20VC: Uber President on Budgeting AI at Uber: How AI Helps and Hurts Uber | Why Autonomy Is Existential | How to Beat DoorDash to #1 in Food | The Untold Stories of Travis Kalanick, Dara Khosrowshahi and China with Andrew MacDonald

Andrew MacDonald, Uber's president and COO with 15 years at the company, provides an in-depth look at Uber's current state as a $160 billion market cap business with $52 billion in annual revenue and 200 million monthly active users. He discusses the company's core principle of filtering all decisions through what's best for Uber, which he credits as the foundation for driving people and building trust without losing their loyalty.

On the AI budget situation that garnered headlines, MacDonald clarifies that CTO Praveen Kumar wasn't warning of runaway spending but noting that usage predictions were difficult. MacDonald emphasizes there is measurable ROI from AI in specific areas—capital allocation processes reduced from 15 hours to 2 hours, forecasting improved from 8 hours to 2 hours, and marketing QA reduced from 2 weeks to 2 days. However, he acknowledges the difficulty in quantifying ROI precisely because freed-up employee time gets redirected to other high-value activities rather than flowing directly to the bottom line. He proposes combining compute and headcount budgets to let trusted leaders allocate resources based on perceived ROI.

MacDonald addresses the membership program disagreement with Dara, admitting he was too focused on short-term price optimization and undervalued Uber One. He now acknowledges membership as the most efficient consumer lever with improving ROI over time as members consolidate mobility and delivery usage, increase LTV, and show lower churn.

On autonomous vehicles, MacDonald frames AV as existential to Uber's business but emphasizes that better technology alone won't determine winners—distribution will. He argues that even if Waymo or Tesla achieve superior autonomous technology, they'll ultimately need Uber's network of 200 million users and distribution advantages. He notes that in China, where multiple AV companies already operate, consolidation to a single player globally is unlikely.

Regarding scaling to 500 million users, MacDonald identifies price as the key inhibitor. Today's UberX at $35 per direction in New York City remains a luxury product compared to the baseline transportation economics that support mass adoption. Achieving scale requires multi-modal offerings (bikes, scooters, trains, autonomous vehicles) all at lower price points.

On the China experience, MacDonald describes Uber burning $52 million weekly in subsidies during the final negotiation phase with Didi, competing "with one hand tied behind our back" due to lack of WeChat access. He views the exit as reasonable given geopolitical realities and praises the company's effort to place Chinese team members in global roles.

MacDonald discusses the Delivery Hero acquisition as strategically important for expanding geographic footprint and accessing locally built brands with strong consumer mindshare (Argentina, Korea, Middle East). He notes delivery has grown faster than mobility and is nearly equal in size, yet receives little investor credit. He's currently running both mobility and delivery after their delivery leader departed, working effectively two jobs.

On competing with DoorDash for food delivery leadership, MacDonald declines to speculate about alternative histories if Travis were leading delivery instead, emphasizing the difficulty of operating challenging businesses with constant trade-offs. He respects Tony Xu and DoorDash's execution.

MacDonald addresses the "disaggregation" risk where AI agents might route consumers to cheapest providers. He argues that unlike e-commerce transactions, rideshare and delivery are managed transactions with complex handoffs, driver-rider interactions, and problem resolution that chat interfaces handle poorly. He credits Brian Chesky's insight that chat may not be the optimal interface for all transaction types.

On headcount, MacDonald believes Uber could operate with fewer people in five years due to AI's power, but will likely employ more people overall because the company will tackle entirely new business areas. He suggests companies extract AI efficiency through tighter target-setting rather than eliminating roles.

MacDonald reflects on lessons from Travis Kalanick—emphasizing creative problem-solving skills and explaining the reasoning behind decisions to create "many versions of yourself" in the organization. From Dara Khosrowshahi, he learned that "management comes from an org chart, leadership comes from the heart," emphasizing leading with both head and heart to build followership.

He credits Rachel Whetstone's advice to "always say yes" to new opportunities as influential to his career longevity and willingness to tackle unfamiliar challenges.