Daniel runs persistent AI agent workspaces across his own projects — travel planning, a research archive, tracking what he owes people by email — daily on Claude Code, Codex, and Gemini CLI, for months at a time.
This is a systems talk rather than a coding talk: there’s code underneath, but he writes almost none of it by hand. The work is defining end goals and the few operating rules that keep a workspace from rotting over months.
He’ll demo the real thing live, including what it costs to keep running and the ways it’s broken on him anyway