A good programmer looks both ways even when crossing a one-way street.
Over the last few months it has become obvious that tomorrow's programmers will do completely different things from what they did yesterday. I believe future programmers will not be coders (people who write code, as they do now) but true engineers.
For the last few days I've been working with Claude Code to set up the Hermes agent on a brand-new Mac mini. I want to play with this piece of art (why not OpenClaw? Just a religious choice).
I've been setting up workflows. One of them I've used in Claude Code for a few months: a humanitarian project that buys things in shops and tracks the logistics. It visits a few shops and other websites, so the agents have to deal with passwords, spreadsheets, and partly personal data.
I'm setting up another profile in Hermes, and it's kind of another person. I set up Matrix chat so I can talk to the agents. I didn't want to do it in Telegram for security reasons.
Claude Code wrote the first versions of the scripts from my directions, and Hermes adapted them. There was nothing dangerous in them. But one agent could access the files of another agent. I don't think it's okay, and this is what I, as an engineer, have to think about.
While Claude Code is a great agent, it's not ideal and can make some stupid mistakes. This is where an engineer's mindset comes in. Claude Code just didn't think much about profile isolation.
I ended up creating a few different macOS users, and I'm testing this setup now. The users are isolated on my Mac mini, and I'm even more paranoid: I created a separate Apple ID for this Mac mini, so if Hermes goes wild, it has less access.
I've also started using a 1Password service account. I'll tell you how it works in future posts.
Do you use personal agents like Hermes or OpenClaw? If so, what are they doing for you?
First published on LinkedIn.
Discuss on LinkedIn