Eric Landau on Building Encord and the Rise of Physical AI

Starting a company is often described as a roller coaster, with extreme highs and lows. The key, according to Eric Landau, co-founder of Encord, is to embrace the ride and try to have fun with it. This perspective comes from his journey from particle physics to quantitative finance, and finally to building a data layer for physical AI.

From Particle Physics to Quant to AI

Eric Landau's career trajectory reflects the evolution of AI itself. He began in particle physics, where systems and strategies were deeply rooted in the physics principles, requiring meticulous data filtering based on particle movement. This approach emphasized domain expertise and feature engineering. He then spent a decade as a quant in high-frequency trading, applying a similar methodology by carefully analyzing market factors before integrating them into machine learning models.

However, the "bitter lesson" of AI, which suggests that scaling systems with more data and compute is often more effective than intricate feature engineering, has become increasingly relevant. This shift is evident in the current AI landscape, where the focus is on feeding vast amounts of data into large machine learning models.

The Leap from Quant to Founder

Leaving a lucrative career as a quant to start Encord was a significant decision for Landau. He recounts quitting during the COVID-19 pandemic, a period of extreme market volatility where his former desk made an unprecedented amount of money. While he was grappling with Python dependencies on his couch, he questioned his choice.

The motivation for this leap wasn't financial. Landau and his co-founder were deeply convinced that AI represented the technological paradigm shift of their generation, akin to the early days of the internet or computing. Despite the financial sacrifice, starting Encord brought a sense of purpose, replacing existential dread with concrete problems to solve and the rewarding feeling of building something impactful.

The Early Days of Encord and the ChatGPT Effect

Encord was founded in 2021, and its initial years were challenging. Landau describes this period as being "in the desert" for a couple of years, struggling to convince people of the importance of their work. The turning point arrived with the release of ChatGPT. This event significantly shifted the conversation around AI, making people more receptive to the technology and its potential.

For Encord, ChatGPT didn't create an immediate, binary product-market fit. Instead, it accelerated a gradual process. Their product was continuously improving, and after ChatGPT, the market started moving towards them at a faster pace. Landau recalls a moment of realization when a sale closed for a company they didn't even recognize, signifying that their product was gaining traction independently of their direct involvement.

Building a Sales Organization Through Trial and Error

As a physicist and quant, building a sales organization was a new challenge. Landau admits it was a process of trial and error. They brought in sales teams that didn't work out, learning hard lessons along the way. He highlights the accuracy of Y Combinator's advice, even when founders make the mistake of not following it. He recounts a specific instance where they hired someone against YC's explicit advice, only to realize their error months later. This experience taught them the importance of trusting their gut and making difficult decisions faster.

The Evolution to Physical AI

Encord initially focused on computer vision, believing it to be the hardest modality and a strong foundation for their long-term vision of multimodal AI. Their thesis was that AI systems, like humans, would naturally become multimodal, integrating various sensory inputs for better decision-making.

The shift to physical AI was a natural progression. While they started with other use cases, the market for multimodal AI in physical applications, such as robotics, autonomous vehicles, logistics, and manufacturing, began to grow significantly. Landau emphasizes the enormous opportunity in physical AI, noting that 80% of economic activity involves manipulating or moving things in the real world. He even suggests that in 5 to 10 years, robots could outnumber people.

Encord's strategy involves constantly engaging with smart people, including prospects, customers, industry leaders, and other founders, to understand emerging problems and synthesize new opportunities. This outward-looking approach, rather than isolated internal discussions, guides their strategic bets.

Diverse Applications and Data Scale

Encord serves a wide range of customers, including those in healthcare. One of the most interesting applications they've encountered is facial recognition for cows, demonstrating the diverse and sometimes unexpected uses of AI. Humans are poor at recognizing individual cow faces, but algorithms excel at it, aiding in health monitoring and tracking.

In physical AI, autonomous driving is a more mature use case, with some applications already in production. However, the biggest growth area for Encord is in robotics, encompassing humanoids, consumer robotics, and manufacturing.

Encord collects data through various means, including a dedicated facility in the Bay Area where robots perform specific tasks in controlled environments, akin to film sets. They also ingest production data from customers, which includes video streams, sensor data, audio, and language – all multimodal data.

The core of Encord's product involves managing, curating, and annotating this vast amount of data. They describe it as "finding a million needles in a billion haystacks," highlighting the challenge of selecting the most relevant data from an overwhelming volume. They deal with multiple petabytes of multimodal data, exceeding the amount used to train GPT-4.

Global Presence and Competitive Landscape

While Encord started in London, with Landau still based there, his co-founder is in San Francisco. Their philosophy is to be where the important people are: customers, talent, and investors, in that order. Given that most of their customers are in the US, particularly the Bay Area, they've established a significant presence there, including their data collection facility.

Regarding talent, London offers a strong pool of AI and engineering professionals from universities and other companies, providing a comparative advantage. However, the competition for talent is intensifying globally.

Encord operates in a competitive landscape, but Landau views competition positively, as it pushes them to improve. Their differentiator lies in their focus on physical AI and the scalability of their data platform, capable of handling petabytes of data – a foundation they've spent considerable time building.

For small companies just starting in physical AI, Landau advises using open-source tools initially. Encord's product is designed for companies at an inflection point, moving from proof-of-concept to production, or experiencing a significant influx of data, when they require scale and robust data management.

The Roller Coaster of Entrepreneurship

Reflecting on his journey, Landau reiterates the roller coaster analogy for starting a company. His advice to founders is to embrace the highs and lows, understanding that one always follows the other. He believes that with enough experience, one can learn to make the journey enjoyable, even finding excitement in challenges. This perspective, he admits, is a relatively recent realization, born from navigating numerous fluctuations and gaining control over his mental state.