AI has largely been something we encounter through a screen. You type a question into ChatGPT, Claude, or another chatbot, and it responds. Maybe you speak instead of type, but the basic experience is still digital: you communicate something to the AI, and the AI communicates something back.
This has had extraordinary uses, but it also gives us a narrow idea of what AI can be.
Everything humans understand about the physical world comes through some form of input. We see color and distance. We feel temperature. We hear sound. We notice motion. But our senses are imperfect, which is why humans have spent centuries building tools that measure the world more precisely.
A machinist uses calipers and micrometers instead of eyeballing a dimension. We use thermometers instead of guessing something’s temperature. We use scales, humidity sensors, depth gauges, compasses, accelerometers, GPS receivers, light meters, and thousands of other instruments because better measurements help us make better decisions.
Now imagine giving AI access to those instruments.
Instead of waiting for a human to read a sensor, interpret the number, enter it into another system, and decide what to do next, the measurement can go directly to software that understands the context around it. The AI can interpret what the sensor is reporting, compare it with other information, communicate the result to a person, and potentially trigger an action.
That sounds like the sort of project you might expect to require a robotics lab, an engineering department, and a very large budget. But not necessarily.
In fact, you can start experimenting for about the price of a nice dinner.
BLOG: What Students Are Asking Us About AI
A Tiny Device That Changes the Conversation
I recently built a device for Internality as a test of exactly this idea. It is roughly two inches by two inches, has a small screen, and can be held in your hand. You can touch it, talk to it, and it connects to a chatbot hosted on a Hetzner server.
It even mounts to LEGO bricks.
The device has a small “brains” module, a battery pack, and the ability to connect to other components. Those components can include soil sensors, light sensors, motion detectors, LIDAR, clocks, compasses, accelerometers, and plenty more. If a sensor exists and can communicate with a small computing platform, there’s a good chance you can make its information available to an AI system.
And this main unit, capable of doing so much, only costs about $45.
Sensors can cost anywhere from a few dollars to more than $100 depending on what they measure and how sophisticated they are. A basic soil sensor, for example, might add roughly $8. That means you can put together a simple, functioning proof of concept for around $50.
But bring up AI and most people won’t immediately think of devices like these. Many don’t know this type of accessible build is possible.
Mention “custom AI hardware” and it’s easy to picture a six-figure development project, a team of electrical engineers, custom circuit boards, industrial designers, manufacturing contracts, and months of development. All of those things might eventually become necessary if you decide to turn a successful prototype into a commercial product. A rugged, secure, production-ready device is very different from something built to test an idea.
But most companies don’t need to start by manufacturing 10,000 units. They can start with $50 worth of hardware and a good idea.
BLOG: How to Prepare Professional Services Firms for AI
What Happens When AI Can Understand a Flower Pot?
A flower pot is a simple way to understand the opportunity.
Put a moisture sensor in the soil and it gives you a measurement. That isn’t new. Gardeners and agricultural businesses have been using sensors for years.
But the measurement alone is only data. Say the sensor reports a particular moisture level. Is it good? Is it bad? Does this plant need to be watered now, tomorrow, or not for several days?
Connect the sensor to AI and the interaction becomes more useful. The system can receive the measurement along with information about the plant, its preferred conditions, recent readings, or other environmental factors. Instead of forcing a person to interpret a number, it might simply report that the plant is healthy and doesn’t need attention today. Or it might recommend watering within the next 24 hours.
You don’t even need the output to be another app notification. You could connect the system to a small servo and have it physically raise a flag. Green means everything is fine. Yellow means check soon. Red means take action.
Now you have something that senses the environment, interprets what it measures, and communicates a recommendation in a way that anyone can understand immediately.
Take the idea one step further and the device could potentially initiate an action itself. If the soil reaches a defined condition, it could trigger a watering system.
Of course, not every company needs an AI-powered flower pot. The point is the architecture: sense something, interpret it, communicate it, and potentially act on it. Once you understand that pattern, the number of possible business applications gets much larger.
AI Doesn’t Have to Wait for Someone to Type Something
Most business software still depends on humans to translate the physical world into digital information.
