Jul 29, 2026 · 1h 0m · mad
The Biggest AI Deployment Nobody Talks About | Samsara CEO Sanjit Biswas
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The MAD Podcast, host Matt Turck interviews Samsara co-founder and CEO Sanjit Biswas about running one of the world's largest deployments of 'Physical AI'. Biswas explains how Samsara leverages edge-to-cloud hardware, real-time video analytics, and autonomous AI agents to digitize real-world industrial operations, prevent hundreds of thousands of road accidents, and support frontline workers.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 23.3% of the talking time here. How this is scored →
Matt as informed peer 5.4 Guest teaching 5.3 Guest disagreement 0.4 Matt pushing back 1.3
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:12 Rejecting the Public Scraping Premise
Sanjit firmly dismisses the notion that foundation models alone can solve physical operations, stating that real-world operational tokens cannot be crawled on Reddit or simulated online.
Hardest push from Matt ▶ 48:48 Pushing on In-Cab Surveillance and Worker EvaluationMatt refuses a purely optimistic framing of dash cams, pushing Sanjit on whether continuous AI monitoring turns into an invasive 'Big Brother' judge of worker quality.
Biggest teaching moment ▶ 57:13 The 125-Year Utility Grid StatSanjit educates Matt on the sheer scale of the energy bottleneck, sharing a firsthand stat that a major utility must build more grid capacity in the next 5 years than in the preceding 125 years combined.
Matt holds his own ▶ 40:10 Dissecting Agentic Autonomy vs Rule-Based WorkflowsMatt displays deep systems understanding by drilling into the exact recipe of autonomous agents, probing whether agentic reasoning is replacing deterministic rules or merely sitting on top of them.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Podcast Channel Subscription Interlude | 4 | 4 | 0 | 0 | Matt sets up the episode by distinguishing physical AI from purely digital agents and robotaxis, inviting Sanjit to define the space. Sanjit explains that physical infrastructure represents un-digitized data trapped in the real world. | |
| Why Physical AI Is Harder and Higher-Stakes | 5 | 5 | 1 | 2 | Matt asks why Silicon Valley historically neglected physical operations and asks if it is fundamentally more unforgiving. Sanjit highlights that physical operations account for 40-50% of global GDP and frames real-world danger as a data opportunity. | |
| Samsara's Product Architecture and Scale Metrics | 5 | 6 | 0 | 1 | Matt cites Samsara's financials ($2B ARR, 30% growth) while Sanjit details their scale metrics, revealing that their safety systems process 25 trillion data points and prevented roughly 380,000 accidents in a single year. | |
| Sanjit Biswas's Journey: MIT, Meraki, and Samsara | 4 | 5 | 0 | 0 | Matt traces Sanjit's journey from MIT RoofNet and Meraki to Samsara, questioning how founders master new industries without prior domain experience. Sanjit explains the necessity of hands-on site visits and beginner's mind. | |
| Samsara's Hardware Stack: Asset Tags, Gateways, and AI Dash Cams | 6 | 6 | 0 | 1 | Matt proposes an anatomical framework for Samsara's hardware and software layers and asks about IoT wireless protocols like LoRaWAN. Sanjit demonstrates physical hardware tags, explaining Bluetooth Low Energy mesh architecture and edge AI dash cams. | |
| Cloud Ingestion, Persona-Based UI, and Moats | 5 | 5 | 1 | 1 | Matt asks about cloud ingestion pipelines and presses on whether foundation models will commoditize Samsara's software layer. Sanjit explains that physical operations data cannot be crawled from the public internet. | |
| Data Flywheels, Edge-Cloud Balance, and Model Strategy | 6 | 6 | 0 | 1 | Matt inquires about cross-customer data flywheels and technical model architectures like CNNs versus generative vision-language models. Sanjit details municipal pothole detection in Chicago and the split between edge inference and cloud reasoning. | |
| Samsara's Agentic AI and Warranty Automation | 6 | 5 | 0 | 2 | Matt explores Samsara's Agent Studio and asks about the balance between pure agentic reasoning and deterministic workflow rules. Sanjit demonstrates automated warranty reconciliation and walkaround inspection processing. | |
| Future AI Horizons: Continuous Video Compute and Cerebras Speed | 6 | 5 | 0 | 0 | Matt and Sanjit discuss near-term AI developments, focusing on continuous 8-hour driver ride-along video tokenization and the impact of ultra-fast wafer-scale inference from Cerebras. | |
| Addressing "Big Brother" Concerns with Frontline Exoneration | 6 | 5 | 1 | 4 | Matt pushes on the 'Big Brother' surveillance dynamic of permanent cab cameras acting as an inescapable judge of driver performance. Sanjit counters by emphasizing frontline exoneration from false claims and cultural change management. | |
| Autonomous Mixed Fleets and American Industrial Growth | 6 | 6 | 1 | 2 | Matt probes long-haul autonomous trucking timelines and overall American industrial dynamism. Sanjit reveals that electric utilities are planning to triple their grid capacity within five years, driven largely by data center demand. |
Statements from this episode (18)
Physical operations industries account for 40% to 50% of global GDP
“All of these industries, they make up about 40, 50% of world GDP.”
