For the last few years, using AI usually meant opening a browser, sending a request to a cloud service and waiting for the answer to come back. That model is starting to change in two directions at once. AI is moving deeper into the devices around us, while some of the intelligence is also moving back onto hardware we actually own.
Google’s latest move with Google Home is a good example. The company has introduced support for Model Context Protocol, or MCP, allowing third-party AI agents to interact with the smart-home environment rather than relying only on Google’s own assistant. In practical terms, an AI agent could analyse information from compatible cameras, help troubleshoot devices or interact with home systems through natural language. Google has also placed restrictions around sensitive actions such as door unlocking, which shows how quickly the conversation moves from convenience to trust. (The Verge)
That makes the smart home much more interesting.
Until now, most smart-home systems were still fundamentally app driven. Open one app for the camera, another for the lights, another for the thermostat. MCP creates the possibility of an AI agent becoming the interface that sits above all of them.
At the same time, AI is moving in the opposite direction too: away from the cloud and back onto the PC.
Perplexity’s new Portable Computer can run local AI workflows on supported Windows and Linux systems. Instead of sending every task to a remote model, users with powerful NVIDIA RTX hardware can run a local model, work with local files and execute longer multi-step tasks without depending entirely on cloud credits. The catch is hardware: the current setup requires a supported RTX GPU with at least 24GB of VRAM. (Perplexity AI)
That is still enthusiast-level hardware, but the direction matters.
We spent more than a decade moving computing into the cloud because it was easier, cheaper and more scalable. AI may now push part of that workload back onto our own machines because local processing offers something the cloud cannot always give us: lower latency, more privacy and potentially lower recurring cost.
The interesting part is where these two trends meet.
Your PC may increasingly run the AI locally.
MCP may give that AI access to the tools around you.
And the smart home may become less about opening apps and more about giving an agent permission to act.