The Machine That Watches Everything: How Ambient.ai Is Disrupting Reactive Security

He argues that the value proposition scales in both directions. A large enterprise with a fully-staffed GSOC can use Ambient.ai to do more with fewer people, extending proactive coverage across more sites than its human analysts could manage alone.

A mid-sized organization with 50 cameras and no dedicated monitoring operation gets something it has never had: genuine real-time situational awareness, delivered by AI, at a cost and complexity level that a full GSOC operation would never make practical.

“Day zero, they’re getting the same kind of security that a Fortune 10 global company that would have invested in a GSOC and 50 to 100 people on site would have gotten, because AI is doing everything a GSOC does, virtually,” Shrestha explains.

That framing carries significant implications for the integration community. The traditional path to enterprise-grade security has run through large system integrators, substantial hardware investments, and ongoing staffing costs.

Shrestha is arguing that AI compresses that path considerably, making sophisticated security operations accessible to a much broader customer base.

Video as Ground Truth

Shrestha is emphatic that video is the foundational layer of everything Ambient.ai does, and his reasoning shapes the company’s entire product philosophy.

“If I asked you to go secure a stadium in the best possible way, our best bet would be to hire 200 people and stand them everywhere around the stadium, observing and reporting," he says. "If they do see something, say something, and we respond, every incident can be prevented.”

The logic extends to sensors. Shrestha does not dismiss non-video sensors such as millimeter wave, time-of-flight, or LiDAR, and he brings credibility to the subject: before founding Ambient.ai, he built time-of-flight cameras at Google.

His position is specific: Any sensor that detects an event creates an immediate verification problem. Something triggered. Now what? “You have two options, he says. "You send a guard to take a look, or you pan a camera to it and visually verify it.”

The camera is always the endpoint. The AI is always the brain that makes sense of what the camera sees. For integrators who have built practices around sensor-based detection systems, the message is that those systems become more useful inside an AI-driven platform, not less. The platform, though, has to be the organizing intelligence.

The Remote Monitoring Question

Perhaps the most pointed disruption Shrestha describes involves remote video monitoring, a service around which many integration firms have built significant recurring revenue.

The traditional RVM model works because the environment it monitors is simple: A car dealership after hours, a construction site at night, a retail location outside business hours. The environment is sterile.

An analytic alerts, and a central station operator pulls up the camera, verifies the event, and calls it in or dismisses it. The human verification step is the business.

Shrestha believes that model will not survive contact with a truly agentic system. “If you have a small site and you deploy a product like Ambient, you don’t really need a human to look at the analytic and verify it, because the AI is that good. Not only that, you’re monitoring daytime too. Even when you have all this activity at the site, it can still see something suspicious. The environment does not need to be sterile.”

The implications are significant. RVM has historically worked at sites too small to justify a GSOC and too active during business hours to use analytics reliably. Shrestha's pitch is that agentic AI closes both gaps simultaneously. 

“Once truly agentic products are deployed at these sites, I don’t think remote video monitoring will exist the same way we think of it today,” he adds.

The Trust Problem

None of this matters if end-users refuse to adopt it. Shrestha is clear-eyed about the conservatism of the security buyer. “Trust is the word," he says. "The big thing is you just have to prove the technology works in the field.”

Shrestha says the company actively designs for transparency in how the platform communicates its reasoning: it generates plain-language descriptions of what it sees, so an operator reading an alert understands why the system flagged something. “Once they read that and understand, they kind of get it that it’s not the same foreground, background, motion detection analytic – it’s doing something deeper.”

There is also a technical answer to a question the industry is starting to ask about AI reliability: Who watches the AI? Shrestha described the use of ensemble models, a practice where multiple independent AI model paths analyze the same event simultaneously, each biased toward a different perspective or confidence threshold.

“The [AI] can analyze from different perspectives," he explains. "One model path may say, 'this really looks like a weapon.' Another model may say, 'it is a weapon, but it’s not a real weapon.' It may just be a Nerf gun.”

The ensemble approach builds a layer of internal verification into the AI’s own decision-making, reducing the risk of a single model generating a confident but wrong conclusion. For security applications where a false response can be as costly as a missed threat, that architecture matters. It is also the kind of technical detail that tends to sway conservative buyers, because it demonstrates that the vendor has thought carefully about failure modes, not just performance metrics.

What Comes Next

Ambient.ai has raised $146 million across four funding rounds, with its most recent closing in April 2025. The company counts Andreessen Horowitz, Y Combinator, and Allegion Ventures among its investors, along with George Kurtz, the CEO of CrowdStrike.

In March 2026, it announced a global reseller agreement with Securitas Technology, a move that signals an intent to scale beyond direct enterprise sales and into the broader integration channel.

The company currently goes to market with three agentic product lines, and the roadmap is already moving. In May, Ambient.ai announced new capabilities for Ambient Access Intelligence, its access control platform module, adding real-time diagnostics for chronic door infrastructure problems and live visibility into physically unsecured doors across a customer’s entire site.

Shrestha says the company intends to expand from its current three products to nearly 10 over the next two years, all oriented around automating discrete functions within a security operation.

“There aren’t many purely doing AI agentic software in the industry right now," he says, "and we think that market and that opportunity is growing.”

Whether or not that milestone lands on schedule, it is building toward a security operation where AI handles volume, pattern recognition, and initial response decisions, and where human operators handle judgment, leadership, and crisis management.

Shrestha describes the target ratio as 80% automated, 20% human, and he believes the timeline is shorter than most of the industry expects. “In the next two to three years, we’ll see a pretty strong change happening in the industry.”

For integrators, the relevant question is positioning. Agentic AI is arriving in physical security, whether the channel leads it or reacts to it. The firms that move early, build competency around these platforms, and help customers navigate the transition from reactive to proactive security postures will find themselves on the right side of that change. The ones that wait may find the conversation has moved on without them.