How to unlock AI’s full potential – Uber’s CEO, Dara Khosrowshahi

When Uber chief executive Dara Khosrowshahi spoke about artificial intelligence at the World Economic Forum in Davos, his message cut against the celebratory tone that often dominates conversations about corporate AI adoption. The real divide, he argued, is no longer between companies that use AI and those that do not, but between those willing to fundamentally rewire how they operate and those content to stage what he described as a “pretend transformation.”

Many companies, he said, have learned how to talk fluently about AI without allowing it to meaningfully reshape their businesses. Tools that summarize documents, generate presentations, or draft emails may save time, but they do not alter decision-making, incentives, or outcomes in a way that creates lasting advantage. These uses are visible and easy to deploy, which makes them attractive to executives eager to signal progress, yet they largely leave existing structures intact.

What differentiates serious adopters, in his view, is whether AI is embedded into the organization’s core processes rather than bolted on at the edges. That shift requires companies to confront an uncomfortable reality: most large organizations are not just collections of people and products, but dense webs of rules, policies, and exceptions built up over years. AI, Khosrowshahi suggested, exposes how brittle and outdated many of those frameworks have become.

Uber’s own journey with AI-powered customer service illustrates this tension. Early efforts focused on training AI systems to follow the same policy manuals used by human agents. While this approach delivered modest efficiency gains, it also reproduced the limitations of the old system. The breakthrough came only when Uber’s engineers scrapped the inherited rulebook and redesigned the system from first principles.

Instead of encoding thousands of if-then rules, developers defined a small set of underlying objectives. One of the most important, Khosrowshahi said, was ensuring that customers felt satisfied at the end of an interaction. From there, the AI was allowed to reason through how best to achieve that goal, rather than mechanically enforcing policies designed for a different era. Letting the system optimize for outcomes rather than compliance, he argued, proved far more powerful.

This approach has broader implications for how companies think about control and trust. Traditional corporate systems rely on rigid rules to manage risk and ensure consistency. AI systems that reason toward goals introduce flexibility, but also require leaders to accept less predictable paths to those outcomes.

For many organizations, that trade-off is deeply unsettling, which helps explain why genuine transformation has been slower than the pace of AI investment might suggest.

Khosrowshahi framed this challenge bluntly, describing companies themselves as “a bunch of policies.” To unlock AI’s full potential, those policies must be questioned, dismantled, and in many cases rewritten entirely. That process, he acknowledged, is rarely smooth. Internal resistance, missteps, and operational failures are part of the journey.

“You have to survive through a bunch of car crashes internally,” he said, capturing the trial-and-error reality behind glossy AI announcements.

Uber’s internal tooling reflects how deeply the company is pushing into this transition. Developers are using AI-assisted coding and reasoning tools such as Anysphere’s Cursor and Anthropic’s Claude, integrating them directly into software development workflows. This signals a move beyond experimentation toward treating AI as infrastructure rather than novelty.

Khosrowshahi’s remarks come amid rising corporate spending on AI, driven by expectations of productivity gains and competitive pressure. A recent RBC Capital poll found that 90% of IT professionals expect their organizations to increase AI spending this year. Yet alongside that surge is a growing unease that AI may not be delivering the transformative gains many executives promised shareholders and employees.

Concerns about skill erosion, over-automation, and the hollowing out of institutional knowledge are becoming harder to dismiss. Khosrowshahi did not deny these risks, but his comments implied that shallow adoption may be the greater danger. Companies that deploy AI superficially may bear the cultural disruption of change without reaping strategic rewards, leaving workers disillusioned and systems no more effective than before.

His argument ultimately reframes the AI debate away from tools and toward leadership choices. The question is not whether a company has an AI strategy, but whether it is prepared to rethink how decisions are made, how success is defined, and how much legacy structure it is willing to abandon. In that sense, separating the pretenders from the real deal in AI adoption is less about technology sophistication and more about organizational courage.

Khosrowshahi indicates that AI becomes a catalyst for redesigning how the business functions at its core for firms willing to do the hard work. For those that are not, it risks becoming another layer of automation draped over systems that were never built to learn, adapt, or reason in the first place.