The AI Fluency Framework
B · Useful, limited One-page summary of the 4D AI Fluency Framework by Rick Dakan, Joseph Feller and Anthropic; a clear, reusable competency model for training, with no evidence behind it on this page.
Summary based on the full text·Original: about 2 minutes to read
In short
A single-page summary of the AI Fluency Framework (2025), released under CC BY-NC-SA 4.0 by Rick Dakan, Joseph Feller and Anthropic. It defines AI fluency as working with AI systems in ways that are effective, efficient, ethical and safe, and breaks it into four competencies, the "4Ds". It also names three modes in which people work with AI.
Key numbers
- Fourinterconnected competencies: Delegation, Description, Discernment, Diligence
- Threemodes of AI interaction: automation, augmentation, agency
As reported by the source.
English·PDF·opens www-cdn.anthropic.com
What it reports
- Delegation: setting goals and deciding whether, when and how to use AI.
- Description: describing goals well enough to get useful outputs and behaviour.
- Discernment: judging how useful the outputs and behaviour are.
- Diligence: taking responsibility for what is done with AI and how.
- Three modes: automation (AI executes a task on instruction), augmentation (human and AI work together as thinking partners), agency (humans configure AI to perform future tasks on their own).
- Competencies are described as bundles of skills, knowledge, insights and values.
How they got it
A summary page for a framework developed by two academics with Anthropic for AI fluency courses. No research method, data or validation is presented on this page.
Limits
- One page: definitions only, without the course material, examples or assessment rubrics.
- No evidence that the four competencies predict better outcomes.
- Co-authored by an AI vendor; the corpus index had mislabelled this file as a labour-market paper.
- Non-commercial licence restricts reuse in paid training.
Why it matters for companies
The 4Ds give learning teams a simple vocabulary for AI skills that goes beyond prompting: deciding what to hand over, and checking and owning the result, are two of the four. The three-mode split is useful for role design, because the skills for supervising an agent differ from those for co-writing with a chatbot. Companies would still need their own assessments to measure fluency against this model.
Summary based on the full text. Read the original for the full argument and data.Open PDF on www-cdn.anthropic.com