Andy Fang | No Priors Summary

1. Key Themes

Agentic Commerce is Already Driving Measurable Behavior Change

DoorDash's "Ask DoorDash" natural language interface is producing results that traditional UX could never achieve. Fifty percent of users are ordering from new restaurants, and grocery basket sizes are 40% larger — both historically stubborn metrics.

"50% of trajectories of people using Ask DoorDash for restaurants, 50% of those trajectories are people ordering from places they've never ordered from before, which is huge because that's one of the hardest metrics historically for DoorDash for us to move." — Andy Fang 00:01:56

DoorDash Has Been a Robotics Company Since 2018 — Long Before It Was Fashionable

Most people think of DoorDash as a food delivery app. The founders have been systematically investing in autonomy for nearly eight years, quietly building an infrastructure that is only now becoming visible.

"We've actually been looking into robotics autonomy probably much longer than people thought, like since 2018, actually. Back when it wasn't obvious autonomy and robotics was going to be a thing." — Stanley Tang 00:07:26

The "Build the Technology First" Approach to Robotics Is Fundamentally Broken

The dominant paradigm in the autonomy/robotics startup world — build a general capability, then find use cases — is the wrong approach. DoorDash's experience with every major player in the space confirmed this.

"It always felt like these companies weren't really focused on a use case. It always felt like they kind of build the technology first and then retroactively try to go find a problem to fit into... when we went through YC, we're always taught to build something people want... but then when it comes to hardware and hard tech and AI and robotics, people just kind of do the opposite." — Stanley Tang 00:11:46

DoorDash Dot Occupies a Purposefully Engineered Gap in the Autonomy Market

Sidewalk robots are too slow (2–3 mph) for DoorDash's average 3–5 mile, 15-minute delivery window. Robo-taxis are designed for people, weigh 4,000 pounds, and can't solve the first-and-last-100-feet problem. Dot — 300 pounds, 20+ mph, navigating both roads and bike lanes — was built specifically for suburban last-mile delivery because nobody else was building it.

"If you were to start first principle... what would that look like? And we looked around. Turns out no one's really doing that. It's not a sidewalk robot. It's not a robo-taxi. We felt like it was probably something in between... an autonomous motorcycle or a scooter or bike profile vehicle... 300 pounds... 20, 25 miles per hour." — Stanley Tang 00:14:56

Drop-Off Location Data Is a Genuine, Non-Replicable Moat

Where food is actually delivered — which driveway, which gate, which front door — is data that exists nowhere else in the world. This solves a hard robotics problem that neither Google Maps nor any autonomy startup can address.

"Where did the human dasher drop it off historically? And that is... that first and last hundred feet problem. Like you don't, that data doesn't exist anywhere else. It doesn't exist in Google Maps. It only exists at DoorDash." — Stanley Tang 00:32:25

Scaling Autonomy in the Real World Is Mostly NOT an Autonomy Problem

The edge cases that kill scale are operational, not algorithmic: dirt on camera sensors, leaves creating torque differentials, boot-up scripts crashing across 500 robots, regenerative braking causing electrical shocks in extreme stops. These are invisible until you're operating daily at scale.

"When you have to do something seven days a week, 10 hours a day at scale, things start breaking... there's some leaves on the ground... half your wheels, the right two wheels are on the leaves. The left two wheels are still on the asphalt. All of a sudden the torque you have to send to the wheels is very different." — Stanley Tang 00:28:35

The Hardware Problem Has Quietly Become the Binding Constraint in Robotics

Five years ago autonomy was the hard problem. Today, as autonomy matures, manufacturing, supply chain, and component reliability are emerging as the actual blockers to scale — a shift that the industry hasn't fully digested.

"It's kind of funny. It's like when we first started like five years ago, like everyone thought hardware was a commodity and now it's starting to look like hardware is starting to become the bottleneck... we hand built the first hundred robots ourselves and which is not an issue, but then okay, the next thousand or 10,000, well, we're going to have to now start to think of things like supply chain and component reliability." — Stanley Tang 00:37:34

AI Spend at Enterprise Scale Needs Rigorous ROI Benchmarking

DoorDash's internal AI spend went up 20x from January to June 2025. They built a public benchmark (DashBench) to measure coding task performance and are now working to benchmark non-technical functions — a model other large enterprises should replicate.

"I was looking at it a week ago. I think our spend in June went up like 20X versus what the spend was in January... clearly this has got to get some sort of return." — Andy Fang 00:41:25

More Autonomy Will Mean More Dashers, Not Fewer

Counterintuitively, robotics and AI will expand the total delivery market so dramatically — through lower prices unlocking latent demand — that human dashers will actually grow in absolute number over the next decade.

