AI Fluency · The Five D Framework
How fluent are you with AI?
We start from Anthropic's four D framework and add a fifth D we consider crucial: Discovery, knowing the tool itself. Take the quiz, see how you score, then turn the same lens on your own conversations.
5D: the five Dimensions of AI fluency
Each Dimension covers a different part of working well with AI. The quiz ahead tests whether you can recognise them in real life.
Description
Communicate clearly with AI.
Delegation
Decide what to do with AI vs. yourself.
Discernment
Evaluate what AI gives you.
Diligence
Use AI responsibly and accountably.
Discoverynew · ours
Know how the tool itself actually works.
Tap any pill to see that Dimension's indicators. The first four are Anthropic's, catalogued across 24 behavioural indicators; Discovery is ours. The self-check prompt further down scores a practical subset of all five.
This means you can:
The fifth D: Discovery
Anthropic's four D framework treats "the AI" as a generic partner you talk to, hand work to, judge, and use responsibly. It gestures at picking the right tool for a job, under Delegation, to get a better result. It never reaches the thing underneath all four: knowing how the specific tool in front of you works. What it can see right now, where its answers come from, what it can and can't do on the surface you're using, and what happens to the data you send it. That's Discovery, and it's a skill you can be fluent or clumsy in.
It lies close to Delegation, so it's worth being precise. Delegation asks what should I hand off, and which approach gets a better result? It optimises, and it assumes you already know the toolbox. Discovery asks what is this tool: what can it see, what does it know, what can it do here, what happens to my data? One is being efficient with the system; the other is being right about it. Anthropic's own Delegation indicators (“anticipate AI's limitations”, “match task to method”) only work if you already hold that map of the tool. Discovery is the skill of building the map.
It matters for safety and AI literacy. You can have perfect intentions about Delegation and still lose work because you didn't understand how the tool actually behaves, as in this documented case.
You can also be fluent in all four Ds and still write a prompt the tool can't carry out. Anthropic's own
AI Fluency tutorial is a case in point: its self-check prompt tells Claude to
“pull from any surface — chat, Claude Code, Cowork” and to tag every quote by surface,
yet the tools it relies on (recent_chats, conversation_search) index claude.ai
chat only. Asked to analyse that same prompt, Claude itself reports that the Claude Code and Cowork tags
“aren't achievable.” So it works only partly, and only in claude.ai chat; pasted into Claude Code,
where those tools don't exist, it can't run as written (observed 8 July 2026, still present on 10 July 2026;
Anthropic may since have revised it: source).
That gap is Discovery, and Discovery is what makes real Description, Delegation and Discernment possible.
Anthropic is now building in this direction itself. Its new Reflect feature (beta, July 2026) analyses your usage against the four Ds natively, inside claude.ai chat, with Cowork conversations listed as “coming soon” and Claude Code not covered. Then on 16 September 2026 Anthropic merged Cowork into ordinary Claude conversations, and has not yet said what that does to Reflect's scope; the “coming soon” line names a product that no longer exists on its own. Reflect scores the four Ds; knowing this kind of thing about your tools is the fifth, Discovery.
Ten scenarios
Ten moments of working with AI. Some are prompts typed to Claude; others are situations, a decision or action someone takes around the AI. Each is labelled, so you know which layer to read: a prompt asks you to look at how they talk to Claude; a situation asks you to look at what they chose to do. Pick the dimension it shows most clearly.
Why two forms? Description lives in the words of a prompt; delegation, diligence and much of discernment live in the decision around it, including the choice not to use AI at all, which no prompt can show. Real moments often touch more than one dimension (Anthropic's own examples overlap too), so pick the primary one. Arguing for another isn't wrong; noticing the overlap is itself a sign of fluency.
Question 1 of 10
Prompt to Claude
“”
You scored 0/10
Turn the lens on your own work
This prompt scores your recent conversations across all five Dimensions and builds you a visual scorecard. Claude in chat and Claude Code work differently, so here's a correct version for each. Pick the one that matches where you're pasting it.
Use in a claude.ai chat. It works best with memory features turned on. It analyses your main chat conversations only. If you're not inside a Project, conversations that happened inside Projects will not be included. If you use Anthropic's new Reflect feature, the prompt will invite you to bring its insights in as extra evidence.
Paste this into a new Claude Code session. It will produce two results, both in HTML:
a scorecard for your AI Fluency, and a usage report from the /insights command (or the
closest similar feature available to you), which informs the Discovery score.
Both are exploratory reflections, not formal assessments. Diligence and Discovery often happen in what you don't paste, so they may read "not observed" even when you practise them. That is expected, and worth noticing.
Both prompts adapt the one in Anthropic's tutorial; the corrections and the Discovery section are ours.