now emerged as the top deployer.
- “Finance actually led this [within] Uber for a couple years … at least until agentic coding came along,” Nedkov said at Gartner’s annual conference for corporate finance leaders held in National Harbor, Maryland.
Dive Insight:
The comments come amid reports of tensions at Uber over AI spending, driven by concerns about coding tools and token-based usage patterns that are pushing internal costs higher than expected. The development highlights a broader trend known as “token-maxxing,” an industry term used to describe heavy AI token usage.
Uber’s internal concerns over AI spending surfaced in a report by The Information last month, which said Chief Technology Officer Praveen Neppalli Naga disclosed the company had burned through its entire 2026 AI coding budget within the first four months of the year, driven by rapid adoption of tools such as Claude Code across engineering teams.
The issue was later echoed by Chief Operating Officer Andrew Macdonald in a Rapid Response podcast interview, where he indicated it was hard to show that rising AI token usage was delivering a return on investment.
“That link is not there yet,” he said, adding that while usage statistics were rising sharply, it was difficult to draw a direct line between those metrics and improvements in consumer services.
Uber’s rollout of Claude Code has accelerated rapidly across engineering teams, with adoption rising from 32% in February to 84% by March, according to Forbes. The report said roughly 95% of Uber engineers were using AI tools monthly by spring, with about 70% of committed code originating from those systems. That contributed to monthly costs ranging from roughly $150 to $250 per engineer on average, and as much as $2,000 for heavy users.
Nedkov said AI adoption in finance remains strong even as engineering has taken the lead overall.
Uber’s finance organization has pushed AI-driven automation across a range of workflows, with more than 96% of invoices now processed by the technology at over 95% accuracy and with reduced human-in-the-loop review time, according to a slide presentation used during Nedkov’s interview with Gartner’s Mallory Barg Bulman. Contract review for revenue recognition is fully automated, while regulatory notice handling has been cut by roughly 70% through multilingual AI workflows.
He also pointed to a “data agent” used for financial intelligence that enables conversational analytics over enterprise data, along with efforts to build a “process source of truth” that maps finance workflows through AI-driven systems. Nedkov also said there is growing interest in exploring new agentic AI use cases.
In an interview following his remarks at the Gartner conference, Nedkov said Uber has “basically democratized” access to AI tooling, allowing employees to build and deploy applications without significant barriers. He said the approach has enabled “a lot of innovation and productivity,” while also driving a significant proliferation of output, which he acknowledged can create “source of truth” challenges.
“But if you have guardrails, you can control that,” he said. “You have sort of a playground where people are developing these things.”
While most of Uber’s finance professionals are using AI, the team remains in the “early innings” of AI adoption, he said.
“There’s no sort of token-maxxing in finance,” Nedkov said. “But you do have to be mindful of that. You have to make sure limits are in place.”
Nedkov declined to comment on reports that Uber’s 2026 AI coding tools budget has already been exhausted. The company didn’t immediately respond to a request for comment.