When AI brings your loved one home safe

When people think about serious artificial intelligence companies, they think OpenAI, Anthropic, or Google. In the hardware space, Tesla or Apple. They don’t usually think about Samsara, but they should.

We process over 25 trillion data points a year, all physical, real-world data that Samsara’s sensor network uniquely positions us to collect. That includes more than 100 billion miles of driving annually, 120+ billion API calls, and coverage across 99% of major U.S. roads. We run machine learning continuously across millions of devices and in the cloud. That’s real-time edge inference in concert with the cloud and large-scale, distributed cloud training, all processing an enormous scale of intricate, real-world multimodal signals.

If you step back and look at the volume and diversity of the data we process, physical-world signals across vehicles, job sites, and infrastructure, it becomes clear that we are operating in a problem space that is both large and deeply complex.

The scale and complexity of our models, combined with the stakes of the environments they operate in, make this one of the hardest and most meaningful challenges in AI.

A problem space that gets more complex the deeper you go

Building something in the real world, at scale, is drastically harder than in a controlled environment.

Running these models on millions of devices means working with limited compute, not racks of GPUs but hardware closer to a smartwatch. They have to operate continuously on highways and construction sites, in rain and snow, at night, and with unreliable connectivity. They have to operate across different camera placements, vehicle types, and human behaviors.

Some companies can narrow their operating domain and expand gradually. We can’t. Our customers do not get to choose ideal conditions. They drive in bad weather. They drive on every type of road and offroad. Their work does not pause for perfect circumstances, and our systems can’t either.

It is not enough to mostly work. It has to work amazingly well, at high frequency, at scale, and across enormous environmental diversity. That intersection of hardware constraints, environmental variability, scale, and human interaction is what makes this not just technically interesting but one of the hardest challenges that exists in AI.

Where AI has to work, and “close enough” is not enough

Most importantly, we are dealing with real people’s lives, not just the frontline workers using our products but also the communities around them.

This isn’t a sandbox or demo environment. It isn’t a place where you can shrug and say, “We’ll fix it in the next release.” When you’re operating at highway speeds, with heavy equipment, or in school zones, small errors are not abstract. One moment of failure can mean someone doesn’t make it home.

So when we talk about getting it right, that is not marketing language. Every day, by conservative estimates, four people make it home safely because of our technology who otherwise would not. We will never know who they are. There is no headline about it. But for those people and their families, it’s everything. And they may never know that the outcome was different because our products did their job.

A lot of AI right now is measured in engagement, productivity gains, or benchmark improvements. Our AI saves lives. And that fundamentally raises the bar.

The way we build

It’s because our AI has to operate in the real world that we don’t look like any other AI company.

We are in the field. Our team, from the C-suite to new hires, spends time on highways and job sites, in hard hats and safety vests, understanding how systems behave under real constraints. We build where the work actually happens.

We ship quickly because getting into customers’ hands is how we learn. Feedback from the field accelerates the innovation loop in a way internal debate never can. We build something strong, deploy it, see how it performs in the field, and improve it.

Culturally, we are low ego and low politics. Engineers ship ideas in days or weeks, and those changes can reach millions of devices and way more people. You do not need layers of approval to make an impact. Ownership here is not only real but an expectation.

That mindset starts at the top. Recently, our Chief Product Officer, Johan Land, topped the ARC-AGI-2 leaderboard ahead of full teams from Google, OpenAI, and Anthropic with a system he built independently in his free time. Everyone here stays hands-on: they build, experiment, and seek out new technical frontiers.

That is the bar we set for our teams. We are looking for engineers who are deeply curious, technically rigorous, and motivated by real-world impact. Builders who want to ship, learn fast, and solve hard problems where being right matters.

Why do I work here?

In order of importance to me:

• This work matters at a human level. We are preventing accidents. We are reducing injuries. We are saving lives.

• We support the infrastructure that keeps the world running. Utilities, construction, transportation. There is an entire network of people most of us never think about who keep power flowing and water running. We help them operate more safely and more efficiently.

• The technical problems are genuinely hard. They sit at the intersection of scale, distributed systems, edge ML, human behavior, and environmental variability. This is operational, not theoretical, AI.

• It is an engineering culture of ownership. Anyone, from CEO to IC, can ship solutions into production in a matter of weeks and see their ideas deployed at scale.

Samsara is not building AI for clicks or demos. We build physical AI that supports the manual, often invisible work that keeps your and my world running and helps the people doing it make it home safely.

Come build with us.