Andrew MacDonald | 20VC Summary

1. Key Themes

Distribution as the Ultimate Moat in Transportation

MacDonald argues that regardless of who wins the autonomous vehicle technology race, distribution will ultimately determine who wins commercially. He draws a direct parallel to how McDonald's and Starbucks — despite massive 1P channels and fixed assets — still work with delivery marketplaces because utilization of expensive fixed assets is the overriding economic imperative.

"In the end, distribution wins. And look, of course, if only one player gets to the finish line on the technology side, then that is a problem for us. But that is not the future that I think we think will exist." 00:27:12

Autonomy Is Existential — But Not in the Way Most Think

MacDonald frames autonomy not as a threat to Uber's existence but as a product superiority curve that only moves in one direction. The existential risk is not participating. He also makes a subtle but important point: the majority of Uber's trips — in India at $2.53 and Brazil at $3.54 average fares — will not be addressable by AVs for decades, meaning the AV disruption is geographically concentrated in premium Western markets first.

"Autonomy is as bad as it's ever going to be today, right? And every single day it's going to get better. Then that's going to be the business. And that's going to be how people get around. And if Uber doesn't have autonomy on our platform, and we will, we are investing actively and aggressively to bring it to market. But if we didn't, then it certainly would be existential for our core business." 00:00:00

Membership Programs as the Highest-ROI Consumer Lever

MacDonald openly admits he was wrong to deprioritize Uber One in favor of price subsidies. The key insight is that membership LTV compounds over time — members ride more, cross-pollinate into Uber Eats, churn less, and are more resilient to competitive pressure. He calls it the single most efficient long-term consumer dollar deployed.

"The LTV of Harry just goes up over time with membership. You're less likely to churn. You're more resilient from a market share perspective. Like there's all these downstream long-term impacts that sort of multiply the value of that first dollar I put into membership." 00:10:04

AI ROI Is Real But Nearly Impossible to Precisely Quantify at Scale

MacDonald gives the most honest large-enterprise AI assessment heard publicly: the gains are real (capital allocation processes shrunk from 15 hours to 2, forecasting from 8 hours to 2, marketing QA from 2 weeks to 2 days), but the direct line from those gains to headcount reduction is blurry because freed time gets absorbed by other high-value work. The practical solution he proposes: combine headcount and compute budgets into one pool and let trusted operators allocate.

"If we really believe that AI is making our employees 10 percent or 20 percent or 30 percent more efficient, then next year we should just not increase headcount. Or we should increase it by 2 percent instead of 10 percent. Or we should decrease it by 5 percent and say you all should be getting more done with less." 00:37:41

Price Is the Single Biggest Barrier to Uber's Next 300 Million Users

MacDonald says the honest answer — which his IR team won't love — is that price is the primary inhibitor to scaling from 200 million to 500 million monthly users. The mass of global transportation happens at price points far below UberX, making affordability the structural ceiling on TAM expansion, not geography or product.

"What I mean by price is when you think about the businesses we operate, primarily mobility and delivery, the vast majority of the transactions... happen at a price point that is like way lower than our core products. Like taking an UberX to and from work every day in New York City for like 35 bucks a direction, that's still a luxury product." 00:16:29

The Innovator's Dilemma Is Real and Actively Fought at Uber

MacDonald describes how a $225 billion gross bookings business structurally consumes organizational attention, capital, and engineering capacity that would otherwise go to new ventures. Uber's "Growth Bets" program — dedicated teams of 100-150 people carved out of larger business units — is their structural answer. He's candid that startups building from zero still often outpace internal incubators because they lack the fat of incumbency.

"The thing you've already built is so big that it just swallows up your organizational capacity to do anything else. And even if you're able to stand up other businesses, it's impossible for those businesses to get the resourcing, attention, distribution, marketing dollars, engineering capacity, whatever it is, it just gets swallowed up by the whole." 00:12:54

The Disaggregation of UI (Agent-Led Booking) Is a Strategic Threat Uber Is Actively Resisting

MacDonald reveals that Uber has historically refused aggregator API requests — services that would show real-time Uber pricing alongside Lyft and others — because he wants Uber to remain the front-end consumer experience. He's now navigating a more nuanced version of this with AI agents, choosing selective participation with frontier labs rather than blanket openness.

"I've been against that. I mean, I want to be the front end. I want people to start at the Uber app for the Uber experience. And I think today we win that first look with 200 million consumers growing every month." 00:46:30

The Delivery Hero Acquisition Is About Geographic Compounding, Not Just Revenue

MacDonald frames the Delivery Hero deal not as a revenue grab but as a one-shot solution to the slow burn of country-by-country delivery market building, combined with the acquisition of entrenched local consumer brands in markets like Argentina, Korea, and the Middle East that would be extremely expensive to displace organically.

