Uber CTO Praveen Neppalli Naga says: With 95% of its engineers now using AI, the role of engineers is shifting from writing every line of code to...

Uber's engineers are spending less time writing code—and more time telling machines what to write. That's not a concern, according to the company's CTO. It's the whole point. The ride-hailing giant, which once disrupted how cities move, is now betting that the next disruption is inside its own engineering teams—and that getting ahead of it is better than getting caught flat-footed.Praveen Neppalli Naga said in a LinkedIn post this week that 95% of Uber's engineers now use AI tools every month, and that its internal coding agent is producing 1,800 code changes a week entirely without human authoring. Engineers review and approve those changes, but they no longer write them. The agent's share of all code changes at the company has gone from under 1% to 8% in just a few months."There is zero human authoring," Naga wrote.

From autocomplete to autonomous agents—a shift hiding in plain sight

The more significant number isn't the volume. It's the nature of the work. Naga says 84% of AI users at Uber are now using agent-style workflows, where engineers hand off entire tasks rather than accept line-by-line suggestions. Claude Code adoption at the company nearly doubled in two months, from 32% to 63%. Traditional IDE-based tools have largely plateaued. Even within those IDEs, around 70% of committed code is now AI-generated—written, not suggested.

Put simply, the job is changing. Engineers aren't speeding up the same work—they're doing a different kind of work altogether.

A smooth story elsewhere has some rough edges

Uber's numbers land in the middle of a much messier industry-wide conversation about what AI-generated code actually costs companies. Amazon called an all-hands engineering meeting earlier this month after a string of outages, some tied directly to AI coding tools. Its internal Kiro agent once attempted to fix a problem by deleting and rebuilding an entire environment.

A CodeRabbit study found AI-written code had 1.7 times more issues than human-written code across 470 pull requests.Anthropic, whose Claude Code underpins tools at Uber and Spotify, rolled out a review feature this month specifically to catch errors in AI-generated output—at up to $25 per pull request. The creator of Claude Code himself, Boris Cherny, has said the models are still "not great at coding" and that vibe coding falls short when maintainability matters.

What engineers actually do now—and whether that's enough

Naga frames the shift as an upgrade, not a downgrade. "The role of the engineer is shifting—from writing every line to architecting systems and reviewing AI-generated code," he wrote. He also noted that Uber's adoption wasn't mandated from the top—the strongest uptake came from engineers quietly experimenting on their own. Andrej Karpathy, who coined the term vibe coding, recently put it differently: "I've never felt this much behind.

The profession is being dramatically refactored."Anthropic CEO Dario Amodei has said software engineering could be "fully automatable" within 12 months. Whether that's a prediction or a sales pitch depends on who you ask—but either way, Uber's CTO seems to be betting it arrives on schedule.