I’ve been using ChatGPT for increasingly long-term and interconnected work, and I’ve repeatedly run into a problem: my work doesn’t really organize itself into conversations.
I initially thought this was mainly an organizational problem- something between Projects and individual chats, perhaps with a small visual map. But as I worked through it, I came to think the organizational problem and the context problem are closely related.
The useful thing to persist isn’t necessarily the conversation. It’s the current state of the thing being worked on: what is current, what changed, what was superseded, what remains unresolved, what sources support it, and what other work it relates to. Conversations can then remain the history of how that state developed.
I’ve since read through this forum and found several proposals for nested Projects, persistent project state, improved history retrieval, knowledge graphs, and better continuity. So I don’t think the individual pieces of this idea are unique- in fact, the amount of similar feedback seems significant.
What I haven’t seen expressed quite the same way is bringing those pieces together around a persistent layer above conversations: Projects/domains → persistent workstreams → current state → conversations and sources.
ChatGPT would maintain most of that structure automatically and select the context relevant to the task rather than simply relying on everything it can remember or retrieve. The user could optionally inspect the structure- perhaps through a small mind-map or outline view- and correct, isolate, move, or remove things when ChatGPT’s interpretation is wrong.
The goal isn’t maximum memory or a manually maintained knowledge graph. It’s reliable continuity with the minimum sufficient context and visible user control.
I’ve attached a one-page visual that lays out the model, potential benefits, and the design principles I arrived at. I’d be interested in whether others who use ChatGPT for long-running work see the organizational problem and the context problem as parts of the same underlying issue.