Netradyne ships AI agents that coach drivers and assemble crash reports

Netradyne, whose dashcams sit behind the windshield of roughly half a million commercial trucks and delivery vans, rolled out a new platform on Tuesday that turns the video and telematics stream from those vehicles into work that AI agents can do on their own. The platform, called Netradyne Intelligence, is the company's first major product release that pitches its fleet safety technology as a set of agents that act, not just alert.

The product lands as a wave of "enterprise AI agent" launches have been reshaping what enterprise software is supposed to look like. It is the latest in a string of enterprise agent platforms that put a real credential model around what the agent can touch. Beyond Identity shipped a device-bound passkey product for agent identity on Monday. AppViewX shipped agent credentialing on Tuesday morning. Databricks, Salesforce, ServiceNow, Microsoft, and Google have all told the market that the next chapter is agents, not copilots. Netradyne is the first fleet-safety vendor of any size to put the same framing in front of buyers, and the bet is that the same plumbing that has been selling dashcams to fleets can be sold a second time as the work surface for AI agents.

What Netradyne Intelligence does for fleet managers

Netradyne Intelligence is a layer that sits on top of the company's existing Driveri dashcam, which combines four inward- and outward-facing camera views, a GPS feed, an accelerometer, and on-device AI to flag risky driving events in real time without depending on a cloud round trip. The new platform reads the same telemetry stream and turns the events into three concrete agent workflows. The first is a Coaching Agent that watches the manager-defined safety goals, scores drivers against them, and walks a fleet safety manager through a coaching session, complete with the relevant video clips, the driver's recent alert history, and a recommended improvement plan. The second is an Incident Response Agent that, when a collision is detected, automatically pulls the full collision record, including the video, the first-notice-of-loss paperwork, a crash reconstruction, and the driver's history, and routes it to the right stakeholders. The third, a Driver Recognition and Reporting Agent, is scheduled to ship in the third quarter of 2026 and is meant to take the manual work out of running a safety rewards program.

The platform is built on the same edge-AI stack that Netradyne's hardware already runs, which is the part of the announcement that matters most to a fleet operations buyer. The AI agents are not calling out to a third-party foundation model for every decision. They are calling out to a fleet-specific model that was trained on roughly 30 billion miles and 150 billion minutes of real-world driving telemetry that Netradyne's installed base has been feeding back to the company for years. The training set is not just dashcam video. It is dashcam video tied to the GPS, the accelerometer, the truck's CAN bus, the telematics, and the outcomes, including collision records and the resolution of past insurance claims. That is the part of the announcement that makes the product meaningfully different from a generic AI agent layer bolted on top of a camera.

The pricing story is more measured than the announcement makes it sound. Netradyne is bundling the platform into the existing Driveri subscription, and the Coaching and Incident Response agents are available to existing customers now, with no extra per-driver fee through the end of the third quarter. The Reporting and Recognition agent, when it ships, will be priced as a separate module. The company is also committing to a "no additional cloud storage" stance for the agents, meaning the video, the crash reconstructions, and the coaching session recordings stay on the same storage the fleet was already paying for. That is a deliberate answer to a complaint that the fleet industry has had about most AI products in the category, which is that the unit economics break the moment the vendor starts charging per inference or per gigabyte of data the AI touches.

How the platform reads 30 billion miles of drive data

The 30 billion mile and 150 billion minute training set is the headline number, but the technical story is more about how the data is segmented than how big it is. The training set is broken into two parts. The first is the raw event stream: video clips, accelerometer traces, GPS pings, and the Driveri system's own risk scoring. The second is the labelled outcome set: every collision the company has been able to confirm through insurance records, every safety incident a fleet reported back to Netradyne's customer success team, and every false positive that the on-device model flagged but that a human reviewer later overruled. The labelled set is what makes the agent workflows more than just a different UI on top of the existing dashcam. The Coaching Agent is essentially a model that has been trained to predict which combination of recent events in a driver's history is most likely to lead to a collision in the next 30 days, and to recommend the smallest possible set of coaching actions that the safety manager can take to defuse the pattern.

The edge-AI piece is what the company is leaning on hardest in the announcement, and it is also the piece that is hardest to verify from the outside. The on-device model in Driveri has been running inside the dashcam hardware for years, and Netradyne says the new agents do not require a hardware swap. That is significant because a hardware swap across an installed base of 500,000 vehicles is a non-starter for most fleets, and it is the reason Netradyne's competitors in the dashcam market, including Samsara, Motive, Lytx, and Geotab, are not offering the same kind of agent workflows yet. The bet is that the on-device model can do most of the inference locally, and that the agents only call out to the cloud for the heaviest decision points, like the crash reconstruction and the cross-driver pattern analysis. Avneesh Agrawal, the company's CEO and founder, framed the platform on the launch call as a way to "turn insight into consistent, timely action at scale," which is industry-speak for "the AI does the work the safety manager used to do."

The third piece of the platform is a video search tool called Video LiveSearch that the agents sit on top of, and that is the part of the announcement that fleet safety managers are most likely to feel first. The tool lets a manager type a natural-language query, like "show me every hard-brake event on I-95 in the rain last Tuesday," and get back a ranked list of video clips from the relevant vehicles in seconds. That sounds like a small thing, but it is the kind of tool that has been on the wishlist of every fleet safety manager for a decade, and the fact that it is now available without leaving the Netradyne product is a meaningful step forward. It also makes the agent workflows easier to use, because the Coaching Agent and the Incident Response Agent can both pull video into the workflow without the manager having to switch tools.

Where the Netradyne platform goes next

The near-term roadmap is mostly an extension of the three agent workflows, with the Reporting and Recognition agent scheduled to ship in the third quarter of 2026 and a fourth agent, focused on compliance and driver qualification file management, scheduled to be announced in the fourth quarter. The compliance agent is the one that the large private fleets are most likely to be waiting for, because the manual work of keeping a driver qualification file current is one of the biggest sources of waste in a safety operations team. Netradyne is not alone in seeing the opportunity, and the company is going to be competing for that workflow against Motive, which launched a similar compliance product in May, and against Geotab, which has been quietly building a compliance agent of its own on top of its MyGeotab platform. The race to automate the back office of a fleet safety operation is now a real race, and Netradyne is the first of the three major dashcam vendors to put a stake in the ground.

The bigger question is what the platform means for the rest of the enterprise AI agent market. The fleet safety story is not the most glamorous part of the enterprise AI conversation, but it is one of the first places where a large installed base of edge devices is being asked to do real agentic work, not just chat with a model. If the Netradyne platform works as advertised, it will become a reference design for every other category where a vendor has an installed base of edge devices, including the industrial robotics vendors, the security camera vendors, the point-of-sale vendors, and the field service management vendors. The product is also a useful data point for the broader conversation about enterprise AI agents, which has been dominated by software-only platforms like Salesforce, ServiceNow, and Microsoft, and which is now starting to look like it is going to be a physical-world story as much as a software story.

The other near-term question is whether the platform can hold up under independent benchmark scrutiny. Netradyne is making big claims about the accuracy of the on-device model and the safety outcomes of the agent workflows, and the company has not yet published an external benchmark. The most credible test will be the independent safety reports from the large fleets running the new product in production, which the company says it will publish in the fourth quarter of 2026. If the reports show the kind of safety outcome improvement that the company is implying, the platform will be a major step forward for the fleet safety industry. If they do not, the product will still be useful, but the gap between the marketing claim and the real-world impact will be the kind of gap that fleets tend to notice the second time around.