Tanay Jaipuria interviews Mehul Nariyawala, co-founder of Matic, about how the company shipped home robots to over 10,000 homes after an 8-year journey from demo to deployment. The core insight: robotics companies should start by solving existing customer problems in tedious markets rather than leading with flashy capabilities. Matic chose robot vacuums—a 15% household penetration market where iRobot had generated $700M annually without real innovation in two decades. Nariyawala learned this lesson the hard way at his first company, Flutter (a gesture-control app acquired by Google), which won #1 in 73 countries but never became a viable business because nobody actually wanted gestures. The filter is simple: existing markets only ask "which one?" not "why do I need this?"—a massive advantage when customers already understand the problem. Form factor decisions matter far more than most robotics companies realize. Unlike the hundreds of identical disc-shaped vacuums that dominated the market, Matic built a square robot roughly double the height because the shape follows from the actual job. Disc robots were round because they had no intelligence and just bounced around blindly; a circle can pivot on its axis to escape corners. Matic's precise 3D vision eliminated that constraint, so they designed the robot around cleaning corners and sides, with cameras positioned at the height of a crawling child. On the hardware-software tradeoff, Matic made a Tesla-style bet: five cameras plus a $150 NVIDIA GPU instead of adding sensors that would explode complexity. Each sensor addition means roughly three more software engineers, worse calibration headaches, supply chain fragility, and exponentially higher manufacturing costs. The payoff is that software updates now ship improvements every few weeks while the hardware stays relatively simple to scale. The gap between a flashy demo and production is brutal in robotics. Matic's "clean this" gesture demo worked in fall 2018 but didn't ship until eight years later. A great demo gets you 20% of the way; the other 80% is the grinding work of productization—firmware, operating systems, testing infrastructure, edge observability, and the app itself. Consumers are ruthless about failure on tasks they consider trivial; Matic gets emails complaining about a single piece of popcorn left behind. That demands a 99.9% reliability bar for production, which takes roughly the same effort to achieve each additional nine as the last one. Real deployment data from thousands of homes is now Matic's competitive moat, since simulation and teleop get you to 80% but can't bridge the sim-to-real gap without actual customer data at scale.