This lesson is being recorded

Every video of this course is being uploaded to my YouTube channel. Subscribe so you do not miss it.

Bruno AshfordBruno Ashford @brunoashfordSubscribe

Module 1 · Lesson 1.3 · 12 min

The 4Ds: Delegation, Description, Discernment, Diligence

Apply Anthropic's AI Fluency framework as a management routine for every task you hand to AI.

Module 1
Agentic AI, Explained for Owners

Think like a manager of AI, not a user of it

Bruno Ashford
Bruno Ashford · Module 101 / 12
1.1 From chatbot to agent

What actually changes

ChatbotAgent
Works inOne replyA loop of steps
Uses your filesOnly what you pasteOpens, reads, writes
Gives youText to processFinished work
You ask forContentAn outcome
Bruno Ashford · Module 102 / 12
1.1 The agent loop

Plan, act, check, repeat

  1. 1ReadYour request and context
  2. 2ActUse a tool on your files
  3. 3CheckLook at the result
  4. 4RepeatUntil done or it needs you
Bruno Ashford · Module 103 / 12
1.2 The Autonomy Ladder

Three rungs of trust

  1. 1SuggestIt drafts, you act
  2. 2Act with approvalIt works, asks first
  3. 3Act aloneIt works, you review later
Bruno Ashford · Module 104 / 12
1.2 The rule
The cost of a mistake sets the rung, not how clever the AI looks.
Bruno Ashford
Bruno Ashford · Module 105 / 12
1.3 The 4Ds (Anthropic AI Fluency framework)

A management routine for AI

  1. D1DelegationSplit the task, keep the judgment
  2. D2DescriptionBrief it like a new hire
  3. D3DiscernmentReview like a manager
  4. D4DiligenceOwn what goes out
Bruno Ashford · Module 106 / 12
1.3 Description

The five-part brief

Two minutes of briefing saves twenty of fixing.

  • Goal: what done looks like
  • Context: who, why, which files
  • Rules: standards, tone, never do
  • Output: format, length, where to save
  • Check: report what you read and changed
Bruno Ashford · Module 107 / 12
1.3 A truth that hurts
Most bad AI output is a bad brief.
Bruno Ashford
Bruno Ashford · Module 108 / 12
1.4 Where AI earns its place

Score every task

Score15
ImpactSaves minutesSaves hours or speeds revenue
RiskInvisible, easy fixReaches client, money, legal
EaseIn someone's headWritten, digital, repeatable

Opportunity score = Impact x Ease / Risk

Bruno Ashford · Module 109 / 12
1.4 Never delegate

Some calls stay human

The agent can prepare the facts. You make the call.

  • Hiring and firing
  • Final pricing on big deals
  • Legal or medical judgment
  • Apologising to an upset client
Bruno Ashford · Module 110 / 12
1.5 Choosing your tool

Three desks, one office

ToolBest forSetup
Claude chatThinking, one-off writingNone
CoworkOne-off tasks on your filesDesktop app, paid plan
Claude CodeRepeatable, shareable processesInstall, paid plan or API
Bruno Ashford · Module 111 / 12
Recap
Pick the task, set the rung, brief it well, check it like a manager. The tool comes last.
Bruno Ashford
Bruno Ashford · Module 112 / 12
01 / 12

Use ← → to move. The marker highlights text and the pencil draws on the slide; navigation locks while you annotate. Undo removes the last mark, Esc exits.

1Delegation: split the task

Do not ask whether a task can go to AI. Ask which part of it can. Gathering data, drafting and formatting usually can. Deciding, approving and handling the relationship usually cannot. Write the split down before you start.

2Description: use a five-part brief

Goal, context, rules, output and a final check. It takes two minutes to write and saves twenty minutes of back and forth.

Prompt for Claude
Goal: [what done looks like]
Context: [who it is for, why, which files or folder]
Rules: [my standards, tone, things to never do]
Output: [format, length, file name, where to save]
Before you finish: list what you read, what you changed, and anything you were unsure about.
3Discernment: review like a manager

Spot check at least three facts. Ask where numbers came from. Look for gaps that were filled with confident guesses. Your job is not to redo the work, it is to judge it.

Prompt for Claude
Show me where each number in this report came from, and list any assumptions you made that I did not give you.
4Diligence: own what goes out

Review before anything reaches a client, a bank or a regulator. Keep sensitive data out until your security rules are in place. When a brief works twice, save it; in Module 5 it becomes a reusable skill.

5Run the one-minute pre-flight

Before each new task, run the checklist on the 4Ds cheat sheet: part delegated, rung chosen, brief complete, review method clear, no sensitive data. It takes a minute and prevents most of the problems I see in the field.

Here is a truth that hurts a little: most bad AI output is a bad brief. When owners tell me the AI 'does not get our business', I ask to see the request. Usually it is one line, with no context, no standard and no finish line. We would never brief a person like that.

Anthropic's AI Fluency framework describes four skills for working well with AI: Delegation, Description, Discernment and Diligence. I teach them as a management routine. Delegation is choosing the work. Description is the brief. Discernment is the review. Diligence is owning the result.

Key points

  • Delegation: split the task, keep the judgment.
  • Description: goal, context, rules, output, check.
  • Discernment: spot check and ask for sources.
  • Diligence: you own what goes out under your name.
  • Most bad output is a bad brief.

Avoid

  • One-line briefs with no context or finish line.
  • Reviewing only the tone and never the numbers.
  • Blaming the tool instead of improving the brief.
  • Rewriting good briefs from scratch every time instead of saving them.
Real-world example: a roofing contractor

Mike's first attempt was 'write a follow-up for the Johnson quote'. The result was generic and offered a discount he never approved. With a five-part brief (goal, the quote file, no discounts, mention the 10-year warranty, under 120 words, list the details you used) the draft was ready to send with one small edit. Same AI, same task. The only thing that changed was the manager.

Bruno Ashford Subscribe