AI Fluency · The AI In-Basket

What is the AI Fluency Framework?

The AI Fluency Framework was developed in 2023 and 2024 by Rick Dakan (Ringling College of Art and Design, Florida) and Joseph Feller (University College Cork, Ireland). Version 1.1 was published in December 2025 by Ireland’s Higher Education Authority.

In 2025, the two authors created the free course "AI Fluency: Framework & Foundations" with Anthropic, offered through Anthropic Academy. The framework describes the competencies that help people work well with AI, grouped into four: the 4Ds.

The 4Ds in the game

Each competency is practised at a specific moment in the game.

D

Delegation

Délégation

In short

Starting from a clear goal, decide what the person does, what AI does and what they do together.

Its parts

Problem Awareness (understand the goal and the work), Platform Awareness (know what tools can and can’t do), Task Delegation (split the work).

In the game

This is the heart of the game. Picking a bin is deciding whether to delegate. The four questions make you think about the problem before the tool. The made-up number trap is a reminder of what the tool can’t do.

D

Description

In short

Tell AI clearly what you want, to get useful outputs and a useful collaboration.

Its parts

Product Description (the result you expect), Process Description (how to go about it), Performance Description (how AI should behave).

In the game

The "What we said" note prepares the description: what you’d need to tell AI for it to help well. The "With your own AI" option has you use a real request, with an expected result and a precise format.

D

Discernment

Discernement

In short

Critically judge what AI produces, how it got there and whether the collaboration is working.

Its parts

Product Discernment (quality of the output), Process Discernment (logic of the reasoning), Performance Discernment (how AI collaborates).

In the game

The AI’s view arrives after the team’s choice: you judge it instead of following it. When AI invents the tasks, the team also judges whether the tasks ring true.

D

Diligence

In short

Take responsibility for why and how you use AI, and for what you produce with it.

Its parts

Creation Diligence (choose and use tools ethically and legally), Transparency Diligence (be open about AI’s role), Deployment Diligence (check and stand behind what you share).

In the game

The traps around personal information and human relationships. What are you allowed to put into the tool? Should you say that AI helped? Who checks before it goes out?

Three ways to work with AI

The framework describes three modes of interaction between a person and AI. You’ll find them in the bins.

Automation

Automatisation

The person defines a specific task and AI carries it out.

In the game: AI does it.

Augmentation

The person and AI work back and forth as thinking partners and shape the result together.

In the game: AI helps me.

Agency

Agentivité

The person sets up an AI’s knowledge and behaviour so it can act on its own later, sometimes with other people.

In the game: Partly "AI does it": an assistant that answers employees’ questions on its own, with a check now and then, comes close.

"I keep it" and "Eliminate" aren’t modes of interaction. They are delegation decisions: choosing not to hand the task to AI, or not to do it at all. They are often the most useful decisions.

What Anthropic’s data shows

Since February 2025, the Anthropic Economic Index has tracked how people use Claude at work. It sorts conversations into two broad groups: automation, where AI does the task with little human involvement, and augmentation, where it works alongside the person. In the first report, augmentation came first (57%). In September 2025, automation pulled ahead for the first time.

In February 2026, the AI Fluency Index looked for framework behaviours in nearly 10,000 conversations. The most common one: refining the result with AI. But when Claude produced a finished document, people checked facts and reasoning less often. That’s exactly the habit the made-up number trap is trying to protect.

These reports classify conversations after the fact. The AI Fluency modes describe how people choose to work. They complement each other, but they’re not the same grid.

Sources

Play now