Eric Landau, co-founder of Encord, shares his journey from particle physics and quantitative trading to founding a company focused on building the data layer for Physical AI. He explains how his career path mirrors the evolution of AI—from heavily engineered, domain-specific models to large-scale machine learning driven by vast amounts of data. Despite leaving a lucrative quant job during a highly profitable market period, Eric was motivated by a desire to be part of a technological paradigm shift and to work on meaningful problems rather than for financial gain.
Encord’s early years were challenging, with slow progress and difficulty convincing the market of their vision. The release of ChatGPT marked a turning point, increasing market awareness and accelerating demand for AI solutions. Eric describes product-market fit as a gradual process rather than a sudden event, highlighting a memorable moment when a sale closed with a company they didn’t fully understand, signaling broader adoption. Building a sales team was a trial-and-error process, involving multiple hires and firings before finding the right fit, underscoring the importance of trusting one’s instincts in hiring decisions.
Initially, Encord focused on vision AI, considering it the hardest modality, with a long-term goal of multimodal AI systems that integrate various sensory inputs. Over time, the company pivoted towards Physical AI, which involves robotics, autonomous vehicles, logistics, and manufacturing. Eric emphasizes that while Physical AI is still emerging, it represents a massive opportunity since 80% of economic activity involves manipulating the physical world. He envisions a future where robots outnumber humans, operating autonomously and connected to centralized systems.
Encord collects multimodal data—including video, sensor, audio, and language—from both production environments and a dedicated facility in the Bay Area where robots perform tasks for data gathering. The company manages vast amounts of data, curating, annotating, and evaluating it to train AI models at scale. Despite being based in London, Encord maintains a strong presence in the Bay Area to stay close to customers and talent, balancing geographic advantages between Europe and the US. Competition in the space is intense, but Eric welcomes it as a driver of innovation and differentiation, with Encord’s strength lying in handling petabyte-scale data for Physical AI applications.
Eric concludes with advice for founders, likening the startup journey to a roller coaster with highs and lows that should be embraced rather than avoided. He stresses the importance of enjoying the process and learning to manage one’s mental state to handle the inevitable fluctuations. For early-stage companies in Physical AI, he recommends starting with open-source tools and moving to platforms like Encord’s when scaling becomes necessary. Overall, Eric’s insights highlight the evolving landscape of AI, the challenges of building a startup, and the promising future of Physical AI as the next major technological platform shift.
