What does AI Fluency mean?

The ability to use AI in ways that are efficient, effective, ethical, and safe. AI Fluency is about how well you work with AI, not how many tools you use, how much you automate, or whether you can build models yourself.

What are the four competencies in the 4D Framework?

Delegation, Description, Discernment, Diligence. The same 4 Ds from AI Fluency: Framework & Foundations, applied to building software. The other options are software lifecycle stages, not AI Fluency competencies.

According to the Builder's Toolkit, where is AI weakest?

Empathy, Judgment, and Shipping. AI is strong at producing code, tests, and docs. Understanding users, making judgment calls, and deciding what is ready to ship stay with the builder.

Which of the following is safe to delegate to AI?

Generating a first draft of implementation code. A first draft is something you will review and revise. Shipping decisions, problem fit, and user needs are judgment and empathy calls, where AI is weakest.

Which of the following are true about AI-generated code? Select all that apply.

AI can generate code that runs but misses the intended problem; AI may recommend libraries or APIs that are outdated or don't exist; AI has predictable blind spots like concurrency, caching, and race conditions. Working code is not the same as correct code, and AI can confidently cite packages or APIs that are stale or made up. It will not reliably flag security issues on its own, so you have to ask and check.

What is the correct order of the Description Chain?

User voice → product requirement → technical spec → AI instruction. Start from what users actually say, turn it into a product requirement, then a technical spec, and only then write the instruction for AI. Each step narrows the one before it.

"The code works, but no one wants to use it" is a failure at which discernment lens?

Experience Quality. The code passes Functional Integrity because it works. Experience Quality asks whether people find it useful and want to use it.

Before shipping something AI helped build, what does Diligence require?

Understanding what the code does and being accountable for the outcome. Passing tests and AI self-review are useful checks, but they don't replace responsibility. You own what you ship, so you need to understand it.

Why should you write acceptance tests before writing any code?

To give both you and AI a shared definition of "done" from the start. Acceptance tests written first describe what success looks like, which gives AI a clear target and gives you an objective way to judge its output.

AnthropicFree courseScore: 9/9

AI Fluency for Builders Quiz Answers

By Jackson Sophat · September 23, 2026

Quick answer

The AI Fluency for Builders course from Anthropic defines AI Fluency as the ability to use AI in ways that are efficient, effective, ethical, and safe, and applies the 4 Ds to building software:

Delegation: Hand AI first drafts of code. Keep empathy, judgment, and shipping decisions yourself.
Description: Follow the Description Chain: user voice → product requirement → technical spec → AI instruction.
Discernment: Check AI-built work for functional integrity, experience quality, production readiness, and responsible impact.
Diligence: Understand what the code does and be accountable for the outcome before you ship.

I passed the final quiz with a score of 9 out of 9 (100%). Every question and its correct answer is below.

Take the free course on Anthropic Academy

All 9 quiz questions and answers

Question 1 of 9
What does AI Fluency mean?
AnswerThe ability to use AI in ways that are efficient, effective, ethical, and safe
AI Fluency is about how well you work with AI, not how many tools you use, how much you automate, or whether you can build models yourself.
- The ability to automate every development task (incorrect)
- The ability to use every AI tool available (incorrect)
- The ability to use AI in ways that are efficient, effective, ethical, and safe (correct)
- The ability to build AI systems from scratch (incorrect)
Question 2 of 9
What are the four competencies in the 4D Framework?
AnswerDelegation, Description, Discernment, Diligence
The same 4 Ds from AI Fluency: Framework & Foundations, applied to building software. The other options are software lifecycle stages, not AI Fluency competencies.
- Define, Draft, Deploy, Debug (incorrect)
- Discovery, Development, Deployment, Debugging (incorrect)
- Data, Design, Delivery, Documentation (incorrect)
- Delegation, Description, Discernment, Diligence (correct)
Question 3 of 9
According to the Builder's Toolkit, where is AI weakest?
AnswerEmpathy, Judgment, and Shipping
AI is strong at producing code, tests, and docs. Understanding users, making judgment calls, and deciding what is ready to ship stay with the builder.
- Empathy, Judgment, and Shipping (correct)
- Documentation and scaffolding (incorrect)
- Architecture and implementation (incorrect)
- Code generation and testing (incorrect)
Question 4 of 9
Which of the following is safe to delegate to AI?
AnswerGenerating a first draft of implementation code
A first draft is something you will review and revise. Shipping decisions, problem fit, and user needs are judgment and empathy calls, where AI is weakest.
- Deciding whether a feature should ship (incorrect)
- Deciding whether the product solves the right problem (incorrect)
- Generating a first draft of implementation code (correct)
- Determining what the user actually needs (incorrect)
Question 5 of 9
Which of the following are true about AI-generated code? Select all that apply.
Answers
- AI can generate code that runs but misses the intended problem
- AI may recommend libraries or APIs that are outdated or don't exist
- AI has predictable blind spots like concurrency, caching, and race conditions
Working code is not the same as correct code, and AI can confidently cite packages or APIs that are stale or made up. It will not reliably flag security issues on its own, so you have to ask and check.
- AI can generate code that runs but misses the intended problem (correct)
- AI may recommend libraries or APIs that are outdated or don't exist (correct)
- AI will always surface security vulnerabilities without prompting (incorrect)
- AI has predictable blind spots like concurrency, caching, and race conditions (correct)
Question 6 of 9
What is the correct order of the Description Chain?
AnswerUser voice → product requirement → technical spec → AI instruction
Start from what users actually say, turn it into a product requirement, then a technical spec, and only then write the instruction for AI. Each step narrows the one before it.
- User voice → product requirement → technical spec → AI instruction (correct)
- AI instruction → technical spec → product requirement → user voice (incorrect)
- Product requirement → AI instruction → user voice → technical spec (incorrect)
- Technical spec → user voice → product requirement → AI instruction (incorrect)
Question 7 of 9
"The code works, but no one wants to use it" is a failure at which discernment lens?
AnswerExperience Quality
The code passes Functional Integrity because it works. Experience Quality asks whether people find it useful and want to use it.
- Responsible Impact (incorrect)
- Experience Quality (correct)
- Functional Integrity (incorrect)
- Production Readiness (incorrect)
Question 8 of 9
Before shipping something AI helped build, what does Diligence require?
AnswerUnderstanding what the code does and being accountable for the outcome
Passing tests and AI self-review are useful checks, but they don't replace responsibility. You own what you ship, so you need to understand it.
- Understanding what the code does and being accountable for the outcome (correct)
- Confirming that all AI-generated tests pass (incorrect)
- Checking that the latest model was used (incorrect)
- Getting AI to review its own work one more time (incorrect)
Question 9 of 9
Why should you write acceptance tests before writing any code?
AnswerTo give both you and AI a shared definition of "done" from the start
Acceptance tests written first describe what success looks like, which gives AI a clear target and gives you an objective way to judge its output.
- To comply with standard engineering practices (incorrect)
- To give both you and AI a shared definition of "done" from the start (correct)
- To limit how much AI can change in your solution (incorrect)
- To make AI generate better syntax (incorrect)