# The 4D framework: how to stop prompting like a beginner

> Delegation, Description, Discernment, Diligence: four habits that explain why some developers get great results from AI and others just pull the lever and pray.

Source: https://devaiper.com/blog/4d-framework-ai-fluency
Published: 2026-10-08
Topics: AI fluency, Prompt engineering, Productivity

**Short answer:** When AI gives you something bad, ask which D failed: what you delegated, how you described it, how you checked it, or how responsibly you shipped it. Nine times out of ten the problem is upstream of the model.

You type a prompt, hit enter, and hope. Three matching functions? Great. Nonsense? Pull the lever again. That is a slot machine with a very expensive subscription, not a workflow.

People who are consistently good with AI do something different, and it fits in four words that all start with D. The framework was developed by Rick Dakan and Joseph Feller; the examples here are developer-flavored and mine.

## What is AI fluency?

Not "knowing the best prompt". Prompts expire and models change every month. AI fluency is working with AI **effectively, efficiently, ethically and safely**, and the 4Ds are the habits behind that.

## 1. Delegation: what do I hand over?

The usual mistake is not delegating too little. It is delegating without thinking. Delegation has three parts:

- **Problem awareness**: can you explain the goal to a colleague in two sentences? If not, AI will not fix that. It will produce a confident wrong answer, faster.
- **Platform awareness**: what is this tool good at and where does it fall apart? Strong at drafts, refactors, explanations and tests. Weak on your private code it has never seen and on facts after its training date.
- **Task delegation**: some work is yours (judgment, taste, the final call), some is the AI's (boilerplate, first drafts, the fourteenth unit test), and a lot is shared (you think out loud, it pushes back).

The test: **if this goes wrong, can I tell?** If yes, delegate. If no, keep your hands on it.

## 2. Description: how do I ask?

Three kinds of description, none of them magic words:

- **Product**: what should the output look like? "Write a function" versus "a typed Python function with a docstring, no external libraries, returning `None` on empty input".
- **Process**: how should it get there? "List the edge cases first." "Ask me questions before you write code." You are allowed to tell it to ask you questions. Almost nobody does.
- **Performance**: how should it behave? Direct and critical, not flattering. Explain it like I am a mid-level dev new to Rust. Give me the downsides.

## 3. Discernment: is this any good?

This is what separates an AI-assisted developer from a person who pastes things into production. Judge the output through the same three lenses: is the **product** right (does it run, is the claim sourced), did the **process** make sense (any skipped step or invented API), and did the **performance** follow your instructions.

Description and Discernment are a loop, not two steps. Every bad answer is a free hint about what your prompt was missing. And beware the confident voice: AI does not say "hm, not sure". It says "Certainly!" and then invents a function that does not exist.

## 4. Diligence: am I being responsible?

- **Creation**: choose tools and inputs with care. No customer data or API keys in a tool you have not checked.
- **Transparency**: be honest with your team, clients and readers about where AI helped.
- **Deployment**: verify before you ship, then own it. "The AI wrote it" is not a code review defense.

## Use it tomorrow

Next time AI hands you something bad, do not just regenerate. Ask which D failed. Was it the task you delegated, the way you described it, the way you checked it? You control all three, which is the good news.

## FAQ

### What are the 4Ds of AI fluency?

Delegation (what you hand to AI and what you keep), Description (how you ask), Discernment (how you judge the result) and Diligence (how responsibly you create, disclose and ship).

### Is the 4D framework only about prompts?

No. Prompts are only the Description part. Delegation decides whether AI should do the task at all, Discernment decides whether the output is good, and Diligence covers privacy, honesty and ownership.

### How do I know what to delegate to AI?

Ask: if this goes wrong, can I tell? If you can verify the result cheaply (tests, a quick read, a diff), delegate it. If you cannot, keep your hands on it.