What if everything could just talk to each other?

Software first, then hardware

I was wondering the other day how we could connect various bits of software to each other, even where APIs don't exist.

It's a problem I bump into all the time. A lot of the tools businesses rely on, including plenty of the ones used in legal work, were never designed to talk to anything else. Normally, if two systems need to share information, someone has to build a bridge between them. That bridge is an API: a strict rulebook that says ask for exactly this, in exactly this format, and you'll get exactly that back. If the rulebook doesn't exist, you're stuck copying and pasting.

AI is starting to change that. An AI agent can increasingly work with software the way a person would: read what's on the screen, understand what it's looking at, and move information from one place to another. It isn't perfect, and it still needs a human keeping an eye on it, but the gap between "these systems don't talk" and "these systems talk" is narrowing.

Then I wondered: what about hardware too?

That's more difficult. You can't ask an agent to read the screen of a fridge that doesn't have one. Unless, that is, the devices themselves can speak.

A small brain in the appliance

Picture a fridge that knows what's running low. A car that knows its charge and where it's parked. A coffee machine that knows it's nearly out of beans.

Each has enough intelligence to understand its own little world. None needs to be a brilliant conversationalist. Frankly, I don't need an interesting fridge.

The conversation happens through a personal agent on your phone, watch or a speaker on the kitchen worktop. You tell it what you want. It works out which devices to ask.

In this imagined version, the agent checks the fridge, asks whether the car has enough charge for a shopping stop and suggests a route home. One request, rather than three apps and a small domestic negotiation.

I'm not describing a product you can buy. I'm wondering whether this is where the pieces could lead.

Would AI replace APIs?

Not in the tidy way that headline suggests.

An API is an interface that lets software ask another system for information or get it to do something. Plenty of devices already share standard interfaces. They don't all need a new connection written for every possible pairing.

What might change is the work we do around those interfaces. An agent could take an ordinary request, work out which systems are relevant and use their interfaces on our behalf.

The plumbing still matters. So do permissions, security and a reliable way of knowing what each device can actually do. Asking in plain language doesn't remove the need for precise instructions underneath.

Less fiddling for the person using it, perhaps. Not the end of the API.

Could the conversation stay in the house?

This is where edge AI interests me. Some processing could happen on the devices themselves, or nearby, instead of sending every request to a distant cloud service.

That could mean quicker responses and less information leaving the home. I'd rather my fridge and my car kept as much of their conversation there as possible.

But local doesn't automatically mean private. Someone still needs to decide what is stored, who can access it and whether another part of the service sends information elsewhere. A chip in the kitchen isn't a privacy policy.

And not every appliance needs its own AI model. An ordinary sensor connected to a capable hub might do the job perfectly well.

Someone has to make it worth building

My suspicion is that this only becomes ordinary when a big platform makes it worth manufacturers joining in. Apple is an obvious example. So are Google and Amazon. That is a commercial hunch, not a prediction about anyone's product plans.

There is already work here. Matter is a smart home standard intended to help compatible devices from different brands work together. We aren't starting from nothing.

What I'm imagining goes further: an agent understanding a request across devices, checking its permissions and knowing when it needs to come back to you. Compatibility is part of that, not the whole answer.

The useful outcome isn't one brand controlling every appliance. It's enough common ground that buying a new coffee machine doesn't require a new digital household.

The business question isn't about the fridge

Move the same idea into a business and it becomes more interesting. A stock system notices a shortage. An agent checks the delivery schedule and prepares an order.

Convenient. Until "prepare" quietly becomes "place", or it orders from the wrong supplier.

These are the questions I'd want answered before letting it loose:

1. What is it allowed to do? Reading stock levels, suggesting an order and spending money should not be treated as the same permission.
2. When does a person approve it? Set limits around value, suppliers and unusual requests. Decide what happens when the agent is unsure.
3. Who carries which risk? Check what suppliers promise about their services, what their terms exclude and how the different parts fit together.
4. Can we see what happened? Keep a useful record of the instruction, approval and action, with a practical way to stop or correct things.

Those are commercial design questions before they become a disagreement. They are also a reason to consider commercial legal support alongside the technical work, rather than after the first problem.

This is general commentary, not advice on a particular system or contract. Silva is a legal advisory and AI consultancy, not a law firm. Any support depends on the agreed scope and what we can and cannot help with.

Thinking aloud, not a forecast

I spend a fair amount of my working day alongside AI tools now. My mate Claude gets plenty of use. It has changed how I think about the gaps between systems, and how much of the work in those gaps a person still needs to do.

I may be ten years early or barking up the wrong tree. But if devices become easier to coordinate, I suspect the interesting question won't be whether they can talk.

It will be what we've allowed them to agree.

Frequently asked questions