Deep dive · Jump here anytime when you want more on this idea
Delegation, Description, Discernment, Diligence
After this, you'll be able to break any AI task into the four skills that decide whether it works: what you hand off, how you ask, how you judge the reply, and how you own the result. You'll spot which of the four you skip most.
Before you start
Complete Briefing AI like a new colleague first; this lesson groups that briefing skill, and three others, into the four skills that decide every result.
The idea
Most people think getting good with AI is about knowing better prompts. It isn't. It's four separate skills, and almost everyone is strong at two of them and quietly weak at the other two. Researchers at Anthropic gave these four skills names, the 4 D's: Delegation, Description, Discernment, and Diligence. Once you can see them, you stop guessing why one chat went great and the next one fell apart.
Delegation is deciding what to hand off. You set a clear goal before you ask, hand over a task that is the right size (not your whole project, not something you could type yourself), and pick the right tool for the job. Description is saying it well: enough background, who the output is for, the exact shape you want, and one example of what good looks like. Discernment is judging what comes back instead of trusting it. You check the facts, push on the weak parts (hardest to do when the answer looks polished), and ask for a better version instead of taking the first draft. Diligence is owning the result: you verify before you ship, you know when not to use AI at all, and you keep a person in the loop on decisions that matter. Diligence is the one most scorecards skip, and the one that separates a careful user from a sloppy one.
Here is the same task done two ways. A weak run: you type 'write a follow-up email to a client', paste the reply straight into your inbox, and send it. A fluent run: you say who the client is and what they care about (Description), ask for three short options (Delegation, right-sized), read them and notice one invents a meeting date that never happened (Discernment), fix it, and reread the final version yourself before it goes out under your name (Diligence). Same tool, same minute of work, completely different result. The difference was the four skills, not the prompt.
Now find your own weak D. Think about the last real thing you did with AI. Walk through the four out loud. Which one did you skip? Most people skip Discernment or Diligence because the output looked finished, so they stopped checking. That blind spot is the single most useful thing to know about yourself, and you can have an AI name it for you in five minutes at the end of this lesson.
Try it (6 min)
Watch out for
Paste this into Claude
I want to get better at working with AI, measured by four skills: Delegation (deciding what to hand off, setting a clear goal, picking the right tool), Description (giving enough context, naming the audience, asking for a specific format, showing an example of good), Discernment (checking facts, catching weak logic, asking for a better version instead of the first draft), and Diligence (verifying before I ship, knowing when NOT to use AI, keeping a human in the loop on big calls). Here is the last real task I did with you: [describe in 3 or 4 sentences exactly what you asked for, what you gave me, and what you did with my answer]. Walk through all four skills one at a time. For each, tell me honestly whether I did it well, partly, or not at all, with one sentence of evidence from what I just described. Then name the ONE skill I think I'm good at but the evidence says I'm not. Be specific and don't flatter me.
What good looks like
When this breaks
You can now
Name your weakest D out loud and point to one specific moment in a recent task where skipping it cost you. If you can do that, you can see all four.
Key takeaways
Getting good with AI is four skills, not one: what you hand off, how you ask, how you judge the reply, and how you own the result. Your weakest D sets your ceiling, so find it and fix that one first.
Go deeper