Startup School

Dmitri Dolgov on why a working demo is 1% of the product

Dmitri Dolgov· Co-CEO at Waymo
·~49 min·English·Y Combinator
RoboticsAI InfrastructureAI SafetyMultimodal
TL;DR

Waymo's co-CEO turns two decades of self-driving into a seven-lesson playbook for physical AI, where a working demo is 1% of the work, every added nine of reliability costs roughly 10x, and the real moat is the evals, simulation, and hundreds of millions of autonomous miles backed by publicly audited safety proof.

01Physical vs Digital AI

Human Lives, Not Tokens

Building AI for the physical world means clearing four gaps that set it apart from digital AI — the cost of an error, the latency budget, the missing data, and the day-one validation bar — because you are moving atoms, not bits.

the cost of a mistake can be measured in human lives, not tokens. There's simply not an undo and a retry button.

Dmitri Dolgov, YC Startup School
Key Insight
The four gaps invert the digital playbook: 'ship, then find edge cases on live users' becomes 'prove it, then ship.' That one inversion — validation before deployment instead of after — is what makes physical AI a different kind of problem, not merely a harder version of the same one.

02Demo vs Product

A Working Demo Is 1% of the Job

Waymo's demo drove itself in 18 months, but the shippable product took 15 more years, because reliability climbs an exponential ladder where every added nine of performance costs roughly ten times the last.

at scale, the long tail is the problem space. It's your entire problem statement. When you drive millions of miles per week, a rare event that might happen once in a million miles, that just becomes your daily reality.

Dmitri Dolgov, YC Startup School
Key Insight
The trap is not that the tail is hard, it is that the tail is invisible at demo scale. A once-in-a-million-miles event is unlikely to appear in a 100,000-mile demo, so founders extrapolate a solved problem from a curve that has not yet met its hardest inputs — which is why Dolgov's rule is to count your nines before you count your demo views.

03Architecture

Your Nines Pick Your Architecture

Most technologies ride a steep early curve that tends to flatten, so the recurring trap is picking the tech with the fastest ramp only to hit a plateau below the reliability your product actually needs.

common failure mode is picking the tech that gives you the fastest early ramp, riding that steep curve, feeling like you're winning

Dmitri Dolgov, YC Startup School
Key Insight
Dolgov leaves an escape hatch: riding the steep curve on purpose, to prototype or to learn, is fine as long as you are honest that you are building a demo and not the product. The deeper discipline is refusing to over-fit to the present at all — the same reason he warns against betting the company on today's hardware prices, now six sensor generations and several cost cuts old.

04Riding the Waves

Add Capability, Subtract Complexity

Waymo rebuilt its driver around every AI wave — CNNs, then transformers, now VLMs and world models — but the rare muscle is not adopting new tech, it is forcing each wave to also simplify and unify the stack instead of fragmenting it.

set your launch bar to demand both breakthrough performance and at the same time radical simplification and unification.

Dmitri Dolgov, YC Startup School
Key Insight
The counterintuitive move is treating simplification as a launch requirement, not a cleanup task. Most teams grade a new model on capability alone and quietly accrue complexity; Dolgov's bar rejects any win that fragments the stack, because that compounding complexity is exactly what stops you from catching the next wave.

05The Bitter Lesson

Structure That Channels Scale, Not Fights It

Sutton's bitter lesson says methods that scale with compute and data beat handcrafted knowledge, and Dolgov's refinement is that structure which fights scale loses while structure that channels it wins.

bet on a system that's maximally learned and minimally constrained and leverage structure intentionally to boost performance and scaling laws both in training and in evaluation.

Dmitri Dolgov, YC Startup School
Key Insight
This dissolves the 'end-to-end versus hand-engineered' debate into a better question: is your structure load-bearing scaffolding that scale climbs, or a cage that scale outgrows? Waymo's structure-augmented end-to-end keeps the learned core but materializes a representation it can validate in real time.

06Simulation

Your Simulator Is as Hard as Your Agent

Trying a risky action and watching its consequence is rarely something you can do safely on real roads, so closed-loop simulation is essential for physical AI — and a good simulator is itself a giant world model, as hard to build as the driver it trains.

the problem of building a good realistic simulator is just as hard as building the agent itself.

Dmitri Dolgov, YC Startup School
Key Insight
The tell is that Waymo can now simulate scenarios it has never encountered — a plane landing on the freeway, an elephant in the intersection — and train against them in closed loop. When your simulator can manufacture the long tail on demand, you stop waiting for rare events and start rehearsing them.

07The Ecosystem

You Don't Build One AI, You Build Three

Shipping a physical agent takes three AIs sharing one foundation — the agent that acts, the simulator it practices in, and the critic that judges it — wired into a flywheel where real-world deployment keeps making all three smarter.

once you have these three pillars you can create an incredibly powerful flywheel to accelerate your progress.

Dmitri Dolgov, YC Startup School
Key Insight
The critic is the piece most teams skip. An agent and a simulator alone just spin faster; without a rigorous judge defining better, the flywheel accelerates in no particular direction. The scarce asset is not the model or the sim, it is a trustworthy definition of improvement.

08The Moat

Eval Is the Moat, Not the Model

Model architectures are well known and spread fast, so the durable advantage is not the model but the evals, metrics, and hundreds of millions of autonomous miles of publicly audited proof that are far harder to replicate — turning trust into the real moat.

If you can't quantitatively define what good enough means, you're not really building a product. You're just iterating on your demo.

Dmitri Dolgov, YC Startup School
Key Insight
The safety math is the payoff for the entire playbook: roughly 17 times fewer serious-injury crashes than human drivers, which Dolgov frames as a serious injury prevented every eight days. Eval is not a gate you clear before shipping, it is the compounding asset that becomes the moat, because trust is earned in public one audited mile at a time.