Exploring the Benefits and Frustrations of Running AI Locally

In a personal technology column, the writer describes setting up the Hermes Agent on an M5 Ultra Mac Studio that boasts 256 GB of memory, enabling the use of sizable local language models like the 125-billion-parameter Qwen 3.8 Flash Next. Early experiments involved creating a scripted morning briefing, categorizing a Steam library of over 400 games, and performing financial and laptop-spec analyses, all kept on-device to avoid cloud exposure.

The author notes that while the briefing initially failed due to the Mac sleeping, it now runs reliably, and the Steam organization task completed quickly after granting a temporary API key. Smaller tasks such as data crunching proved useful, yet larger automation projects, like benchmarking laptops, remain work-in-progress. Overall, the piece conveys both the promise of privacy-focused local AI and the practical frustrations that still accompany its deployment.