A small, square screen measuring 240 pixels on each side cycles through three different displays every five seconds. It shows AI tool quotas, room temperature and CO2 levels, and the air conditioner’s energy consumption in watts. The device used is a GeekMagic SmallTV Ultra, a budget-friendly gadget available for about 15 euros on AliExpress. Marketed as a small office weather station, it features an ESP8266 chip and a 240 by 240 pixel screen. Originally, the device could only display a clock and a basic weather forecast. To unlock its full potential, the user spent an hour interacting with an AI agent called Claude Code to add extra functions. The GeekMagic SmallTV and SmallTV Ultra models share the same hardware, including the ESP8266 chip and screen, but differ in the software embedded within them. The more advanced SmallTV Pro uses a more powerful ESP32 chip but is pricier. For the user’s needs, the Ultra was sufficient. To explore the device’s capabilities, a network scan identified it, and accessing its configuration page revealed an HTTP interface. This interface allows users to send images, change themes, and adjust the speed of scrolling text. It is available from the start but not advertised, and it doesn’t require a password. This makes the device accessible to any device on the same network, which can be a concern in shared environments. The AI agent helped write and run scripts that generate images and send them to the screen every two minutes. However, some problems arose that needed manual fixing. For example, the screen sometimes froze and required an undocumented command to refresh, a sensor provided an incorrect data value, and some devices went offline. These issues show that AI still struggles with detecting certain physical and data-related problems. The dashboard cycles through three displays: AI tool quotas, room temperature and CO2 levels, and air conditioner metrics. The CO2 levels are shown in color-coded ranges for easier interpretation. The system was built using Home Assistant, an open-source home automation platform, with assistance from the AI agent. This project shows how low-cost devices can be used effectively in home automation, offering a cheaper alternative to more expensive commercial solutions. While the AI helped speed up development, human oversight was still needed to debug and ensure accuracy. The system also depends on interfaces that are not officially supported, requiring regular updates and troubleshooting.