I'm building a personal AI assistant called Donna to help manage my businesses, projects, daily priorities, and meetings. I use ChatGPT for planning and conversations, with project information stored in my GitHub-based AI Workspace.
What I'd want my AI teammate to remember:
My current projects, goals, decisions, working preferences, and where we left off. Just as important, I'd want Donna to consistently follow our established workflows without needing reminders.
Where this would help me:
Our daily scrum. Donna prepares by reviewing my calendar, Trello tasks, project status, and previous commitments. During our conversation, she should use that preparation to guide us through the complete checklist.
The biggest benefit:
Less time repeating context or reminding Donna about steps she's supposed to follow. We've made progress, but consistency across sessions is still a challenge.
What stood out from the lesson:
The distinction between CLAUDE.md, SOUL.md, and SKILL.md made me think about separating project instructions, an assistant's communication style, and reusable workflows. I'm already using structured project instructions and workflows, but I can see opportunities to improve how they're organized and loaded.
One question I'm left with:
How do we let an AI teammate learn from experience without allowing its instructions or behavior to drift away from what we've approved?
For me, the ideal AI colleague isn't one that seems more human. It's one that remembers the right context, follows through reliably, and gets better through reviewed, measurable improvements.