Uber's Engineers Didn't Ask Permission
84% of Uber's developers adopted agentic AI tools in one quarter. No committee approved it. No pilot program blessed it. The enterprise adoption debate just ended.
Somewhere around December, an engineer at Uber started using Claude Code. Probably didn’t file a ticket for it. Probably didn’t ask their manager. They just started using it because it made them faster, and the person sitting next to them noticed.
By February, 84% of Uber’s engineering org was on agentic coding tools.
That’s not a pilot program. That’s not an executive mandate. That’s a behavior that spread through an organization the way behaviors actually spread: one person shows another person, that person shows two more, and suddenly it’s just how things are done. Three months. No committee required.
The Part That Should Scare You
I keep thinking about the speed. Not the adoption number itself, but how fast the curve moved. Claude Code went from 32% to 63% usage inside Uber in one quarter. For context, most enterprise software rollouts take 12-18 months to hit 60% adoption WITH executive sponsorship, training programs, and change management consultants. This happened organically. Engineers just… switched.
Uber built three internal tools on top of the agentic stack while this was happening. Minion runs background agents. Validator handles security review. Autocover generates tests. They didn’t build these because someone wrote a strategy doc. They built them because the tools created possibilities that didn’t exist before, and engineers are engineers.
Meanwhile, Atlassian announced they’re cutting 1,600 people. Their CEO called it “self-funding an AI pivot.” Their stock is down 50% this year. I don’t think these two stories are unrelated. One company’s engineers moved without asking. The other company’s leadership is still trying to figure out what the plan should be.
What Nobody Talks About
There’s a detail in Uber’s data that matters more than the headline number. Token costs are becoming a real budget line item. CTOs are starting to manage token spend the same way they manage AWS bills, with dashboards and budgets and cost allocation to teams.
This is what happens after adoption. The conversation shifts from “should we use AI” to “how do we pay for all the AI we’re already using.” The interesting thing is that nobody planned for this transition. It just happened because engineers adopted faster than finance could model.
Staff engineers are being told to get hands-on with the tools again. Not “oversee AI adoption.” Not “develop an AI strategy.” Use the tools. Write code with them. The role is snapping back from pure architecture and strategy to production work, because what production work looks like has changed underneath everyone’s feet.
The Dodge That Won’t Work
You already know what the skeptic says. “Uber is a tech company. This won’t translate to the rest of the industry.”
I used to think this was at least partially true, that consumer tech would adopt first and everyone else would lag by years. I don’t think that anymore.
Uber’s adoption was bottom-up. No one in leadership decided this should happen. Individual engineers chose the tools because the tools made them faster. That dynamic isn’t unique to Silicon Valley. It plays out anywhere people write code and want to go home on time. The tools don’t know what company they’re being used at.
The economics don’t change either. If agentic tools save 21,000 developer hours at Uber, they save proportionally the same hours at any company with developers. Developer salaries don’t get cheaper because you’re in a different sector.
What IS different at slower-moving companies is what happens when leadership catches up. At Uber, leadership got out of the way. At a lot of other places, leadership will try to manage it, control it, run it through procurement and compliance and a six-month evaluation. By the time they approve the tools, their best engineers will have been using them secretly for a year, or they’ll have left for somewhere that doesn’t make them fight for basic productivity.
What I’m Actually Watching
The thing I can’t stop thinking about is the tipping pattern. Adoption sat at pilot levels for months. Then it crossed some invisible threshold where enough engineers were visibly shipping faster that the holdouts couldn’t pretend it wasn’t happening. Then it exploded. 32% to 63% in ninety days.
Every engineering org is on this same curve somewhere. Some are at 10% and think they’re “evaluating.” Some are at 40% and think they’re “ahead of the industry.” The ones that hit 80% didn’t get there by planning. They got there by not blocking what was already happening.
The evaluation committees debating whether to adopt are answering a question their engineers decided months ago. The only thing left to decide is whether leadership catches up gracefully or gets dragged.