The AI Fluency Framework

Originally published on LinkedIn, 9 October 2025. Republished here with light editing.

The AI Fluency Framework outlines four interconnected competencies necessary to ensure interactions with AI are effective, efficient, ethical, and safe. These competencies form the "4D Framework": Delegation, Description, Discernment, and Diligence. AI Fluency encompasses practical skills, knowledge, insights, and values that enable adaptation to evolving AI technologies.

The 4D Framework Competencies#

The 4Ds are interconnected collections of skills, knowledge, insights, and values.

1. Delegation#

Delegation involves making thoughtful decisions about what work is appropriate for humans, for AI, or for humans and AI to do together, and how to distribute those tasks. Effective delegation requires both domain expertise and an understanding of AI capabilities.

Key components of Delegation include:

• Problem Awareness: Clearly understanding your goals and the nature of the work before involving AI. This addresses the question, "What am I trying to do?".
• Platform Awareness: Understanding the capabilities and limitations of different AI systems.
• Task Delegation: Thoughtfully distributing work between humans and AI to leverage the strengths of each.

2. Description#

Description is the ability to communicate with AI in ways that create a productive collaborative environment. Clear communication with AI systems upfront saves time and leads to better results. It involves effectively describing goals to prompt useful AI behaviors and outputs.

Key components of Description include:

• Product Description: Defining what you want in terms of outputs, format, audience, and style.
• Process Description: Defining how the AI approaches your request, such as providing step-by-step instructions for the AI to follow. An example is instructing the AI to "Act as a Socratic tutor".
• Performance Description: Defining the AI system’s behavior during your collaboration, such as whether it should be concise or detailed, challenging or supportive.

3. Discernment#

Discernment is the ability to thoughtfully and critically evaluate what AI produces, how it produces it, and how it behaves. This includes assessing quality, accuracy, appropriateness, and determining areas for improvement. Discernment works hand-in-hand with Description in a continuous feedback loop.

Key components of Discernment include:

• Product Discernment: Evaluating the quality of what AI produces, considering factors like accuracy, appropriateness, coherence, and relevance.
• Process Discernment: Evaluating how the AI arrived at its output, looking for logical errors, lapses in attention, or inappropriate reasoning steps.
• Performance Discernment: Evaluating how the AI behaves during your interaction, considering whether its communication style is effective for your needs.

4. Diligence#

Diligence is taking responsibility for what we do with AI and how we do it. This involves using AI responsibly and ethically.

Key components of Diligence include:

• Creation Diligence: Being thoughtful about which AI systems you use and how you interact with them. For example, acknowledging the specific AI model used, like "In this document, Claude 3.7...".
• Transparency Diligence: Being honest about AI's role in your work with everyone who needs to know.
• Deployment Diligence: Taking responsibility for verifying and vouching for the outputs you use or share. Different contexts (personal, academic, professional) may have different expectations for disclosure and verification.

Key Concepts: Three Modes of AI Interaction#

AI Fluency involves interacting with AI systems in ways that are effective, efficient, ethical, and safe. The framework identifies three primary modes of human-AI interaction:

• Automation: AI executes specific tasks based on human instruction. The human defines what needs to be done, and the AI executes it.
• Augmentation: Humans and AI collaborate as thinking partners to complete tasks together. This involves iterative back-and-forth where both contribute to the outcome.
• Agency: Humans configure AI to independently perform future tasks on their behalf, including interacting with other humans or AI. The human establishes the AI's knowledge and behavior patterns rather than specifying exact actions.

