The smart home spent a decade failing at interoperability. Matter and Thread fixed most of the plumbing, and that changed the interesting question: not whether devices talk to each other, but what the system does with what they say.
What the standards actually solved
Matter gave devices a common application layer, so a bulb from one brand works with a hub from another. Thread gave low-power devices a mesh network that survives a router reboot. Together they removed the main reason people gave up: a house full of apps that never spoke.
Where AI enters
Three places, in order of maturity:
- Anomaly detection: a pump running at 3am, a door open longer than usual, a freezer warming slowly. This works today and is genuinely useful.
- Natural language control: asking for a scene instead of programming one. Reliable for simple requests, brittle for compound ones.
- Prediction: pre-heating based on your calendar. Still the weakest, because it is wrong often enough to be annoying.
The local-first rule
Every automation that must work when the internet is down should live on a local hub. Cloud AI can add convenience, but a house that cannot turn on a light during an outage has been built backwards. Look for hubs that keep schedules and scenes on-device.
What to buy first
Start with the boring layer: a Thread border router, a local hub, and sensors that report state rather than stream video. Add cameras last, because they are the most privacy-sensitive and the most cloud-dependent part of any setup.
What is still marketing
Any claim that the house will anticipate your needs without you configuring anything. Detection is real. Anticipation is mostly a demo, and it usually requires more data than a sensible person wants to hand over.
Bottom line: the standards made the smart home work. AI makes parts of it smarter, but only where it observes and alerts. The moment it starts deciding, verify it against a manual switch.