"Delivery Hero, I think, presented a unique opportunity to, in one fell swoop, expand our geographic footprint... Delivery Hero has built a lot of local brands that are really strong. Whether that's true or not, they have consumer mindshare. And they've built Argentina, Korea, the Middle East." 00:51:41

2. Contrarian Perspectives

Distribution Beats Superior AV Technology — Even Waymo or Tesla

Most observers assume the company that wins the AV technology race wins the market. MacDonald argues the opposite: distribution is the dominant force, and even the most technologically superior AV fleet will need Uber's network because expensive hardware requires maximum utilization. The analogy to McDonald's willingly partnering with delivery marketplaces despite having dominant 1P channels is the crux of the argument.

"Whether Waymo or Tesla ends up being the bigger threat, I don't know. I think ultimately it's in both of their interests to put their vehicles on our network... I think everyone will work with us because ultimately we have distribution. And ultimately they have expensive fixed assets that need utilization." 00:26:42

The Majority of Uber Trips Will Never Be Autonomous — for Decades

The entire AV narrative focuses on US cities. MacDonald points out that Uber's two largest volume markets by trips are India ($2.53 average fare) and Brazil ($3.54 average fare), where the economics of autonomy cannot possibly compete with cheap human labor for a very long time. This is never discussed in AV bull theses.

"In India, it's $2.53 or something like that. It's going to be decades until the cost of autonomy compresses to the point where it competes with that cost of human labor. And those markets make up the majority of our trips." 00:24:21

AI Evangelists and AI Skeptics Are Both Wrong — Nuance Is the Only Honest Position

At a moment when the dominant narratives are either "AI is transforming everything now" or "show me the ROI," MacDonald takes a third path: the efficiency gains are clearly real, the ROI just cannot be cleanly isolated in large organizations because time savings get absorbed into higher-value work rather than headcount reduction. Neither the skeptics nor the fundamentalists are right.

"I think AI skeptics were sort of like, see, the Uber COO is saying there's no return on AI, which is obviously not what I was saying. On the other side, there was sort of this like if you were like a fundamentalist AI evangelist, you were saying this guy has no idea what he's talking about. They're obviously doing it wrong because AI is God." 00:34:46

The Chat Interface Is NOT the Right Front-End for Managed Transactions

Against the prevailing assumption that AI agents will mediate all consumer commerce, MacDonald (echoing Brian Chesky) argues that managed, experiential transactions — ride-hailing, hospitality — don't reduce well to plain-language chat queries. The experience that goes wrong is what exposes the fragility of the agent-mediated model, and that problem is nowhere near solved.

"I think he was right. Like, some experiences are more visual. Some experiences are more managed... It is a managed transaction. And so that sort of worst fear hasn't played out yet." 00:48:16

Exiting China Was Strategically Correct — Despite the FOMO

The easy narrative is that Uber made a mistake exiting China. MacDonald argues that given the geopolitical reality, a US company was never going to be the dominant mobility platform in China regardless of capital deployed, the business was burning $52 million per week on subsidies alone, and the outcome — a stake in Didi — was better than almost any Western tech company achieved in China.

"I don't think it was plausible that we were ultimately going to be the market winner. I mean, even for geopolitical reasons alone, like the notion that a US tech company would ultimately be the largest mobility service in China. I just don't think it's something that was ever plausible." 00:31:40

3. Companies Identified

Uber Global mobility and delivery platform. Central subject of the entire conversation. Key stats mentioned: 300 million trips per week, 200 million monthly consumers, $52 billion revenues in full year 2025, approaching $250 billion in gross bookings, operating across 75 countries in mobility.

"We're doing like 300 million trips a week... We're not far off from being a $250 billion company." 00:00:00 / 00:12:06

Waymo Alphabet's autonomous vehicle subsidiary. Identified as one of the most likely winners in the AV race and a potential future partner/threat to Uber.

"Do I think Waymo and Tesla will ultimately be winners? Yes, I do. I don't know who's going to bet against either of those, but I think there will be more winners." 00:25:21

Tesla Identified alongside Waymo as a likely AV winner. MacDonald declines to rank the two as relative threats.

"Who do you think is a bigger threat, Waymo or Tesla?... I think there's going to be more than two winners." 00:24:51

DoorDash Uber Eats' primary US competitor and current market leader in US food delivery. Praised as an excellent operator.

