After setting up a local AI agent named Hermes and reading back through his execution diaries, I realized something important.
I had assigned grand titles to my agent network—Coordinator, Engineer, Librarian, Researcher—but in reality, the setup was just chaotic. As a layperson, the process felt overwhelmingly complex. The agents spent a lot of time guessing my intent, burning through tokens with no clear direction, and constantly throwing out suggestions without real verification.
I found myself blindly approving prompts and making choices without fully understanding the consequences. Every AI recommendation sounded completely logical, but knowing how easily large language models can hallucinate, any of it could have been totally wrong.
In the hands of an expert, these tools are powerful. In the hands of a beginner, they can easily create chaos and offer a false sense of security.
Having said that, I can’t deny the results. The agents successfully helped me finish my OpenMediaVault installation and upgrade my network to 2.5G and 10G by finding and installing the correct drivers. They spun up multiple services in separate LXC containers and successfully configured Nextcloud—something I had previously spent hours failing to do on my own.
It has been an incredible playground. It perfectly demonstrated what it means to run local LLMs and how to set them up. I know for a fact I could not have accomplished this without Hermes.
Yet, I am still left with a massive question mark: does this actually lead to anything productive, and could I ever reliably repeat this success next time?