How AI broke the smart home in 2025

In 2025, the promised revolution of generative AI in the smart home, exemplified by upgrades like Alexa Plus, has resulted in assistants that are more conversational but significantly less reliable at fundamental tasks like controlling lights or running routines. Experts suggest that Large Language Models (LLMs) were not designed for the precise, repetitive nature of command-and-control systems, introducing unwanted stochasticity (randomness) and complexity when trying to integrate them with existing device APIs. While companies are deploying these early access versions to gather data and improve, users are essentially beta testers dealing with inconsistent performance. The trade-off is between the limited accuracy of old systems and the vastly expanded potential of new LLM-based agents, which companies prioritize for their ability to chain complex services, even if basic functionality suffers in the interim.

(Source：The Verge)