Uber Engineers Built AI Version of CEO Dara

Uber CEO Dara Khosrowshahi says employees have built an internal chatbot modeled on him, nicknamed “Dara AI”, and some teams use it as a practice room before presenting to Uber’s top leadership.

What “Dara AI” is

According to Khosrowshahi, teams “make the presentation to the Dara AI” first, using the bot to stress-test their narrative and refine slides before the real meeting. In his telling, the bot acts like a high-pressure rehearsal partner that helps polish decks to the level he typically sees at exec reviews.

What’s notable isn’t just the novelty of “an AI CEO,” but the workflow it implies: employees are treating executive feedback as something AI can simulate early, so the real meeting becomes less about basic clarity and more about decisions.

The hard numbers Khosrowshahi shared about AI inside Uber

Khosrowshahi attached unusually specific adoption stats to the story:

~90% of Uber software engineers use AI in their work

~30% are “power users” who use AI deeply enough that they’re “rethinking the architecture” of how systems should be built

He framed this as more than speedups in coding: engineers are acting both as “brick makers” (shipping components faster) and architects (redesigning systems with AI as a core tool).

Why Uber’s CEO thinks the “AI boss” idea still has a ceiling

Khosrowshahi’s caution was blunt: today’s models can process lots of information, but they still struggle with real-time learning and decision-making based on new, unfolding inputs, the kind executives deal with daily. He suggested the day models can learn “in real time” is when leadership roles become genuinely “replaceable.”

So “Dara AI” (as described) is closer to a simulation of how Dara reacts than an autonomous executive.

The hiring tradeoff: more engineers or more AI agents + GPUs

Khosrowshahi laid out a pragmatic fork in the road.

If AI makes engineers ~25% more efficient, he said he may want to hire more engineers to move faster.

But he also floated the opposite: rather than increasing headcount, Uber could add AI “agents” and buy more GPUs from Nvidia.

That’s a real executive-level signal: AI isn’t only a productivity tool; it becomes a budgeting choice between people vs. compute.

The broader context: Uber’s AI push isn’t just internal tooling

The CEO also pointed to Uber’s operational dependence on AI across the core business, and noted expansion into new use cases, including an AI Solutions division that pays contractors to help train AI for external clients.

What we don’t know yet

Even with the headlines, several key facts aren’t publicly clear from Khosrowshahi’s description:

How “Dara AI” was trained (internal docs? meeting transcripts? public interviews?)

What guardrails exist (privacy, security, hallucination controls, who can access it)

How widespread usage is inside Uber (Khosrowshahi acknowledged that isn’t clear)

Whether it’s a one-off experiment or a formal internal product with governance

Those gaps matter because “AI that mimics an executive” is exactly the kind of tool that can accidentally leak sensitive strategy, unless it’s tightly controlled.

Why this story is bigger than a funny internal bot

Lots of companies talk about “AI transformation.” Uber’s “Dara AI” anecdote shows what that looks like when it becomes culture:

Employees don’t wait for IT to deploy a tool, they build one.

AI is used not only to write code, but to shape decision-making inputs (how presentations are crafted, framed, and defended).

Leaders start openly discussing a future where the constraint is no longer labor alone, but compute allocation (agents + GPUs).

In other words: this isn’t “AI helping employees.” It’s employees reorganizing work around AI first, and treating human executives as the final checkpoint, not the starting point.