Samsara processes 25 trillion data points annually from millions of vehicles
“We were talking about 25 trillion data points, GPS, video, third-party API integrations, all kinds of data flowing into the system. Millions and millions of vehicles, for example we're talking about, you know, millions of frontline workers that are using our a…”
Samsara helped prevent roughly 380,000 road accidents in the past year
“We believe we help prevent about 380,000 car crashes you know, road accidents in the last year.”
About 10% of Americans do not regularly wear their seatbelts
“This is an interesting stat, but in the U S approximately 10% of people don't regularly wear their seatbelt.”
In-cab audio alerts break distracted driving habits via immediate negative reinforcement
“And the idea is they can self correct self coach. And that is like the aha is it breaks the cycle or the bad feedback loop of I'm going to look at my phone. If you get a sort of like audio alert in the moment and it happens many, many times, you tend to break …”
Physical AI moats require real-world edge data you cannot crawl online
“These are not the tokens you're going to find online. Like you can't crawl Reddit and find out about what happened on a construction site, right? Nor can you do you know, test time reasoning about it. You can't just, you can simulate all kinds of environments,…”
Samsara-connected vehicles drive 99% of U.S. roads every single day
“The Samsara system and a given day, we're driving 99% of the U.S. Roads, usually multiple times a day.”
Samsara uses a hybrid AI stack including frontier and open-weights models
“We use multiple models from different labs simultaneously. The open source models are pretty compelling. I think it's having its moment now, but we've been seeing and this is really, I think from the academic communities, open source and really open weights mo…”
Samsara AI agents reduce warranty labor from hours to under a minute
“One of the demos I did on stage is we have a warranty agent. When it sees a fault code, it can basically crack the service manual, look at your
specific like OEM negotiated warranty agreements and then correlate the two and say, yes, this specific issue, give…”
Most commercial fleet warranty claims go unfulfilled due to paperwork burdens
“The reality, by the way, is most of those warranty claims just are unfulfilled, right? Nobody has the time to go do all that work that I just mentioned and do the paperwork.”
Continuous AI driver video monitoring becomes commercially viable in 1-2 years
“That's basically a massive amount of video computation, right? So you can run a tokenizer. It just turns into like a lot of compute. You can do that today. It's pretty expensive and costly, but it works and it's doable. We're pretty optimistic that the, you kn…”
Cerebras cloud achieves 10x inference speedup over fast GPUs on Gemma
“Say, if you run Gemma four, On your GP, you might get like a hundred tokens per second if you have a fast card. If you run it in their cloud, you get anywhere from 800 to 1500 tokens per second. So call it 10 X faster.”
The primary use case for commercial dash cameras is driver exoneration
“Something people don't often think of is, you know, if you have a dash camera, what is it used for? The majority use case is actually exoneration.”
Home Depot achieved a 65% reduction in auto claims using Samsara
“I mentioned Home Depot earlier. They saw like a 65% reduction in their claims, like auto claims.”
Autonomous robots will perform construction site grading within five years
“We see robots being able to do that in the next five years.”
Long-haul trucking represents only a minority of commercial fleet vehicles
“On the trucking side, it's also important to realize, like, there's long haul trucking and kind of, like, moving stuff from point A to point B. That tends to be a minority fraction of what the commercial vehicles on the road are doing. Most of the commercial v…”
Specialized commercial vehicle autonomy could take 10 to 20 years
“So for that reason, we think the adoption might be a bit slower, but it's not like a no, it's just, it might take 1020 years. And these are industries where, again, the equipment's highly specialized. Like if you look at cement mixers or you know, garbage truc…”
Major US utility plans to triple 125-year grid capacity in five years
“I was just in the field last week with a large energy utility. And they've been involved in grid modernization, and they shared with me a really interesting stat. They said, you know, over the last 125 years, we built a certain amount of grid capacity in, in m…”