Here is a number that should bother anyone who has ever stood up a camera: in a 2024 industry survey, 62 percent of security alerts were ignored entirely. Not misjudged. Not acted on late. Ignored. The feeds were live, the events were captured, and no human did anything, because no human could. As one write-up on the problem put it, when every feed is an equal priority, none of them is.

This is the dirty secret of the sensor age. We got very good at deploying cameras and very bad at watching them. More feeds did not make anyone’s job easier. They made it bigger. A single operator can stare at one screen and still miss the thing that matters because they blinked, or got tired, or were looking at the other nine tiles. And a camera pointed at something interesting does exactly nothing for your shared map on its own. Someone still has to see it, understand it, and translate “there is a truck at the tree line” into a marker everyone can act on. That translation, human eyes to shared picture, is the bottleneck, and it does not scale.

For a small team, it is worse. You do not have a wall of monitors and a shift of analysts. You have a couple of people who are already busy doing the actual job. Asking one of them to also babysit a video feed is asking them to do neither well.

The fix finally fits on a phone

For years the answer to “nobody can watch all the feeds” was to ship the video to the cloud and let a big server analyze it. That works until it does not: you cannot stream high-resolution video out over a field uplink, you cannot wait a fifth of a second for a cloud round-trip when something is happening right now, and you cannot send sensitive footage off-device in a lot of the situations we care about.

What changed is that the analysis moved onto the device. In 2026, running computer vision at the edge, directly on the hardware holding the camera, is not exotic. It is the default. Modern object-detection models run at thirty-plus frames per second on the kind of neural processing unit that ships in an ordinary phone. The phone in your pocket can now watch a video stream and recognize what is in it, in real time, without sending a single frame anywhere. Low latency, no bandwidth cost, and the footage never leaves the device. All the reasons edge AI took over commercial security apply double in the field.

Which means the bottleneck we just described, human eyes translating a feed into a marker, can be handled by the phone itself.

What ICU Babushka actually does

ICU Babushka is that idea, built and proven. You point a phone at what you want watched, or feed it a video source, and its onboard AI watches the stream for you. When it sees something, it automatically places a CoT marker on the ATAK map for it. The detection lands on the shared picture on its own, in real time, with no human in the loop for the seeing-and-marking step.

Sit with what that changes. A phone becomes an unattended sensor that populates the common operating picture by itself. Instead of a feed that only helps if someone happens to be watching the right tile at the right second, you get a watchful node that flags what it sees, to everyone, the moment it sees it. Your teammates do not have to be looking at a camera. They are looking at the map, and the map now updates itself when the AI spots something worth a marker.

And because Babushka runs the whole chain on the phone, the same on-device design we build everything around, it carries the field-grade properties for free. The recognition happens locally, so it works when your comms are thin, because it is not depending on a cloud to think for it. The video does not have to leave the device to be useful, which matters both for bandwidth and for keeping sensitive footage where it belongs. And the marker it produces is small, cheap CoT data that rides your existing network easily, even when full video never could. The heavy thing, the video, stays put. The light thing, the detection, travels.

Under the hood, the same project also lets the phone serve its live video on demand and paint its field-of-view cone on everyone’s map, so a teammate can tap the marker and actually watch the stream when they want to. But the headline capability is the autonomy: the phone watches so your people do not have to.

Keep the human where the human belongs

We want to be careful and clear about what this is, because this is a place where hype gets dangerous and we do not traffic in it.

ICU Babushka is situational awareness. It sees, and it marks. It puts a flag on the map that says “something the AI recognized is here.” What happens next is a human decision, made by a person looking at that map with all their judgment and context intact. The AI is doing the tireless, unglamorous watching that humans are bad at, spotting the thing and raising a hand. It is not deciding what the thing means or what to do about it. That is not a limitation we are apologizing for. That is the correct design. The value is in freeing your people from staring at screens so they can spend their attention on judgment and action, which is the part that actually needs a human.

Used that way, it is a force multiplier for a small team. One phone can hold an overwatch on an approach, a perimeter, or a chokepoint, and quietly flag movement onto the shared map while your two or three people get on with the mission. It is the analyst you do not have, doing the one job that does not require being a person.

Who this is for

Anyone with more to watch than eyes to watch it, which, in the sensor age, is everyone. Small security details holding a perimeter with no one to spare for a monitor. Search-and-rescue teams wanting an extra set of tireless eyes on an area. Public safety operations that have cameras but not the staff to stare at them. Anyone whose feeds are, right now, functionally just expensive storage because nobody has the bandwidth to watch them live.

Turn one of those feeds into a phone that watches itself and speaks up on the map. If you want it tuned to detect what your mission actually cares about, that is exactly the kind of thing we build to spec. Email hello@spicycorp.com, or book a call.