"DoorDash is an excellent company. I think Tony's a tremendous entrepreneur and founder. They operate really well. They move quickly. They're aggressive. They take risk. They're well capitalized." 00:56:28

Delivery Hero European food delivery conglomerate being acquired by Uber. Valued for geographic footprint (Argentina, Korea, Middle East) and entrenched local brands.

"Delivery Hero has built a lot of local brands that are really strong... And they've built Argentina, Korea, the Middle East. Like, these are leading brands that consumers identify with, have high household awareness." 00:52:11

Didi Chinese ride-hailing giant. Uber merged its China operations with Didi in 2016. Discussed in context of the $52M/week subsidy war and the ultimately favorable exit terms Uber received.

"We were burning 52 million a week in China just on price subsidies." 00:00:00

Revolut European fintech super-app. Referenced as the world's best operator of parallel product experiments.

"Nick Staronsky from Revolut. I've interviewed a thousand founders. He's the single best founder I've ever interviewed. And it's because he runs 26 product experiments at once." 00:14:11

Anthropic Frontier AI lab. Referenced in context of AI model selection, cost management, and the risk of frontier labs competing with enterprise clients.

"You do not need the latest and greatest model from Anthropic or OpenAI to ask like, you know, tell me who the president was in 1945." 00:40:25

OpenAI Frontier AI lab. MacDonald personally prefers OpenAI's voice engine above all other AI tools. Also discussed in context of agent-mediated consumer transactions.

"For me, I use both. I use voice so much. Like, it's my single most used AI feature by a mile. Probably 50x anything else... And I find OpenAI's voice engine incredible." 00:58:41

Fireworks AI AI infrastructure provider. Mentioned as an external partner Uber uses for intelligent model routing.

"Do you work with providers like Fireworks to enable efficient routing? Yes. So we work with external providers." 00:40:00

Postmates US food delivery company acquired by Uber. MacDonald notes it had strong brand, geographic pockets, and followership despite being capital-constrained at time of acquisition.

"Postmates has a strong brand, a strong followership, and some strong geographic pockets. And I think we've been able to build on those." 00:57:56

Glovo European food delivery company (Delivery Hero subsidiary). Oscar Wilde mentioned as its leader.

"I know Nicholas really well. Interviewed him. Really like him. Brilliant guy. Know Oscar Wilde from Glovo." 00:50:56

Amazon Referenced as the gold standard for membership programs (Amazon Prime) that Uber One is benchmarked against.

"We're not at like Amazon Prime or Costco levels, but we're getting to within spitting distance." 00:07:10

Costco Referenced alongside Amazon Prime as a membership program benchmark.

"We're not at like Amazon Prime or Costco levels, but we're getting to within spitting distance." 00:07:10

SoftBank Referenced as a major capital provider during the free-money era of ride-sharing competition, investing in multiple competing ride-share platforms simultaneously.

"You had players like SoftBank investing in the market as well. And you remember those days, like that was free money era." 00:29:29

Airbnb Referenced in context of the disaggregation debate — whether chat/agent interfaces are the right front end for experiential, managed transactions.

"Don't get me wrong. Airbnb is an amazing business. But if you remove the transactional booking travel, well, then it just becomes experiential booking travel." 00:49:07

McDonald's Used as the primary analogy for why AV OEMs (like Waymo/Tesla) will ultimately partner with Uber despite having 1P channels — same reason McDonald's works with delivery platforms despite owning its stores.

"McDonald's and Starbucks also are strong leaders in their individual verticals... They also ultimately work with the marketplaces. And we're able to come to a good economic agreement that works for both sides." 00:25:48

Starbucks Referenced alongside McDonald's in the delivery marketplace partnership analogy.

"McDonald's and Starbucks also are strong leaders in their individual verticals." 00:25:48

Lyft Referenced as a comparative pricing option in the context of agent-mediated ride comparison queries.

"Compare the prices of Uber, Lyft, and Waymo and get me the cheapest one." 00:47:17

Palantir Alex Karp (CEO) referenced for his public comments on AI ROI at large enterprises and the risk of frontier labs cannibalizing enterprise clients.

"Alex Karp came on CNBC or CNN and said like, no, the ROI question is still there to validate what you said." 00:36:53

D-Matrix AI semiconductor company. Sid Shait (co-founder/CEO) mentioned as a JP Morgan banking client in the sponsor segment.

WeChat Referenced as the platform Uber was blocked from in China — described as equivalent to being denied access to email or phone numbers in the US market.

"We were not able to operate on the WeChat platform. Like trying to compete in China and not having access to WeChat, you know, it's like trying to compete in the US without like email or a phone number." 00:31:14

4. People Identified

Andrew MacDonald President and COO of Uber, longest-tenured active employee (14+ years). Runs all business initiatives across mobility and delivery. Currently directly operating the delivery business in addition to his day role.

