In his blog post, former COO Rui Xu recounts a year spent trying to build affordable humanoid robots at a YC-backed company that ultimately failed to secure Series A funding and folded by late 2025. Drawing on 15 years of hardware experience, Xu distills six lessons about where the startup, and many like it, went wrong, with a focus on hard engineering realities that can’t be glossed over by hype or AI enthusiasm.
The first lesson warns against what Xu calls “Large Model Chauvinism,” the belief that advanced AI alone can make up for weak hardware design. At his startup, the team debated adding simple mechanical safety features such as joint end stops, opting instead to rely on AI to respect physical limits. That misplaced confidence nearly caused hardware damage.
Xu also challenges oversimplified analogies in pitch decks, equating humanoid robotics to hoverboards or phones ignores the vastly greater mechanical complexity. He stresses that such narratives serve fundraising more than engineering.
A third key insight is that supply chain isn’t a checkbox task but a capability that must be built, especially when dealing with multiple manufacturers and tight tolerances. Relatedly, he argues there is no “commodity” hardware in robotics yet; custom parts still dominate, and undervaluing hardware expertise undermines the team.
Xu’s fifth lesson highlights that bad R&D decisions erode momentum faster than bad luck. Focusing on the toughest problems without converging on solutions quickly drained resources and investor interest. Finally, he cautions against unrealistic timelines, noting that haste often leads to skipped engineering steps that cost more time later.
The article closes with a personal reflection on leadership and a call for engineers to respect physics, timelines, and manufacturing realities if embodied AI is to succeed.