Persistent memory is becoming a default feature of AI assistants. For creative work, that sounds ideal: remember the visual language, the project history, the preferred tools, the names of recurring characters, and the decisions made three weeks ago.

But a creative assistant also encounters drafts, guesses, contradictions, abandoned ideas, generated text, and instructions that were only true for one project. If every interaction becomes an equally trusted memory, continuity quickly turns into distortion.

Four fields every useful memory needs

provenanceWhere did this come from? User confirmation, file metadata, an imported document, or an AI inference?confidenceHow certain is the system, and who is allowed to change that confidence?lifecycleShould this persist forever, expire after a project, or be reviewed periodically?correctionCan the user inspect, edit, supersede, or delete it without fighting the system?

A simple record might look like this:

memory: Steven prefers dark editorial visuals | type: preference | source: user-confirmed | confidence: 1.0 | scope: global | expires_at: null

What this changes for creator tools

Memory should be treated as part of the product interface, not an invisible vector database. A project assistant should show why it made a suggestion, distinguish a confirmed preference from a pattern it inferred, and make it easy to say: “That was true for the last campaign, not forever.”

The practical direction is clear: typed memory records, explicit provenance, scoped retrieval, reversible updates, and human checkpoints for anything that changes a durable project truth.

The goal is not to make an assistant remember everything. The goal is to make it remember the right things, for the right reasons, for the right amount of time.