Introduction

For most people, artificial intelligence still lives inside a chat window or a search bar. But a quieter shift is underway: AI is moving out of the cloud and into physical objects that can see, decide, and act without a screen. A Brooklyn-based hardware invention studio called Tomorrow Lab has spent the last year exploring exactly what that looks like, building three devices where intelligence is designed into the hardware from the very beginning.

What Happened

The studio developed three distinct projects, each solving a tangible problem rather than simply adding AI for the sake of it.

A Watcher for the Pulaski Bridge

Near the studio’s Brooklyn location, a drawbridge opens on its own schedule, often trapping cyclists and drivers on the wrong side. Tomorrow Lab created a compact monitoring device that watches the bridge with a webcam and uses local AI vision to predict when it is up. The hardware pairs an Arduino UNO Q board—running a Qualcomm Dragonwing processor under Debian Linux—with an STM32 microcontroller for real-time control. Because the vision model runs entirely on the device, the system works offline inside a box about the size of a sandwich.

A Pet Bowl That Knows Who Is Eating

In a second build, the team tackled a familiar issue for multi-pet households: one animal finishing its own meal and moving to another’s bowl. They designed a feeder that recognizes which pet is approaching and only dispenses food for the right animal. The core is an Analog Devices MAX78000 microcontroller with a built-in neural network accelerator, allowing image recognition to happen instantly at the edge. No images are sent to the cloud, and there are no ongoing cloud fees.

A Networked Leak-Detection System

The third project focuses on home and facility safety. Using Golioth, a cloud platform for managing connected hardware, the team built a suite of sensors that detect gas or water leaks. Individually, a sensor can flag a problem. Together, the network streams readings to a shared dashboard where patterns can reveal the true source of a growing leak. Firmware and detection logic can also be updated remotely after the devices ship.

Why This Matters

These builds illustrate a broader principle: an AI system is only as capable as the sensors feeding it, the mechanism it controls, and the power budget it runs on. When hardware and intelligence are developed together instead of bolted on afterward, everyday objects can gain new abilities without relying on constant connectivity. The same low-power recognition used in the pet bowl could apply to access control or quality inspection, while networked leak sensors show how fleets of devices can improve over time through over-the-air updates.

Key Takeaways

  • AI is shifting from software-only products into physical objects that interact with the real world.
  • Each project began with a real-world problem, then selected the right chip or platform for the job.
  • Local, offline inference reduces cloud costs and keeps data on the device.
  • Connected sensor networks can identify patterns that a single sensor might miss.
  • Designing hardware and AI together from the start unlocks capabilities that neither could achieve alone.

Conclusion

The next generation of AI products will likely look less like apps and more like ordinary objects with extraordinary awareness. Teams building them will need to think in atoms and algorithms at the same time, choosing sensors, chips, and mechanisms with the same care given to models. Tomorrow Lab’s three builds offer a working preview of what that future looks like.