Blab AI
Robotics Founders Discuss Vertical Strategy and Stack Ownership at Bessemers March 2026 Robotics Day
On March 12, 2026, Bessemer Capital’s inaugural Robotics Day unfolded in San Francisco, drawing three founders who have been shaping the future of physical AI.
The panel, moderated by investor Alexandra Sukin, featured Ury Zhilinsky, a former senior staff engineer at Waymo and founding technical member of Mind Robotics; Sandy Hefftz, founder and CEO of Bellboy Robotics; and Ted Stinson, former CEO of Covariant. Together they unpacked practical lessons for new robotics ventures, from selecting a vertical to building a complete hardware‑software stack.
Zhilinsky’s career illustrates the rapid pace of change in robotics. After eight years at Waymo, where the shift from rule‑based systems to end‑to‑end AI was incremental, he moved to Physical Intelligence, a well‑funded research lab. There he observed the same transition unfold in months, noting that the lab could produce a laundry‑folding robot within six months of starting.
Hefftz chose the hotel industry for Bellboy Robotics because it presents a clear pain point: labor shortages. A 600‑room hotel that relies on 150 staff each day to turn rooms over is a prime candidate for automation. The hotel environment is structured—rooms have predictable layouts, lighting is controlled, and tasks repeat daily—making it easier for robots to learn and adapt. Bellboy’s robots first mastered laundry sorting, then expanded to laundry services and event‑venue table‑setting.
Stinson’s experience at Covariant underscores the importance of customer urgency. In Covariant’s early days the team aimed to remain a pure model company, partnering with traditional automation providers. However, the delay in gaining customer insight limited the company’s growth. The decision to own hardware, deploy robots, and manage customer relationships accelerated product development and market fit.
All three founders agree that hardware reliability remains a bottleneck. Zhilinsky notes that no single embodiment has proven reliable enough for a clean stack. He argues that until hardware foundations are solid, companies that own more of the stack and iterate together will move faster than those that specialize too early.
A recurring theme was the need for a role that bridges research and real‑world deployment. Stinson describes the “field deployment engineer” position that emerged after researchers resisted spending weeks in a distribution center. The engineer must be technically capable, customer‑empathetic, and able to translate real‑world constraints back to the engineering team.
Funding stories also surfaced. Hefftz explains that Bellboy’s first round came from a hotel owner who directly experienced staffing pain, not from a venture capital firm. This early deployment proved the technology’s viability and opened the door to larger investors. Zhilinsky and Stinson emphasize that domain depth and a clear vision are now the primary signals investors look for.
The panel also touched on the state of AI models for robotics. Many companies build on language and vision models that were not designed for physical deployment. Zhilinsky says that Mind Robotics is collecting egocentric data directly on Rivian’s plant floor to train models tailored to industrial environments. Stinson notes that reinforcement learning for hard tasks requires building new infrastructure from scratch, and a reusable recipe would benefit the industry.
Looking ahead, Stinson predicts that factories and warehouses will shift from retrofitted robots to automation‑first designs. He sees hardware building blocks becoming modular, similar to Lego, and notes that integration and mass‑customization are now closer to reality than ever.
The panel concluded that while vertical robotics companies may dominate the next decade, consolidation is also possible. The key differentiator remains the ability to own a reliable stack and gather context‑rich data from specific environments.
In summary, the March 2026 Robotics Day highlighted that successful robotics founders must choose a vertical with acute pain, build or own a reliable hardware stack, create roles that bridge research and deployment, and maintain deep domain knowledge. Companies that combine these elements are better positioned to scale their solutions across industries.
The panel, moderated by investor Alexandra Sukin, featured Ury Zhilinsky, a former senior staff engineer at Waymo and founding technical member of Mind Robotics; Sandy Hefftz, founder and CEO of Bellboy Robotics; and Ted Stinson, former CEO of Covariant. Together they unpacked practical lessons for new robotics ventures, from selecting a vertical to building a complete hardware‑software stack.
Zhilinsky’s career illustrates the rapid pace of change in robotics. After eight years at Waymo, where the shift from rule‑based systems to end‑to‑end AI was incremental, he moved to Physical Intelligence, a well‑funded research lab. There he observed the same transition unfold in months, noting that the lab could produce a laundry‑folding robot within six months of starting.
Hefftz chose the hotel industry for Bellboy Robotics because it presents a clear pain point: labor shortages. A 600‑room hotel that relies on 150 staff each day to turn rooms over is a prime candidate for automation. The hotel environment is structured—rooms have predictable layouts, lighting is controlled, and tasks repeat daily—making it easier for robots to learn and adapt. Bellboy’s robots first mastered laundry sorting, then expanded to laundry services and event‑venue table‑setting.
Stinson’s experience at Covariant underscores the importance of customer urgency. In Covariant’s early days the team aimed to remain a pure model company, partnering with traditional automation providers. However, the delay in gaining customer insight limited the company’s growth. The decision to own hardware, deploy robots, and manage customer relationships accelerated product development and market fit.
All three founders agree that hardware reliability remains a bottleneck. Zhilinsky notes that no single embodiment has proven reliable enough for a clean stack. He argues that until hardware foundations are solid, companies that own more of the stack and iterate together will move faster than those that specialize too early.
A recurring theme was the need for a role that bridges research and real‑world deployment. Stinson describes the “field deployment engineer” position that emerged after researchers resisted spending weeks in a distribution center. The engineer must be technically capable, customer‑empathetic, and able to translate real‑world constraints back to the engineering team.
Funding stories also surfaced. Hefftz explains that Bellboy’s first round came from a hotel owner who directly experienced staffing pain, not from a venture capital firm. This early deployment proved the technology’s viability and opened the door to larger investors. Zhilinsky and Stinson emphasize that domain depth and a clear vision are now the primary signals investors look for.
The panel also touched on the state of AI models for robotics. Many companies build on language and vision models that were not designed for physical deployment. Zhilinsky says that Mind Robotics is collecting egocentric data directly on Rivian’s plant floor to train models tailored to industrial environments. Stinson notes that reinforcement learning for hard tasks requires building new infrastructure from scratch, and a reusable recipe would benefit the industry.
Looking ahead, Stinson predicts that factories and warehouses will shift from retrofitted robots to automation‑first designs. He sees hardware building blocks becoming modular, similar to Lego, and notes that integration and mass‑customization are now closer to reality than ever.
The panel concluded that while vertical robotics companies may dominate the next decade, consolidation is also possible. The key differentiator remains the ability to own a reliable stack and gather context‑rich data from specific environments.
In summary, the March 2026 Robotics Day highlighted that successful robotics founders must choose a vertical with acute pain, build or own a reliable hardware stack, create roles that bridge research and deployment, and maintain deep domain knowledge. Companies that combine these elements are better positioned to scale their solutions across industries.
