DevOps engineer: Hermes Agent build cost him the thrill

The author grounds his argument in a before-and-after comparison from his own blog. Almost a year ago, on 25/10/2025, he wrote about building a "Self Monitoring Project" — spending a couple of days just migrating data from Google Sheets to InfluxDB and adding a simple web page for entering new data. Even then, he says, he did it only with ChatGPT or Claude.ai, discussing functions and running commands himself.

His new "PersonalAI Health Agent" does more: it imports data from Google Sheets into his self-hosted VictoriaMetrics, plus data from Google Calendar and his own journal, with analysis and reports. That was built in 2 days with Hermes Agent and GPT 5.5, and 99% of it was done by the agent. What struck him was not the speed but the absence of the old reward loop — task, confusion, search, wrong idea, understanding, solution — and the fact that he cannot fully explain why the system was built the way it was. He says the most he can rely on are README.md, AGENT.md and CLAUDE.md files, and asking Codex to explain.

He frames the stakes as a trade-off rather than a verdict, arguing that agents are now the new reality and genuinely help when production is down. What the outcome hinges on, in his telling, is whether engineers deliberately keep some manual work — choosing tasks by how critical a mistake would be — so they preserve the understanding, control and motivation that make them engineers at all.