Uber’s engineering teams have built and are using an internal “Dara AI” that mimics CEO Dara Khosrowshahi’s feedback style, a tool they consult to rehearse pitches and stress-test decks before stepping into executive meetings. The reveal came from Khosrowshahi himself on The Diary of a CEO podcast, underscoring how deeply generative AI is now woven into the company’s decision-making rituals.
The boss-bot is not a public product. It is essentially a leadership simulator that helps teams anticipate what the CEO will challenge, where he will push for data, and how he frames trade-offs. The goal: reduce friction, tighten arguments, and arrive at decisions faster.
Inside Dara AI: how teams rehearse executive scrutiny
According to Khosrowshahi, some groups now present to the AI first, refining narratives and metrics until they can withstand the scrutiny they expect from the real thing. That matches Uber’s self-image as a “giant code base,” where leaders view engineers as system builders as much as product owners.
This is not a chatbot for casual Q&A. Think of it as an executive proxy trained to surface gaps: missing cohort analyses, unit economics that do not square with geographic expansion, or operational edge cases like surge anomalies. In practice, it acts as a force multiplier for preparation.
Adoption and productivity claims from Uber engineers
Khosrowshahi says about 90% of Uber’s software engineers now use AI in their daily work, with roughly 30% operating as “power users” who are rethinking architecture with AI at the center. He describes the productivity shift as unlike anything he has seen in his career.
External benchmarks help contextualize those figures. GitHub reported that developers using its Copilot assistant completed coding tasks up to 55% faster in controlled studies, while developer surveys from organizations like Stack Overflow show a strong majority experimenting with or adopting AI coding aids. Uber’s internal numbers sit on the leading edge of this broader curve.
For a company orchestrating real-time logistics across riders, drivers, couriers, and merchants, even small efficiency gains compound. If AI-driven prep trims minutes from meetings, reduces iteration cycles on experiments, or raises the hit rate on product bets, the operational ripple effects can be meaningful.
Why build a boss simulator to streamline decisions
Executive time is a scarce resource, and leadership expectations are often embedded in unwritten heuristics. A faithful model of a CEO’s style helps teams internalize those heuristics without waiting for a meeting’s live feedback loop. That means fewer surprise objections and more time spent on high-leverage decisions.
There is a cultural angle, too. Uber has long prized quantitative rigor. An AI that insists on clear counterfactuals, guardrail metrics, and sensitivity analyses nudges the organization toward better habits. It also democratizes “institutional memory,” capturing how leadership has weighed similar trade-offs in the past.
The catch with cloning leadership using AI models
Boss-bots carry real risks if misused. They can fossilize a single perspective, amplifying confirmation bias and discouraging contrarian ideas. An AI trained on prior decisions may overweight past context and underweight emergent signals, especially in zero-to-one product areas.
There are governance concerns as well. Any model tuned on executive communications or sensitive planning documents must have strict data controls, clear retention policies, and auditable access. Companies will also need to label outputs as advisory, not authoritative, so humans maintain accountability for decisions.
Finally, tone matters. If employees treat the AI’s verdict as final, it can distort organizational dynamics. The healthiest pattern is using a boss-simulator to sharpen arguments while explicitly encouraging dissenting views and controlled contrarian experiments.
What to watch next as AI twins enter the workplace
Expect these “leadership twins” to proliferate. Enterprises already run AI assistants for customer support, fraud triage, and developer tooling; internal executive models are the next logical step. The question is how deeply they’ll connect into company data—think auto-summarized postmortems, OKR progress synthesis, or scenario planning that updates as metrics shift.
For Uber, the signal is clear: AI is not just writing code; it is shaping how the company decides what code to write. If the internal gains match the CEO’s enthusiasm, other large-scale platforms will likely follow, blending human judgment with AI-fueled preparation to move faster with fewer blind spots.
Khosrowshahi’s admission, first flagged by Business Insider and aired on The Diary of a CEO, captures a broader transition. The future of leadership may be less about solitary judgment at the top and more about teaching machines to ask the right questions—so people walk into the room already knowing the answers.
