I've been using Docker volumes to maintain per-project Claude configuration — a ~/.claude folder with all the session history and context for each project.
The theory was solid: isolate each project's AI context. Keep everything clean and project-specific. No context pollution from one project leaking into another.
Turns out the session storage only keeps the last 30 days. So when I went back to a project from November after taking a break, the entire conversation history from my earlier work was gone.
It's a small problem with a big lesson. When you're working with multiple projects and AI, you lose institutional memory quickly. The AI starts fresh each time. It doesn't remember the decisions you made last month, the architecture you chose, the gotchas you discovered.
That's not necessarily bad — fresh perspective has value. But it also means you need better documentation. You need architecture decision records that survive beyond 30 days. You need README files that capture the thinking, not just the code.
The future of AI-assisted development isn't just about speed. It's about building systems that remember why they were built the way they were.
Part of the #100DaysToOffload documenting agentic development in 2026