Smart home: the truth we would rather not hear
My talk at VarnaConf 2026. Why the smart home is not smart yet, what it is missing and when it should ask the person.
On 22 August I spoke at VarnaConf 2026 with the talk “Smart home: the truth we would rather not hear” („Смарт хоум: Истината, която не ни устройва“). It follows on from the talk in Burgas, which was about how a system is built from a single sensor. In Varna the argument was that even a well-built system rarely deserves the word “smart”.
Automation is not intelligence
Today’s smart home is full of connected devices: lamps, sockets, air conditioners, locks. Most of them, though, wait for a command. The person opens an app, picks a room and a device and presses a button, or says a voice command. That is convenient, but the decision is still theirs. The system only carries it out.
The automations that look smarter usually tie an event to a predefined action: on motion after 18:00, the lamp switches on. The rule is predictable and easy to check, but it does not know why there is motion. If someone gets up for a glass of water at 23:30, the rule will light the hallway and the kitchen at full power. It is carried out exactly, and the result is wrong, because the system sees the event but not the situation.
What is missing
The classic smart home is missing three things:
- Memory: what usually happens in this home, on this day and at this hour.
- Context: what is happening right now. Is anyone home, are they asleep, are there guests, what is the temperature outside.
- Prediction: what is likely to happen next.
A home that has them would react to the same motion at 23:30 differently: it would turn on only the hallway light, at low power, for two minutes. The reactive chain “event, rule, action” is replaced by a closed loop. The home observes, understands the situation, predicts, chooses an action and then checks whether it was right. Without that check the system does not learn, and it is left with rules that go stale over time.
When the home should ask
The hardest question in the talk was what the system should do when it is not sure. An example: the heating usually starts at 07:00, but today the sensors see nobody at home, and the system is only 58% sure. If it acts on its own with that much uncertainty, it will often be wrong. If it asks about everything, it becomes annoying and people stop using it.
The sensible approach is a confidence threshold. Above the threshold the home acts on its own, below it the home asks the person. Critical actions always go through the person, and the safety rules act immediately, whatever the model thinks.
You can try this below. Pick a situation, compare what the rule does with what the home with context does, and move the threshold to see when the system becomes too bold or too annoying.
The best interface for a smart home is one you rarely need to use. When the situation is clear, the home acts. When it is not sure, it asks. When there is risk, the person decides.
Edge and cloud
A home collects some of the most personal data a person creates: when they sleep, when they go out, whether anyone is home, what is being said. If a decision can be made on the spot, there is no reason for the raw data to leave the home. A local decision is also faster, and it keeps working when the internet goes down. Today, small models for forecasting, anomaly detection and recognition run on a small computer at home.
The cloud is still useful for heavy models, updates and long-term analysis. So the architecture is hybrid: fast, personal and critical decisions are made locally, and the rest can go to the cloud.
Security comes first
The more decisions a home makes on its own, the more it costs if someone else takes control of it. The vulnerabilities of smart devices are the subject of our paper on a smart camera from TELECOM 2025. It is one of the reasons I keep as much as possible local.
A home that really helps
Comfort is the easy part. The real test of whether a home is smart is whether it helps a person who needs help. One idea we are working on is a small device for a blind person that reads text aloud and recognises objects entirely locally, without the pictures ever leaving the home. In my view, that is the direction in which the smart home makes the most sense.