"No one's been at the company longer than me at this point." 00:00:00 "I literally have had to schedule an evening shift because there's just no way to fit my operating cadence in." 00:54:11

Dara Khosrowshahi CEO of Uber. Described as a low-ego, high-heart leader whose superpower is creating followership through transparency of reasoning.

"Management comes from an org chart, leadership comes from the heart." 01:00:37 "He will not ask you to do anything he wouldn't do himself. He's the first one over the fence." 01:00:59

Travis Kalanick Uber co-founder and former CEO. Identified as a visionary creative problem-solver who saw the AV future earlier than almost anyone and started Uber's autonomous efforts circa 2012-2014.

"He's a problem solver. Like, he will define what he is and what he looks for in others as creative problem solving. The ability for him to walk into any meeting on any topic, ask a few pointed questions, float a few ideas, and in 15 minutes sort of change the minds... of the people in the room who have spent, like, weeks as experts on this topic is amazing." 00:59:13

Praveen (Uber CTO) Uber's Chief Technology Officer. Made headlines by stating Uber had burned through its AI budget in the first few months of the year; MacDonald clarifies this was a comment on usage unpredictability, not runaway waste.

"Praveen was speaking at an event and generated this headline by saying we were through our AI budget in the first few months of the year. Praveen's our CTO." 00:33:49

Tony Xu Co-founder and CEO of DoorDash. Praised as a tremendous entrepreneur who runs an aggressive, well-capitalized, fast-moving operation.

"I think Tony's a tremendous entrepreneur and founder. They operate really well. They move quickly. They're aggressive. They take risk. They're well capitalized." 00:56:28

Nik Storonsky Co-founder and CEO of Revolut. Named by Harry Stebbings as the single best founder he has interviewed across 1,000 founders. Praised specifically for running 26 parallel product experiments simultaneously with disciplined weekly check-ins and milestone-based funding.

"He runs 26 product experiments at once. He gives them $2 million, tells them to run for a year. Every single week, he checks in for 20 minutes with each of the leaders. And then he determines whether to fund their next round or not." 00:14:11

Alex Karp CEO of Palantir. Referenced for publicly raising the AI ROI question and warning about frontier labs potentially cannibalizing enterprise data.

"Alex Karp came on CNBC or CNN and said like, no, the ROI question is still there." 00:36:53

Brian Chesky Co-founder and CEO of Airbnb. Referenced approvingly for his contrarian insight that chat interfaces may not be the right front-end for managed, experiential transactions — a point initially mocked but which MacDonald validates.

"Brian Chesky got kind of roasted, but I thought it was an insightful point when he said, it's not clear to him that, like, the right interface for hotel booking is a chat interface." 00:48:16

Rachel Whetstone Former Uber SVP of Communications and Policy (joined ~2015-2016). MacDonald credits her with the single best piece of career advice he received: "always say yes" to new opportunities.

"She sent the speech of a commencement address she gave to the whole company in her first week. And in that, the sort of central thesis was always say yes. Just jump at the next adventure." 01:02:54

Oscar Hartmann (Oscar Wilde) Referenced as the leader of Glovo (Delivery Hero subsidiary). Harry Stebbings vouches for him personally.

"Know Oscar Wilde from Glovo. Really like him." 00:50:56

Nicholas (Delivery Hero CEO) Referenced as the CEO of Delivery Hero whom Harry Stebbings knows personally and vouches for.

"I know Nicholas really well. Interviewed him. Really like him. Brilliant guy." 00:50:56

Jeff Bezos Referenced for the quote: if you want to be right most of the time, you have to change your mind a lot — used to frame MacDonald's philosophy on intellectual humility.

"I think it's a Bezos quote, which is if you want to be right most of the time, you got to change your mind a lot or something to that effect." 00:05:43

5. Operating Insights

Combine Headcount and Compute Budgets Into One Unified Pool

MacDonald's most actionable AI budgeting insight: rather than setting separate headcount and compute budgets that create internal political battles and make AI spend look like "overruns," merge them into a single pool and let trusted operators allocate across both. This naturally allows the business to trade human time for compute when the ROI is there.

"I think it would be totally reasonable for Dara to say your headcount budget is X, our compute budget is Y. Just add X and Y together and then spend it as you see fit. And so if you want to spend relatively more money on compute, on inference, on whatever, because you believe that's the highest ROI, do that. But it means you have less for heads." 00:38:42

Pair AI Engineers Directly With Business Operators — Process by Process

Rather than deploying AI top-down through IT or as a generic productivity tool, Uber stands up a dedicated pod of 30 top AI engineers who are physically paired with business operators to redesign individual processes from scratch. The outcomes: capital allocation workflows cut from 15 hours to 2, finance forecasting from 8 hours to 2, marketing QA from 2 weeks to 2 days.

