| Episode | Status |
|---|---|
| Episode | Status |
|---|---|
Sanjit Biswas is one of the rare founders who has scaled AI in the physical world – first with Meraki, and now with Samsara, a $20B+ public company with sensors deployed across millions of vehicles an...
Sanjit Biswas, CEO of Samsara, discusses building AI for physical operations at massive scale—90 billion miles of driving data annually across millions of vehicles. He explains why physical AI is fundamentally different from cloud AI, requiring distributed edge computing at 2-10 watts instead of kilowatts, and how foundation models are unlocking new capabilities like video reasoning and positive behavior recognition. The conversation covers practical challenges of deploying sensors at scale, the importance of go-to-market execution for technical founders, and why autonomy will increase operational intensity rather than replace human workers.
Sanjit explains how autonomy will increase operational intensity rather than replace workers, enabling a 'third shift' between midnight and 8AM when humans typically sleep. He argues that lower costs from automation will unlock massive latent demand, similar to how cheaper delivery changes willingness to pay for parts or services.
Sanjit traces his journey from MIT's RoofNet project through Meraki's $1.2B acquisition to founding Samsara. The through line is building products with real-world impact at technology inflection points. In 2015, he bet on the convergence of ubiquitous connectivity, cloud compute/GPUs, and smartphone-quality cameras to transform physical operations.
Sanjit details the engineering challenges of running AI inference on edge devices with severe power constraints (2-10 watts vs kilowatts in data centers). Samsara uses model distillation and teacher-student approaches to create task-specific models that run on mobile-phone-class hardware, focusing on relevant use cases like road risk detection rather than general knowledge.
Sanjit explains how advancing foundation models enable capabilities impossible 2-3 years ago. Beyond detecting negative behaviors (phone usage, no seatbelt), they can now detect weather conditions, road quality, and positive driver behaviors—allowing recognition of frontline workers doing great jobs 80-90% of the time, which previously went unseen.
Sanjit discusses what differentiates physical AI from pure software: the massive operational effort required to install sensors on millions of vehicles, train frontline workforces, and provide immediate value. Only a handful of companies (Tesla, Samsara, Waymo) operate at this scale, requiring thousands of people for installations and change management.
As a self-described engineering nerd who avoided selling candy bars in school, Sanjit learned go-to-market execution through necessity during Meraki's growth through the 2008 financial crisis. With risk capital turned off, they had to operate at breakeven, treating sales as an engineering problem. This lesson carried to Samsara, which has reinvested ~$3B from revenue and gross margin.
Sanjit sees autonomy as inevitable across all physical operations sectors. Warehouses automated ~10 years ago and workers welcome it for injury reduction. Road autonomy (Waymo) has moved from prototype to preferred option. Construction and job sites are next. He's cautiously optimistic on humanoids, comparing them to where self-driving was 10 years ago.
Samsara's product journey started with GPS tracking in 2015 when customers still used MapQuest printouts and got location updates every 5-15 minutes (while Uber had real-time tracking). A weekend webcam prototype led to their largest product—AI dashcams detecting phone usage. The strategy is concentric circles: core use case, then adjacent use cases, now ~10 products.
Samsara launched during the 3G/4G transition and piggybacked on carriers' investments driven by YouTube consumption, which drove down marginal cost per gigabyte. Starlink represents similar opportunity—massive upfront investment but low marginal cost per device, enabling coverage for remote oil/gas operations where no roads or cellular exist.
By 2030, Sanjit expects AI adoption in physical operations to move from base of curve to mainstream, similar to how apps became ubiquitous. He's excited about AR wearables and visual feedback for hands-free frontline workers, running VLMs to understand environments. AI helps distill sensor data overload into actionable insights, with customers discovering emergent use cases like automated time cards and customer ETA notifications.
Why the Next AI Revolution Will Happen Off-Screen: Samsara CEO Sanjit Biswas
Ask me anything about this podcast episode...
Try asking: