What "AI in the home" actually means

“AI” is now printed on devices that contain a timer. So how do you tell products apart? Look at one thing: what happens when the product is wrong. This page describes four steps. Each step can do something the one before cannot. Three questions at the end help you place any product on the ladder, including ours.

Step 1: Manual and on a clock

You work the house. A switch is a switch. Then the house starts to act on the clock: heating at six, garden lights at dusk, irrigation on Tuesdays.

A well-designed manual house, with good switches in the right places, is more pleasant than many automated ones. If automation stops and the house becomes hard to use, the automation was built wrong.

What it cannot do: a timetable is a guess made once. It is right about the average day and wrong about today. It does not know you are away, that it rained, or that the room is already warm. Much of the disappointment with “smart” heating is a timetable being asked to do more than a timetable can.

Step 2: Reacts to buttons and sensors

Motion switches on a light. One button sets eight dimmers, three blinds and an amplifier input. The house now reacts to the world, not only the clock.

Most high-end homes are here. A button press recalls a saved set of light levels. That is a snapshot from one day. Each rule acts on its own. Nothing remembers, and nothing decides when two rules clash.

That is not a criticism. Done well, this step is very good. But it explains why spending more inside it does not feel like a different kind of house.

Step 3: Understands what you ask

You no longer need the system’s exact words. “Make it cosy in here” turns into actions, even though nobody set up “cosy”. Many voice and chat assistants sit here.

What it cannot do: it still waits for you. In an empty room it does nothing. And a better chat in front of the same rules changes how you ask, not what the house does with the request.

A warning we want to be clear about. Some products connect an AI chat tool straight to the lights and locks. If the AI guesses wrong, the guess can become a real action. This is our view, not a fact about every product. The safer pattern is that the AI can only suggest, and fixed rules decide what happens.

Step 4: Suggests, and fixed rules decide

This is where Casient aims. It is early. It runs in one real home so far, and pilots are under way.

Here is how it works in plain words.

• The AI never talks to devices. Everything it asks for goes through the same checked request path as an app or voice command, and new rules wait for approval.
• Fixed code checks every request. Requests made through Casient are checked against the room behaviours and limits before anything is sent.
• No new rule goes live until your installer approves it. They see what will change first.
• A room’s own behaviour is checked first. A child’s bedroom in Night mode doesn’t light by itself. A switch press still works.
• A log says why. It shows which rule ran, what it overrode and what started it.

Rules are written and approved by your installer. Room behaviours come from the ORBS cards, an open draft standard. Your installer picks and approves them.

Casient does not control locks, fire systems or exits. Emergency lighting is a separate certified system.

Why every decision needs a log

If a system in your house does something, it owes you an answer to one question: why? A useful record is not just “light turned on at 18:42”. It says which rule ran, what it overrode and what started it. Requests made through Casient are logged when the decision is made, not rebuilt afterwards. Every change is logged.

This is also why the AI is kept apart from the devices. The AI never talks to devices. Fixed code checks every request. No new rule goes live until your installer approves it. If the AI is wrong, offline or switched off, the limits set on the bus still hold.

Three questions for any product

“Who can act on my house, and what checks them?” If an AI can act directly, ask what stops a wrong guess. A good answer names fixed rules, not “the AI is smart”.

“What happens when two automations disagree?” If the answer is “the last one wins” or “you set the order in the rules”, then clashes are left to chance or to careful programming.

“Can you show me why it did that last Tuesday?” If there is no log with the rule that ran, you cannot check it, and nobody can fix it with confidence.

The honest summary

Most home products sold with “AI” are step 2 with a step 3 chat. That is useful, easy to buy and well supported. For many households it is the right purchase. If that fits what you want, buy it.

Step 4 is for a narrower wish. You are tired of telling the house things. You want it to follow room rules you can read, and you want each change approved by your installer. That is what we are working on, and we would rather say so than sell you something you do not need.

For more, read why a bridge is not a brain and the language of your installation. Installers can find fix-it notes at KNX notes.

Send this to your installer, or ask them:

• How does the system decide between two requests?
• Can you run a free Readback on our project?

For installers

Installers: run a free Readback

Want to talk it through? Book 20 minutes with Ted.