AI Technical Concepts#

• Generative AI: Artificial intelligence systems that can create new content (text, images, code, etc.) rather than just analyzing existing data. Examples include writing new emails, whereas traditional AI might classify emails as spam.
• Large Language Models (LLMs): Generative AI systems trained on vast amounts of text data to understand and generate human language.
• Claude: Anthropic's family of large language models.
• Parameters: The mathematical values within an AI model that determine how it processes information and relates different pieces of language to each other. Modern LLMs contain billions of parameters.
• Neural Networks: Computing systems similar to, but distinct from, biological brains. They are composed of interconnected nodes organized in layers that learn patterns from data through training.
• Transformer Architecture: The breakthrough AI design from 2017 that enables LLMs to process sequences of text in parallel while paying attention to relationships between words across long passages.
• Scaling Laws: An empirical observation that as AI models grow larger, are trained on more data, and use more computing power, their performance improves consistently. New capabilities can emerge at certain scale thresholds.
• Context Window: The amount of information an AI can consider at one time, including conversation history and shared documents. It has a maximum limit that varies by model.
• Hallucination: A type of error where AI confidently states something that sounds plausible but is actually incorrect.
• Knowledge Cutoff Date: The point after which an AI model has no built-in knowledge of the world, based on its training completion.
• Reasoning or Thinking Models: Types of AI models specifically designed to think step-by-step through complex problems, showing improved capabilities for tasks requiring logical reasoning.
• Temperature: A setting that controls how random an AI's responses are. Higher temperature produces more varied and creative outputs, while lower temperature produces more predictable and focused responses.
• Retrieval Augmented Generation (RAG): A technique that connects AI models to external knowledge sources to improve accuracy and reduce hallucinations.
• Bias: Systematic patterns in AI outputs that unfairly favor or disadvantage certain groups or perspectives, often reflecting patterns in training data.

Prompt Engineering Concepts and Techniques#

• Prompt: The input given to an AI model, including instructions and any documents shared.
• Prompt Engineering: The practice of designing effective prompts for AI systems to produce desired outputs. It combines clear communication with AI-specific techniques.
• Chain-of-Thought Prompting: Encouraging an AI to work through a problem step-by-step, breaking down complex tasks into smaller steps to help the AI follow your thinking and deliver better results.
• Few-Shot Learning (N-Shot Prompting): Teaching AI by showing examples of the desired input-output pattern. The "N" refers to the number of examples provided, helping the model understand what you want without lengthy explanations.
• Role or Persona Definition: Specifying a particular character, expertise level, or communication style for the AI to adopt when responding. This can range from general roles (e.g., "speak as a UX design expert") to specific personas (e.g., "explain this like Richard Feynman would").
• Output Constraints/Output Formatting: Clearly specifying within your prompt the desired format, length, structure, or other characteristics of the AI's response to ensure you get exactly what you need.
• Think-First Approach: Explicitly asking the AI to work through its reasoning process before providing a final answer, which can lead to more thorough and well-considered responses.

6 Techniques for Effective Prompt Engineering:#

1. Provide context: Be specific about what you want, providing details about scope, geography, timeframe, and other relevant parameters.
2. Show examples of what "good" looks like: Providing examples helps the AI understand the pattern, style, or format you're looking for more clearly than descriptions alone. This is also known as Few-Shot Learning.
3. Specify output constraints: Clearly define the desired format, length, structure, or other characteristics of the AI's response.
4. Break complex tasks into steps: Guide the AI's reasoning process by breaking down complex tasks into clear, methodical steps. This is also known as Chain-of-Thought Prompting.
5. Ask it to think first: Encourage more thoughtful, comprehensive answers by giving the AI space to process before responding. This is also known as the Think-First Approach.
6. Define the AI's role: Shape the AI's approach to fit your specific needs and audience by defining its role, tone, or style. This is also known as Role or Persona Definition.

Secret Weapon: If you're unsure how to phrase a request, you can ask the AI for help with prompting to craft an effective prompt for your goal.

Current Limitations of Generative AI#

Despite their capabilities, Generative AI systems have limitations:

• Knowledge cutoff date: Their knowledge is limited to their training data, which has a specific cutoff.
• Potential inaccuracies ("hallucinations"): They can confidently state incorrect information.
• Context window constraints: The amount of information they can process at one time is limited.
• Challenges with complex reasoning and math: They may struggle with tasks requiring deep logical reasoning or precise mathematical calculations.