One self-confirming realisation I've had after working closely with AI agents: I genuinely love designing systems. So much so that I often spend more time tinkering than actually using the systems I build - but there is one exception.
The first AI 'system' I used was originally built by Aman Khan and its called PersonalOS (link to Github repo). I'll admit, this was also the first Github repo I ever pointed Claude Code at and told it to copy for me. It was like magic. It even has a script that does most of the initial setup for you like a software install wizard.
Over time I've adapted my PersonalOS to meet my specific needs, but haven't had to make any major updates in months (which is a good sign of a simple but effective system). Here's what it does for me:
- Replaces Monday/Planner/Asana as my personal project management tool
- Captures and stores context across my email, meeting transcripts, and documents as markdown files
- Gives me a 'morning standup' summary of my top priority tasks, aligned to my goals.
- All first drafts of documents I create start in my PersonalOS and leverage the combined context that the system stores for me.
This isn't a unique system. You may have seen other 'second brain' type setups that are similar. But the part that was great about this one for me at the time was how accessible it was and relatively easy to just get started with.
At the end of the day, its just a few folders filled with markdown files - but it taught me a lot about how to build context for AI agents to help them be better at helping me without the need to repeat myself.