AI Fluency
Also: 4D Framework, Delegation Description Discernment Diligence
Anthropic's framework for working with AI, taught across Claude Academy and built on four practices it calls the 4Ds. **Delegation** is deciding what to hand to a model and what to keep. **Description** is briefing it well enough to get the output you meant. **Discernment** is judging what comes back. **Diligence** is verifying in proportion to the stakes — the single most portable idea in the set. AI fluency sits one level above AI literacy: literacy is knowing what a model can do, fluency is having a repeatable practice for working with one. The framework is deliberately about durable habits rather than product features, which is its strength. It is also scoped to one person working with one model, so it says nothing about the organisational problems — budget, adoption, procurement, failure at scale — that consume most of a real deployment.
In practice
A analyst applying the 4Ds to a quarterly report: delegates the first-pass summarisation but keeps the interpretation; describes the audience and format up front instead of iterating five times; reads the output looking for the specific claims that would embarrass them if wrong; and verifies every number that goes in front of the board while accepting the prose as drafted. That last split — heavy verification on figures, light on wording — is diligence proportional to stakes.
Related concepts