"My prediction actually is in a world where robotics, drones, AI is everywhere, my guess is that in 10 years time, we're actually going to have more dashers doing deliveries, not less... the business growing 25% year over year, like fast forward 10 years time, if we want a 5X from here, 10X from here, well, where's the supply going to come from?" — Stanley Tang 00:45:01

2. Contrarian Perspectives

The Incumbent Data Advantage Is Mostly Illusory — Except When It's Hyper-Specific to the Task

The conventional wisdom that incumbents win because they have "data" is largely wrong. The specific data that matters is deeply task-specific — e.g., exact drop-off coordinates per address — not general customer databases.

"Early on when people were talking about what's going to happen with AI and incumbents and startups, there were a lot of people with a very surface level view of what the incumbent data advantage was... they'd be like, ah, like we have the customer records and database. And I was like, that actually has like very little to do with the thing we're trying to accomplish with an agent." — Sarah Guo 00:32:42

Voice Was the Wrong Modality — But the Underlying Bet on Conversational Commerce Was Right

DoorDash was early on voice and it failed. The lesson is not that conversational interfaces don't work, but that the specific modality matters enormously, and even correct underlying theses can be invalidated by wrong implementation choices.

"It started a couple years ago, honestly... we were bullish on voice as the modality. That ended up not being the thing, but maybe it will in the future, but just that didn't really land." — Andy Fang 00:00:54

There Is Already More Agent Traffic on the Web Than Human Traffic

This single data point — stated in passing — fundamentally reframes what a "product" needs to be designed for. Most product teams are still optimizing for human interfaces.

"One stat that I always like to think about nowadays is just like there's more agent traffic on the web than human traffic... how do we have a DoorDash type experience that plays into that trend?" — Andy Fang 00:04:40

Frontier AI Models Underperform on Real Enterprise Data Versus Scrubbed Demo Data

When real DoorDash enterprise data (unscrubbed) is used, model performance is materially worse than on cleaned benchmark data. This gap between benchmark performance and enterprise reality is a major underappreciated problem.

"When we actually have it with the enterprise data and all the real stuff, it's not performing as well... if you like dumb down the problem, maybe the models do well, but like for some reason, when we actually have it with the enterprise data and all the real stuff, it's not performing as well." — Andy Fang 00:44:19

Hiring for Robotics: Ship in the Real World Beats Pure Research Every Time

Contrary to the prestige of PhD lab work in the robotics talent market, the pitch that actually works is the promise of shipping something real — and it turns out there is deep, frustrated demand for this among experienced autonomy researchers.

"My pitch is really simple: do you want to go work on prototypes and demos and be at a PhD lab? Or do you want to work on something where you can actually ship something in the real world?... especially in the autonomy world for the past 10 years, people were just fed up, just working on something for 10 years and then never actually getting to a point where they actually saw their products being used in the real world." — Stanley Tang 00:26:46

3. Companies Identified

DoorDash The host company. Food and grocery delivery platform operating in 40–50+ countries, 9 million dashers, 3 billion+ deliveries per year, 40 million monthly consumers, growing 25% year over year.

"We do over three billion deliveries a year. There are no two deliveries that look the same." — Stanley Tang 00:18:14

Waymo Autonomous vehicle / robo-taxi company. Cited as proof that the autonomy inflection point has arrived and as a benchmark for what L4 autonomy looks like at scale.

"You've seen Waymo tap it in... fast forward today, you're seeing the Waymos driving. It's happening." — Stanley Tang 00:08:14

Also (spun out of Rivian) Micro-mobility company founded/chaired by RJ Scaringe (Rivian founder). DoorDash partnered with Also specifically to solve the hardware manufacturing and vehicle scaling challenge for Dot.

"We actually partnered up with this company called Also, which is this micro-mobility company that spun out of Rivian. So RJ is actually the board founder and chairman of the company. Why don't we work with someone who knows how to actually scale vehicles." — Stanley Tang 00:38:41

Metis AI-native company acquired by DoorDash to infuse AI-first thinking into DoorDash's large organization.

"One of the reasons why we were so excited to acquire a company called Metis last year was really to just infuse some of that AI native thinking into the company." — Andy Fang 00:40:01

Sunday Robotics company focused on household tasks (specifically mentioned in the context of home robotics and the challenge of distribution of real-world environments). Sarah Guo is an investor.

"I'm an investor in a company called Sunday... one thing that we deeply believe in this company is you can't imagine the distribution. As soon as you make contact with the physical world, you're like, man, if we're trying to do the dishes, why is a cat in the dishwasher?" — Sarah Guo 00:33:02

Tesla Cited alongside Waymo as evidence of the broader autonomy breakthrough reaching maturity.

"You see kind of now with especially with AI, like Waymo has kind of made that breakthrough. I think Tesla's starting to make that breakthrough." — Stanley Tang 00:35:53

4. People Identified

Andy Fang Co-Founder, DoorDash. Leads AI and product direction including Ask DoorDash, DashBench, and the DoorDash CLI. Deeply focused on agentic commerce and enterprise AI ROI measurement.