Someone observes something. They write it down. They type it into a system. They fill out a form. They upload a file. Only then does software get involved.
AI-enabled physical devices can remove some of those translation steps.
A sensor can measure temperature, humidity, distance, vibration, movement, light, orientation, location, pressure, or hundreds of other conditions. A device can identify a person through NFC. A microphone can capture an answer. A camera or LIDAR unit can collect information about physical surroundings. An accelerometer can tell you how something is moving.
AI can then help make sense of those inputs.
That creates an interesting exercise for a business owner: look around your organization and ask where people are currently observing, measuring, recording, checking, sorting, or interpreting something manually.
Where does someone read a measurement and decide what it means? Where does information get written down only to be entered into a computer later? Where does valuable context disappear between an in-person interaction and the software your team uses afterward?
Those are the places where a relatively simple device may be worth experimenting with.
How Could You Better Capture Focus Group and Jury Research?
Let’s think outside the flowerpot.
Imagine a focus group session in which every participant has an NFC card connected to their profile. When it’s time to answer a question, the participant taps the card against a device and responds. The system knows who is answering, captures the response, and associates it with the appropriate participant information.
For a business conducting focus groups or a law firm conducting jury research, that could substantially change what happens after the session.
Traditional research can generate piles of notes, survey responses, spreadsheets, recordings, and observations that need to be organized before analysis really begins. Someone must connect a particular response back to the person who provided it. Someone may have to transcribe what was said. Someone must prepare the information for whatever analysis comes next.
A purpose-built device could make the information structured and AI-ready from the moment it is collected. Then researchers can begin working with the information sooner and in richer ways. The physical device becomes a simple interface connecting people in the room to a much more sophisticated analysis system behind the scenes.
BLOG: Practical AI for Law Firm Intake
Cheap Prototypes Can Change How Companies Innovate
When a prototype costs $50,000, an organization must be convinced the idea is good before anyone builds it. There are budgets to approve, projections to make, stakeholders to persuade, and a great deal of pressure to predict the outcome.
That is a difficult way to innovate because the entire purpose of a prototype is to discover things you don’t know yet.
When the hardware costs $50 or $100, the question changes from “Can we prove this will work before we build it?” to “Why don’t we build a small version and find out?”
Put it in someone’s hands. See whether they understand it. Watch what they do with it. Find out whether the information is actually useful. Discover which assumptions were wrong. Learn what features nobody cares about and which unexpected feature suddenly seems essential.
Then decide whether it’s worth investing more.
This is the same test-and-learn mindset we already use in marketing, product development, and digital experiences. The difference is that inexpensive hardware and accessible AI now make the approach practical for physical experiences as well.
The Hard Part Is Determining Direction
None of these components is especially secret. Sensors are inexpensive and widely available, and small computing modules are increasingly capable. The AI APIs and models are accessible to businesses of almost every size if you know where to look.
The harder and more valuable part is recognizing the right problem to solve.
Not every object needs AI and not every manual process deserves automation. Attaching a chatbot to something does not automatically make it useful. Sometimes, it just makes it more annoying.
The opportunity is to identify moments where better measurements, faster interpretation, cleaner data, or easier interaction can improve a real business outcome.
That might mean giving researchers cleaner information, helping employees identify a problem sooner, collecting field observations without forcing employees to stop and fill out a form, or giving customers an entirely new (and helpful) way to interact with a service.
The device itself is only part of the solution. The important work comes in designing the experience around what people actually need.
Your First AI Device Might Cost Less Than Your Office Printer
We’re still early in the transition from AI as something we visit in a browser to something embedded in the physical environment. And all you might need to start is a little screen, a battery, a sensor, and a very specific problem.
For a relatively small cost in hardware, you may be able to test an idea that your organization would previously have assumed was too expensive, too technical, or simply impossible.
At LaFleur, that’s the part we’re excited about. We can help organizations identify practical opportunities for AI, build custom AI-enabled devices and experiences, connect them to the right systems, and determine whether the idea creates enough value to develop further.
Have an idea for something AI should be able to see, measure, hear, or respond to? Talk to LaFleur. We can help you figure out what’s possible.
