What is the 4D AI Fluency Framework?

The Four Competencies

Delegation is deciding whether, when, and how to involve AI at all, and what to keep for yourself. It is the competency that happens before any interaction: choosing which part of a task is a good fit for a system and which part needs human judgement, relationships, or accountability that shouldn't be handed over.

Description is communicating the task to the AI well enough that it can act usefully — supplying the goal, the constraints, the context, and the format. Prompting lives here. Note that: prompting is a set of skills inside one of the four competencies, not the whole of fluency.

Discernment is evaluating what comes back — judging quality, accuracy, and bias, and noticing when a fluent output is wrong, incomplete, or merely plausible. This is the competency that turns AI output into something you can actually rely on, rather than something that sounds right.

Diligence is taking responsibility for what you do with AI and how: transparency about its use, accountability for the result, and the ethical and practical judgement that you own the output once it carries your name.

The framework's own definition of AI fluency ties these together: the ability to work with AI effectively, efficiently, ethically, and safely. Each adverb maps onto the competencies — and each is a human responsibility the framework declines to hand to the machine.

Read the four together and a pattern is hard to miss: only Description is about interacting with the system. Delegation, Discernment, and Diligence are about deciding, evaluating, and taking responsibility for what happens around that interaction. Three of the four competencies are human judgement, not machine operation. That is the deeper reason prompting gets demoted — not because it's unimportant, but because the framework's authors locate most of the work outside the prompt entirely.

Why "fluency" and not "literacy"

The word choice is deliberate. Literacy suggests a threshold you cross: you can read, or you can't. Fluency is a practice you deepen, and it degrades without use. The framework's authors chose it to make a point that matters for anyone planning training: there is no certificate that makes someone done. The competencies are exercised, not completed.

This is also why the framework is model-agnostic by design. It describes the human side of the collaboration, which is stable, rather than the tool's behaviour, which changes every few months. A framework built around today's interface would need rewriting at the next capability jump. One built around Delegation and Discernment does not, because deciding what to hand over and judging what comes back are constant regardless of how good the system gets, and arguably more important as it gets better, because better output is more persuasive when it's wrong.

The Loop Most People Miss

The four Ds are not a sequence you complete once. Description and Discernment form a loop: you describe, the system responds, you discern what's off, you re-describe, it improves. The actual work of using AI well lives in that cycle, not in a single well-crafted prompt.

This adds a mechanism to the demotion. If prompting is a skill within Description, and Description only does useful work in tension with Discernment, then a clever prompt evaluated by someone who can't tell good output from plausible output is worth very little. The prompt is not the skill. The judgement applied to what the prompt produced is the skill, and it is the half that can't be copied from a prompt library.

That is the framework's quiet argument against the entire prompt-tips economy: the transferable artefact (the prompt) is usually the cheaper half, and the non-transferable capability (the discernment) is the expensive half. Most AI training sells the cheaper half because it is the half you can package.

The Three Modes

The framework also names three modes of working with AI, and the competencies apply differently across them. Automation is the AI executing a specific instructed task. Augmentation is human and AI working as thinking partners. Agency is configuring AI to act on your behalf on future tasks you won't individually supervise.

The modes matter because Delegation and Diligence get harder as you move along them. Delegating a one-off task is low-stakes; the output is in front of you. Configuring an agent to handle a class of future tasks means exercising Delegation for situations you can't see yet and Diligence for outputs you won't individually review. The framework's competencies don't change across the modes, but the cost of being weak at them rises steeply. This is the point where individual fluency runs into its own ceiling — which is the next thing worth saying plainly.