DataFramed #347 Let's Get Physical with AI with Ivan Poupyrev, CEO at Archetype AI
28 snips
Feb 23, 2026 
Ivan Poupyrev, CEO and founder of Archetype AI and former Google ATAP lead behind Soli and Jacquard. He talks about turning raw sensor streams into meaning, how physical AI differs from LLMs, sensor fusion wins like wind-turbine alerts, edge deployment and privacy, and how organizations can pick practical starter projects for real-world intelligence.
AI Snips
Chapters
Transcript
Episode notes
IoT Solved Connectivity Not Meaning
- IoT succeeded by connecting physical devices and creating data pipelines, but it stalled at turning raw data into meaning.
- Poupyrev argues foundation models will convert massive sensor streams stored in clouds into actionable insights and decisions.
Three Layers Of Physical AI Applications
- Physical AI use cases stack: insights (what data means), recommendations (what to do), and automation (act on it and coordinate systems).
- Poupyrev highlights discovery of unknown failure modes and system-level optimization across assets and homes.
Physical Models Must Be Grounded In Measurements
- Physical foundation models differ from LLMs because their primary data are sensor measurements and video, not text, so architectures must be grounded in physical measurements.
- Physical accuracy matters: hallucinations acceptable in creative media are dangerous in control scenarios like power stations.