Samsara uses AI across fleet operations, dash cams, and field teams

John Bicket doesn't run a company most people have heard of, but if you've driven near a delivery truck or a utility van lately, you've probably seen Samsara's hardware without knowing it. The company builds dash cams, telematics units, and asset trackers for fleets — the eighteen-wheelers, forklifts, and service vans that keep supply chains moving. Bicket, Samsara's CTO, recently laid out how AI is getting woven into nearly every piece of that hardware stack, and the picture he paints is less about flashy demos and more about grinding operational reality.

The dash cams are the clearest example. Instead of just recording footage for insurance disputes after the fact, Samsara's cameras now use AI models to spot risky driving behavior as it happens — following too close, phone use, drowsiness — and flag it before it turns into a claim. That shift, from passive recording to active detection, is where a lot of the company's engineering effort is going.

Telematics and asset tracking get the same treatment. Instead of a fleet manager sifting through raw GPS pings and sensor logs, AI is layered on top to surface patterns: which vehicles are burning more fuel than they should, which routes are inefficient, which equipment is likely to fail soon. It's the kind of predictive maintenance pitch that's been promised for years, but Bicket's framing suggests Samsara thinks the underlying models are finally good enough to make it reliable at scale, across thousands of vehicles and job sites.

What's notable is how unsexy all of this is compared to the AI conversations dominating headlines. There's no mention of chatbots or generative writing tools here. This is AI applied to sensor fusion and computer vision problems that have existed for a decade, just now cheap and accurate enough to deploy broadly. Bicket's comments read less like a product pitch and more like an engineering progress report — here's what got easier to build this year, here's what field teams actually use.

And that's probably the more honest state of enterprise AI right now. Not everyone is racing to build the next foundation model. A lot of companies are just quietly retrofitting AI into hardware that's already out in the world, one dash cam and one forklift sensor at a time.