Your "smart" home isn't smart. The agentic home is.

I'm going to say something that might annoy a few people in this industry: most "smart" homes built in the last decade aren't actually smart. They're automated. And there's a real difference, one that's becoming impossible to ignore right now.

For most of the industry's history, "smart home" meant one of two things. You set up a schedule, or you gave a command. Lights off at 11pm because you programmed that. Aircon starts at 4pm because you told it to. Gate opens when you press the button on your phone. Convenient? Yes. Intelligent? Not really. The home is still waiting for you to make a decision. It just executes faster.

What's arriving now is something else entirely. People in the industry are calling it the "agentic home."

The shift from reactive to proactive

The concept is straightforward to describe and surprisingly hard to build properly. A smart home reacts to what you do. An agentic home understands what you need and acts without being asked.

The difference is agency. An AI agent monitors the state of the home, reads context from sensors and calendars and weather data and your own patterns, and makes decisions. Not "turn on the light when motion is detected": that's still just a rule. More like: the system notices it's Friday at 6pm, you're usually home by 7pm, it's cold tonight, your gym bag is by the door, and the last three Fridays you ran a bath when you got in. So it starts warming the geyser at 6:45pm. No instruction required.

This is the shift the industry is moving toward right now. Amazon spent most of 2025 and early 2026 rebuilding Alexa from a voice assistant into what they're calling an agentic platform. Alexa+ can handle multi-step tasks, remember your preferences across sessions, and coordinate actions across different devices without you having to explain each one separately. Homey, another automation platform, shipped a version that opens its system to let outside AI agents actually control connected devices. These aren't experiments. They're mainstream products moving in the same direction at the same time.

The reason it's happening now is simple: AI has gotten good enough to be genuinely useful at this, and the cost of running it has dropped enough to make it practical.

Why this only works on a reliable backbone

Here's where I'll get specific about something that the mainstream tech press mostly ignores.

AI agents are only as good as the data they have access to. And the data is only as good as the infrastructure feeding it.

Most consumer smart home setups run on Wi-Fi and Zigbee and Z-Wave. Those are fine for basic use. But when you start giving an AI agent responsibility for making real decisions about your home, you need clean, consistent, real-time data from every device. You need to know that when a sensor reports a state change, it actually happened, right now, not 30 seconds ago when the mesh reconnected. You need device control that is deterministic: when you tell a light to dim to 40%, it goes to 40%, not somewhere between 35% and 45% depending on whether the mesh caught the command.

KNX is a wired bus protocol. It doesn't have the packet loss problems of wireless. It doesn't have the mesh routing delays. State changes are immediate and reliable. When an AI agent queries the state of a KNX thermostat or a shutter actuator or a DALI lighting zone, it gets an accurate answer. When it sends a command, that command executes.

This is why I keep coming back to KNX as the backbone for high-spec residential projects. Not because it's the only way to automate a home. But because when you want the home to actually think (not just execute), you need infrastructure that can be trusted completely. You can't have an AI agent managing a climate system if it can't reliably read the sensor data or trust that its commands landed.

Running locally matters, especially in South Africa

Most agentic home systems being built right now assume constant cloud connectivity. Alexa+ phones home to Amazon. Google's AI home control routes through Google. That works fine until your internet connection drops, or Eskom decides it's Stage 4 tonight, or your LTE signal is having one of those afternoons.

We run our AI locally. Home Assistant on a local server, combined with local language models (we use Ollama with Qwen), means the automation intelligence sits inside the building. It doesn't need to reach AWS data centres in Ireland to decide whether to turn on your backup lighting or adjust a thermostat. It decides, immediately, on your network, with or without internet.

For a South African home with a Victron inverter setup and solar, this is the logical endpoint: a house that knows its own battery state, its own grid status, its own occupancy, its own energy consumption pattern, and makes decisions accordingly. Not because you programmed 200 automations. Because the AI agent understands the system and responds to conditions as they change.

What this actually looks like

Someone walks in the front door at 7pm carrying groceries. The camera picks them up. The system recognises it's a weeknight. The kitchen lights come up. The counter power sockets activate. The extractor comes on at low speed ready for cooking. The lounge dims to a comfortable evening level. Nobody pressed anything. Nobody spoke to any device.

That's an agentic home. Not magic, just a system with enough context to make a sensible decision.

Getting there requires a reliable physical layer (KNX), a capable integration platform (Home Assistant), good sensors, and an AI layer that can reason across all of it. The 1Home gateway already bridges KNX into Home Assistant and Matter. That's the architecture. The AI sits on top of a system that was built properly from the start.

What changes for architects and specifiers

The architecture decisions you make at design stage are becoming more consequential. An agentic home needs dense sensor coverage: presence, lux, humidity, CO2, door/window states. It needs a physical infrastructure that can carry reliable real-time data. These things are easy to incorporate at first-fix. They're painful to retrofit.

If you're specifying a project now for a client who wants their home to actually anticipate their needs rather than just respond to them, the conversation about what goes in the walls needs to happen in the design meeting, not six months later on site.

Worth a conversation if you're in that process. No pitch, just a discussion about what the project needs.

wayne@knxlogic.co.za | 082 564 3982 | www.knxlogic.co.za