"We have stood up a pod of 30 of our best AI engineers that are partnered with business people or partnered with folks in the G&A functions to go in and go process by process and start sort of ground up with AI. How do you improve that process?" 00:35:34

Force New Internal Ventures to "Sing for Their Supper"

MacDonald envies Revolut's model of giving new product experiments a fixed budget ($2M), a fixed runway (one year), weekly 20-minute check-ins with the CEO, and milestone-based re-funding decisions. The key mechanism: removing the safety net of a large corporate P&L forces the team to operate with startup discipline — faster cadence, leaner resource consumption, less drift.

"I love the cadence of that, by the way. Like operating on weeks, not months or quarters is how a new business should run. I think also having to sing for your supper, like come back and ask for money... you just are slower and you're constantly actually chasing the people who are doing it from first principles." 00:14:31

Teach Your Reasoning, Not Just Your Decisions — It Creates Organizational Multipliers

MacDonald's most important leadership operating principle, drawn from observing Travis Kalanick: the compounding value of a leader comes not from making good decisions, but from explaining why the decision was made — so the organization can replicate the reasoning without the leader being present.

"He would not only give an answer to a question, but he would explain his thinking on why that was the answer... it creates mini versions of yourself. And so I think if you can do that across your organization, where you tell people how you got to an answer, you're amplifying the power of the organization." 01:00:11

Use AI-Driven Model Routing and Cost Visibility Dashboards to Prevent Compute Sprawl

Uber actively routes queries to appropriately-powered models (using external providers like Fireworks AI and internal tooling), publishes internal AI usage leaderboards alongside cost leaderboards, and assigns different model tiers to different roles. The discipline: not every task warrants a frontier model, and making costs visible changes user behavior without requiring mandates.

"The idea that you would like not only publish an AI usage leaderboard, but also an associated cost leaderboard, just so people were aware. The idea that you might choose different models for different tasks from the outset or give different levels of employees different models for different tasks from the outset." 00:39:30

6. Overlooked Insights

The "Employees on Both Payrolls" Phenomenon in Chinese Ride-Hailing — A Preview of AI Agent Competition

MacDonald briefly mentions, almost as a throwaway anecdote, that when Didi merged with Kuaidi (the third major Chinese ride-hailing player), they discovered approximately 200 employees who were simultaneously on both companies' payrolls — effectively double agents. He treats this as a colorful China story, but the structural implication is far larger: in platform markets where the product is nearly identical and the moat is thin, competitive intelligence and talent arbitrage become the actual battleground. This dynamic is beginning to re-emerge in AI, where engineers routinely consult for multiple frontier labs and the boundaries between companies are porous in ways that Western regulatory frameworks haven't caught up to.

"They realized that like of like the 2,000 employees here and the 2,000 employees here, there were like 200 employees that were on both payrolls. And so you sort of had this dynamic where you sort of like realize like, oh, OK, like, you know, this is like real deep competitive gnarly. Like you have employees that are wearing both hats, which was crazy to me to hear because that notion just in like competing in the US, it's not something that in a million years I could see happening. Feels like a frontier AI." 00:30:18

The Membership Program Has a Structural Cost Problem That Limits How Good It Can Ever Get

MacDonald briefly surfaces a structural constraint on Uber One that gets no further discussion: unlike hotel loyalty programs or airline miles — where excess fixed-capacity inventory can be given away at near-zero marginal cost — Uber's purely variable cost model means every benefit given to a member has real cash cost because a driver must be paid for every ride. This means Uber One will always have a harder time generating the high-perceived-value, low-actual-cost benefits that make programs like Marriott Bonvoy or Delta SkyMiles feel magical. This is a non-obvious structural ceiling on Uber's ability to close the gap with Amazon Prime or Costco, and it's a lens that applies broadly to any platform evaluating loyalty programs — variable-cost platforms are intrinsically disadvantaged relative to fixed-cost platforms in loyalty economics.

"With our business, that's tough because I don't have a lot of free to give away on the platform, right? If I want to give you a ride because you're a loyal member, either through a membership program or a rewards program, I still got to pay the driver to provide that ride. It's not like a hotel where you might have excess inventory and so your marginal cost of giving away a room night is pretty low. The beauty of our model is we're primarily a variable cost model... But it also means that we just don't have a fixed capacity to give away." 00:11:13