AI will remain. Your access conditions may not.

AI is unlikely to disappear. It is more likely to become ordinary infrastructure: something people rely on for writing, research, planning, programming, communication, and daily work.

The problem is not simply whether AI exists. The problem is whether one person can continue using it under conditions they can understand and control.

Models are retired. Prices and usage limits change. Accounts can be restricted. Interfaces disappear. Regional availability changes. A provider can decide that an old workflow no longer matters. A user may be able to export some files while still losing memory, permissions, project context, or the exact working environment that made those files useful.

Better models do not solve this continuity problem. A stronger replacement may still know nothing about the user's confirmed memory, preferences, policies, projects, previous decisions, source records, local artifacts, or recovery procedures.

doll is intended to place those durable parts outside the model. A model may be local or remote, large or small, preferred or merely available. Its capability may change. The user's state should not have to disappear with it.

This does not mean every model will behave the same way. It does not guarantee permanent access to frontier-level performance. It means that reduced performance should not automatically become total loss.

The intended end state is a system that can continue locally, switch reasoning engines, preserve confirmed state, audit tool use, restore from verified backups, and degrade to a smaller fallback when necessary.

doll is therefore not a bet on a future in which AI goes away. It is preparation for a future in which AI remains, while the conditions required to keep using it remain unstable and externally controlled.

Origin

This note began during planning discussions for doll. It was selected, reviewed, and published by badjoke-lab because it describes the reason the project exists.