Uber CTO shows how Claude Code can blow up AI budgets
Uber’s chief technology officer has revealed that the ride‑hailing giant’s rapid rollout of AI coding tools, particularly Anthropic’s Claude Code, has forced the company to revisit its 2026 artificial‑intelligence budget after spending more than planned in just a few months. Praveen Neppalli Naga told insiders that engineering teams’ adoption of Claude Code and similar assistants has surged far beyond internal forecasts, with many engineers now integrating these models into daily workflows to generate, review, and refine backend code. The move has delivered speed and productivity gains, but also exposed the steep cost of scaling large‑language‑model usage across thousands of developers.
According to reports drawing on company disclosures and internal commentary, Uber’s AI and software‑development spending is embedded within a broader R&D envelope that reached 3.4 billion dollars in the prior fiscal year. Within that envelope, the portion allocated specifically to AI tools and platform development has been largely consumed in early 2026, driven by token‑based usage fees tied to Claude Code and other coding‑focused models. Engineers were given access to Claude Code in bulk late last year, and usage quickly ramped up, with data indicating that AI‑written or AI‑assisted code now accounts for a rising share of live updates to Uber’s backend systems. This has turned what was initially seen as a supplementary tool into a core, recurring cost center.
Naga described the situation as a “back to the drawing board” moment, emphasizing that Uber did not anticipate the pace at which engineers would embrace AI‑driven coding assistants. He noted that Claude Code has become the dominant tool within the company’s expanding AI toolkit, outpacing other coding‑focused models whose growth has stabilized. The CTO also signaled that Uber is now evaluating additional coding assistants, including OpenAI’s Codex, as it seeks to balance productivity, reliability, and cost control. For the company, the episode underscores the tension between unlocking faster engineering throughput and containing the financial risk of runaway AI‑compute bills.
Analysts point out that Uber’s experience reflects a broader pattern in the tech sector, where once‑modest AI experiments quickly morph into enterprise‑scale deployments with outsized token‑usage profiles. When thousands of engineers regularly invoke models to generate boilerplate, refactor legacy systems, or scan codebases for vulnerabilities, even “small” per‑request fees compound into multimillion‑dollar quarterly bills. The CTO’s remarks imply that Uber will likely introduce more granular cost monitoring, usage quotas, and policy guardrails for AI coding tools, while still preserving their role in accelerating development cycles. For other large employers, the episode serves as a cautionary signal that AI budgeting must now treat language models as permanent infrastructure, not one‑off experiments.
-
10:30
-
10:16
-
10:15
-
10:01
-
10:00
-
09:47
-
09:45
-
09:32
-
09:31
-
09:16
-
09:15
-
09:01
-
08:53
-
08:45
-
08:37
-
08:30
-
08:18
-
08:15
-
08:00
-
07:45
-
07:30
-
07:30
-
07:15
-
07:00
-
20:15
-
19:50
-
19:32
-
19:15
-
18:44
-
18:21
-
18:15
-
18:05
-
18:00
-
17:46
-
17:45
-
17:30
-
17:28
-
17:15
-
17:10
-
17:00
-
16:48
-
16:45
-
16:34
-
16:30
-
16:16
-
16:15
-
16:00
-
15:54
-
15:45
-
15:34
-
15:30
-
15:15
-
15:15
-
15:14
-
15:00
-
14:53
-
14:45
-
14:32
-
14:30
-
14:16
-
14:15
-
14:00
-
13:57
-
13:54
-
13:45
-
13:35
-
13:30
-
13:16
-
13:15
-
13:00
-
13:00
-
12:45
-
12:39
-
12:30
-
12:25
-
12:14
-
12:10
-
12:01
-
11:48
-
11:45
-
11:31
-
11:30
-
11:17
-
11:16
-
11:15
-
11:13
-
11:00
-
10:55
-
10:54
-
10:49
-
10:45
-
10:40
-
10:36
