Generation is excellent at breadth and adaptation
An AI tutor can turn one Frame into a workplace example, a travel dialogue, a role-play or a short quiz. It can change difficulty and respond to a learner’s question immediately. Those are strong affordances because language use is contextual and learners do not all need the same scenario.
Metkagram can benefit from that flexibility when generated activity is anchored to a known Frame, Move or Contrast. The model expands the practice surface; the reviewed object gives the expansion a stable reference point.
A curated corpus solves a different problem
A stable Frame can have an identifier, reviewed constraints, examples, relationships, quality metadata and provenance. The same object can appear in a web page, an API, a teacher export or an AI session without silently changing its definition between requests.
That stability matters for learning history and research. If a learner repeatedly confuses two Frames, the system can refer to the same Contrast tomorrow instead of asking a model to reinvent the distinction from scratch.
The strongest combination separates content authority from tutoring behaviour
An AI tutor can choose a prompt, personalise a situation, ask for retrieval and explain feedback. The canonical corpus can define which Frame is being practised and which relationships have actually been reviewed. This division allows creativity at the interface without pretending that every generated linguistic claim is equally verified.
Metkagram’s architecture follows that principle already: AI can use the corpus, but it does not become the editorial source of truth merely because it can write fluent examples.