Stop Scaring Wildlife. Start Talking to It.

Most of the AI conversations I have are about data centers, agents and enterprise software. This one was about kangaroos, grizzly bears and birds on a runway, and it turned out to be one of the clearest examples I’ve seen of AI solving an expensive physical problem.

I sat down with Sara Nozkova at Disrupt the day after she pitched Flox Intelligence on the Startup Battlefield stage. The company started in 2020 as a spinoff from the AI lab at KTH in Stockholm and is now headquartered in Syracuse, New York. Flox is building what Sara calls a nature intelligence layer for machines and infrastructure: a way for airports, railways and highways to coexist with the wild animals around them, and to speak to those animals in their own language.

Why airports moved first

Airports are not known for buying technology early. So why are they Flox’s first customers?

Because they’ve tried everything else. Pyrotechnics, lasers and radar each have a place, but radar can only detect animals. None of these tools can tell an animal to stay away in a way it understands. And the stakes are hard to ignore: Sara noted that a single bird strike can easily cost more than a million dollars. When one event costs that much, a solution that keeps working is worth a lot.

Fear fades. Communication sticks.

The core problem with traditional wildlife deterrents is habituation. A loud bang or a flash of light might scare animals off for a day or two. Then they learn it isn’t really dangerous, and they come back.

Flox takes a different approach. Animals have been communicating for millions of years: marking territory, coordinating movement, warning each other about danger. Using AI, bioacoustics and wildlife science, Flox is decoding those signals, many of which people could barely hear or see before, and using them to talk back.

The context matters. It’s not enough to know the species. Sara explained that the right signal depends on whether the animal is, for example, a mother with young or an adult male. The message has to fit the animal it’s meant for.

Inside Flox Edge

Behind Sara during our conversation was Flox Edge, the company’s edge AI device. It was making sound while we talked. After five years of validation, Flox launched the devices commercially in May.

The units are mounted on infrastructure: at airports, along railways and highways, and near urban areas. Each one detects an animal, identifies its species, age, sex and behavior, then plays the communication cues for that species in that situation. The animal is guided away instead of being scared off.

What actually breaks: hardware and site-specific AI

I asked Sara what stops a customer from going from a few dozen devices to hundreds. Her answer was honest. Flox put about a hundred prototypes in the field four years ago just to collect data, and it took a long time to get two things right.

First, the hardware. The same device has to keep working through a Nordic winter and a Texas summer. Flox now has manufacturing and software partners to handle that.

Second, the AI. Flox is deploying with kangaroos in Australia and grizzlies in Canada, and each site needs a couple of weeks for the system to learn its animals and landscape. Getting that learning to fit into very small models that run directly on the unit was one of the team’s biggest challenges.

Tiny models, no cloud

This is the part I find most important for anyone building physical AI. Generic AI models fail in this setting, Sara said. A Flox unit has to run around the clock with no connectivity and no cloud, and it has to go from detecting an animal to acting in seconds. Flox’s spec sheet puts that at under two seconds.

So Flox builds very small, site-specific models trained on its own proprietary data for each species and location. Every deployment adds to what the company knows, and that knowledge becomes the next site’s starting point.

The animals are teaching the team, too

What has surprised Sara the most? Some calls don’t get the reaction the team expects. Some animals become more alert, some get curious, and some start interacting with the units. Flox has PhDs in both bioacoustics and wildlife behavior who use the devices as research tools and learn from every interaction.

Who Flox is for

For now, Flox sells to infrastructure: airports, departments of transportation dealing with animal collisions or trying to steer animals toward new wildlife crossings, large agricultural producers and governments protecting livestock from predators, and mining companies and national parks. In Sara’s words, Flox can help anywhere there’s a shared landscape with human-wildlife conflict today.

Consumers are next. Enough people asked that Flox opened a waitlist, and a couple of hundred have already signed up for Flox Home, a smaller Wi-Fi version for private gardens. Sara’s example was keeping deer from eating your plants. That’s on the roadmap, but this year the focus is infrastructure customers.

What breaks next

The obvious takeaway is about wildlife. The bigger one is about AI at the edge.

Flox is a reminder that the hardest AI deployments aren’t always in the data center. They’re on a pole beside a runway in winter, with no network, where a model has to make the right call in seconds for one species in one landscape. Generic models don’t work there. Small models, specific data and years of field validation do.

When a deterrent stops working, the cost isn’t only money. It’s safety, it’s animals’ lives, and it’s infrastructure that can’t do its job. Flox is betting the answer isn’t more noise. It’s learning to speak the language.

What Breaks Next is Mesa Capital Partners’ research series on what enterprises will actually adopt, and what fails along the way. If you’re an infrastructure, transportation or operations leader working out where AI fits in the physical world, get in touch to join the conversation.