Startup School Paris

Eric Landau on why physical AI is won at the data layer

Eric Landau· Co-founder and Co-CEO of Encord at Encord
·~20 min·English·Y Combinator
RoboticsAI CompanyTrainingBusiness Strategy
TL;DR

Encord's Eric Landau explains why physical AI — roughly 80% of the economy — will be won not by cleverer models but by whoever can collect, curate, and evaluate multimodal data at petabyte scale.

01Core Mental Model

The bitter lesson, lived as a career

Landau's path from particle physics to quant to Encord traces the same shift the whole field took — away from hand-crafted expertise and toward scaling raw data and compute.

The common thread is almost the kind of trajectory of AI development in the time that I've had my career.

— Eric Landau, Startup School Paris
Key Insight
His resume looks eclectic — physics, quant, startup — but it's one straight line: each field moved from encoding human expertise to letting scale do the work. Encord is a bet that the same arc now runs through robotics, where the scarce ingredient isn't cleverness but curated data.
02Origin Story

He quit at the top of the market

He walked away from a decade of peak quant pay during the most lucrative month his desk ever had — not for upside, but to escape the feeling that his work didn't matter.

My desk made more money in that month than I think they made in the previous 3 years combined.

— Eric Landau
Key Insight
The tell is what replaced the money — not upside, but the disappearance of an existential question he used to hit the pillow with every night. Founders who optimize for that internal payoff tend to survive the desert years that break people chasing external validation.
03Product-Market Fit

The market came down to the product

Encord never hit a single product-market-fit moment; the product compounded quietly for years until the market shifted and came down to meet it.

Sometimes it's the market comes to the product, you know, the world changes and suddenly the thing that people didn't care about now care about.

— Eric Landau
Key Insight
"The market comes to you" sounds like luck, but it only pays off if the product is already good enough to catch the wave. Encord spent the desert years compounding quietly, so that when ChatGPT shifted the conversation there was something ready to buy — the fit was only ever obvious looking back.
04Building the Team

It took three sales teams

Two sales teams failed before the third worked, and the real lesson wasn't a hiring formula — it was learning to act on his gut faster.

By the third sales team, it started to work.

— Eric Landau
Key Insight
The gap he keeps naming is between knowing and acting — YC's advice was almost always right, yet they still made the wrong hire and kept the person nine months. The scarce skill in hiring isn't judgment; it's the willingness to act on judgment you already have, faster.
05The Physical AI Bet

80% of the economy is physical

Physical AI isn't a niche — moving and manipulating things in the real world is about 80% of economic activity, and that's the market Encord is now built around.

80% of economic activity is manipulating or moving things in the real world.

— Eric Landau
Key Insight
The 80% figure reframes physical AI from a niche robotics story into the largest addressable market in tech. If models are commoditizing, the durable moat shifts to whoever can feed them the physical-world data they can't generate on their own.
06Inside the Product

A million needles in a billion haystacks

Encord's real product isn't collecting data but curating it — pulling the small slice worth training on out of petabytes, then annotating and evaluating it.

Finding a million needles in a billion haystacks.

— Eric Landau
Key Insight
The counterintuitive part is that collecting data is the easy, low-value step — customers are drowning in it. The defensible business is subtraction: deciding what tiny fraction is worth keeping, labeling, and testing against, at a scale most teams can't handle.
07Where to Build

Be where the important people are

Landau's rule is to locate yourself where your top priority lives — customers first, then talent, then investors — even if it splits the founders across two cities.

You should just be where the important people for your company are. So, for us it's customers and then talent and then investors in that order.

— Eric Landau
Key Insight
The framework quietly rejects the reflexive "just move to SF" orthodoxy. By ranking customers over talent over investors, Landau turns location into a per-company decision — and is willing to split his own founding team across an ocean to honor it.
08The Founder Mindset

Just ride the roller coaster

The founder skill Landau values most is accepting that startups swing between extreme highs and lows, and choosing to ride the volatility instead of fighting it.

You should just ride the roller coaster. Like people try to mitigate it, but just ride it.

— Eric Landau
Key Insight
The reframe is subtle: he isn't claiming the volatility shrinks, only that his reaction to it can be trained. It's the bitter lesson turned inward — stop trying to engineer the ride, and just scale your tolerance for it until the swings feel survivable.