"We're seeing people 50% of trajectories of people using Ask DoorDash for restaurants, 50% of those trajectories are people ordering from places they've never ordered from before, which is huge because that's one of the hardest metrics historically for DoorDash for us to move." 00:01:56

Stanley Tang Co-Founder, DoorDash. Drives the autonomy and robotics strategy including Dot, the autonomous delivery platform, and the Also partnership. Deeply operational thinker on physical-world scaling.

"We've actually been looking into robotics autonomy probably much longer than people thought, like since 2018, actually. Back when it wasn't obvious autonomy and robotics was going to be a thing." 00:07:26

RJ Scaringe Founder and Chairman of Also (micro-mobility, spun out of Rivian); founder of Rivian. Brought in as a DoorDash partner specifically for his expertise in scaling physical vehicle manufacturing.

"RJ is actually the board founder and chairman of the company. So if you know, like, why don't we work with someone who knows how to actually scale vehicles." — Stanley Tang 00:38:41

5. Operating Insights

Build Benchmarks Before You Scale AI Spend — Not After

DoorDash watched AI spend increase 20x in six months before building DashBench to measure ROI. The lesson: instrument the value created by AI tools before adoption runs ahead of accountability. They are now extending this to non-technical functions.

"We announced a benchmark called DashBench a couple of weeks ago now that was mainly focused on our ability to figure out how well various models and harnesses performed on coding tasks... okay, like, $30 million of okay." — Andy Fang 00:40:57

Use Open-Weight Models for Commodity Tasks to Recapture Closed-Model Spend

Rather than routing all AI tasks to frontier closed-weight models, DoorDash is actively mapping tasks by intelligence requirement and delegating cheaper tasks to open-weight models — achieving equivalent output at lower cost.

"If there's a way for us to maximize the intelligence, but maybe delegate to open weight models for some of the cheaper tasks we can actually do, we can get the capable level of intelligence but actually pay less than if we were just using these closed weight models." — Andy Fang 00:42:24

The Highest AI Adoption Growth Is in Non-Technical Org — Don't Over-Index on Engineering

DoorDash's fastest growing AI user base is non-engineers: analysts, operators, account managers. Enterprises that deploy AI only in engineering will miss the majority of the productivity opportunity.

"We're actually seeing the highest amount of growth in our organization in terms of seats in the non-technical organizations, because analysts are finding a lot of value in it, our operators, account managers who are trying to figure out, okay, how do I do my QBR with the strategic merchants?" — Andy Fang 00:42:51

When Scaling Any New Modality, Partner for Manufacturing Before You Need It

DoorDash hand-built the first 100 Dot robots themselves. The decision to partner with Also for scaled manufacturing came from recognizing that supply chain and component reliability become binding constraints earlier than most founders expect.

"The problem five years ago was autonomy. Now it's increasingly becoming more about operations, commercialization, hardware manufacturing... we hand built the first hundred robots ourselves... the next thousand or 10,000, we're going to have to now start to think of things like supply chain." — Stanley Tang 00:37:34

Skunkworks First: Test Every Major New Technology Bet With Minimal Resources Before Committing

Every major DoorDash bet — including autonomy — started as a sub-team experiment with minimal headcount. This is how they avoid over-committing to the wrong form factor or approach.

"It was me and half an engineer's time. It was a skunkworks project. It was an experimentation to go: let's go explore. We don't even know what autonomy looks like, how robotics is going to impact our space." — Stanley Tang 00:09:21

6. Overlooked Insights

The DoorDash CLI Is a Silent Launch of a Programmatic Commerce API — and It's Enormous

Mentioned only briefly and almost in passing, the DoorDash CLI is not just a developer tool — it is the infrastructure layer that allows any AI agent to trigger a DoorDash order programmatically. The office pantry camera example is not a cute demo; it is proof that DoorDash is positioning itself as the commerce execution layer for the agentic web. Given that agent traffic already exceeds human web traffic, this could be a bigger revenue surface than the consumer app itself within a decade.

"We've been testing some of that with the recent DoorDash CLI that we launched last week... someone was really excited to use the DoorDash CLI because let me basically streamline my office manager use case for my startup... they just pointed a camera at their pantry shelf, and whenever the shelf was getting empty, like they would fire off the agent to basically restock the shelf." — Andy Fang 00:47:55

DoorDash Is Collecting Physical World Training Data at Scale Through the Dasher Fleet — and Selling It to Humanoid Robotics Companies

Also mentioned extremely briefly: DoorDash launched a product called "Tasks" that uses dashers to collect data points to train world models — i.e., the Dasher fleet is being monetized as a real-world data collection engine for humanoid robotics companies. This is a brand new revenue stream and a strategic wedge into the foundation model supply chain for embodied AI that almost no one is discussing.

"We also launched a product called Tasks a couple months ago where we're having people in the Dasher fleet help basically collect data points to help train some of these world models." — Andy Fang 00:20:05