Just Delegate It: Complete Mind Map

Goal: Master the two core frameworks (Delegation Loop + Delegation Brief), every concept, every distinction the exam tests. Practical and visual.

THE ENTIRE COURSE IN ONE SENTENCECopy Anchor Link

Give AI the job. Watch it work. Check what matters. Then decide.

Two frameworks hold everything:

1. The Delegation Loop - 6 steps (the rhythm)
2. The Delegation Brief - 6 questions (the written form)

THE BIG PICTURE MAPCopy Anchor Link

THE DELEGATION LOOP (the rhythm of every job)

  YOURS                AI WORKS              YOURS
+--------+--------+  +--------+--------+  +--------+--------+
| DEFINE | DELE-  |  |OBSERVE |INTER-  |  | VERIFY | ACCEPT |
|        | GATE   |  |        | VENE   |  |        |        |
| Say    | Hand   |  | Let it | Fix    |  | Check  | Decide |
| what   | over   |  | work   | drift  |  | claims | fit    |
| result | the    |  | without| with   |  | that   | for    |
| you    | job    |  | micro- | small- | | matter | use?   |
| want   |        |  | manage | est    |  |        |        |
|        |        |  |        | fix    |  |        |        |
+--------+--------+  +--------+--------+  +--------+--------+
                          |        |
                          |   <----+ REDIRECT (back to Delegate)
                          |
                      <-------------------+ SEND BACK (back to Define)

AI owns the MIDDLE. You own BOTH ENDS.
AI can HELP with your ends. It cannot TAKE them.

THE 11 CONCEPTS AT A GLANCECopy Anchor Link

CONCEPT 1: ASKING vs GIVING THE JOBCopy Anchor Link

ASKING:                              GIVING THE JOB:
"What are some good courses          "Find three free courses for
 for learning AI agents?"             learning AI agents and put them
                                      in a table with provider, language,
                                      time needed, and a link."

  Result: a list                       Result: a table with links
  Next steps: YOURS                    Next steps: CHECKING
  Can you hand it off? NO              Can you hand it off? YES
  Can you verify it? HARD              Can you verify it? YES (open links)

The difference is NOT length. It's that the job asks for a THING TO EXIST.

To delegate = give AI the responsibility for producing the result.

Works on any plan, any model. Depends on what kind of sentence you type, not what you pay.

CONCEPT 2: OUTCOME, NOT CLICKSCopy Anchor Link

Two ways to hand over the same job:

When steps ARE appropriate: When a source or order of work is REQUIRED -- that's a constraint (Concept 3), not a recipe.

When steps are NOT appropriate: When you're writing them because you're nervous. That feeling belongs at the END (when you check), not the START.

Formula: "When this is done, I will have [the thing]."

CONCEPT 3: THE DELEGATION BRIEF (Core Framework)Copy Anchor Link

Six questions you answer BEFORE AI starts. The work order, not a form.

+------------------------------------------------------------------+
|                    THE DELEGATION BRIEF                           |
|                                                                  |
|  1. OUTCOME       What should exist when the work is complete?   |
|                                                                  |
|  2. CONTEXT       What AI needs to know that it can't know       |
|                   on its own (including any source you give)      |
|                                                                  |
|  3. CONSTRAINTS   Requirements and limits that apply to the job  |
|                                                                  |
|  4. AUTHORITY *   What AI may decide WITHOUT asking you           |
|                   (the line people SKIP)                         |
|                                                                  |
|  5. DELIVERABLE   The FORM the result takes (table, doc, file)   |
|                   shaped so someone can USE it                   |
|                                                                  |
|  6. VERIFICATION* What you will CHECK before accepting           |
|                   (written BEFORE the work starts)               |
|                   (the other line people SKIP)                   |
+------------------------------------------------------------------+
  * = the two lines people skip. These are the most important.

Why it matters: Every time a result disappoints you, one of these six was unanswered. Now you can say WHICH.

It's a thinking tool, not a form. Nobody fills in six boxes for a two-line job. But knowing which question was missing is the diagnostic.

Worked ExampleCopy Anchor Link

Outcome: three free online courses for learning AI agents, compared,
with one recommended for a complete beginner.

Context: I have never built or used an AI agent, and I do not know
how to program. I can study about five hours a week.

Constraints: only courses that can be started for free, and only
courses that do not expect programming knowledge. Search for current
information rather than answering from what you already know.

Authority: choose the sources yourself. If two courses are equal,
pick the shorter one without asking me. Say clearly if a course is
paid, or if it expects me to know how to program.

Deliverable: a short table with provider, language, time needed,
whether a certificate costs money, and a link. Then a two-sentence
recommendation.

Verification: mark anything you could not confirm as "unverified"
instead of guessing.

How the Brief Maps to AI Fluency's Four CompetenciesCopy Anchor Link

CONCEPT 4: RESEARCH (Current Information)Copy Anchor Link

AI has two ways of knowing things:

The rule: When the job needs what is true TODAY, make AI search, and CHECK that it searched.

How to verify it searched:

• A found fact comes WITH a source you can open
• A remembered fact arrives ALONE
• No source = don't assume it looked, regardless of what it says about itself
• A source is NOT a date -- find the date on the PAGE yourself

Add this line to force transparency:

For each item, say whether you found it by searching just now
or already knew about it, and give the date of the page you used.

Web search vs Deep research mode:

• Web search = one fact that one search settles
• Deep research = job whose deliverable IS the report (takes minutes, reads many pages)

"A citation is not verification. Checking the citation is verification."

CONCEPT 5: THE SOURCECopy Anchor Link

When the job needs something AI can't search for (your syllabus, your office notice, a message thread).

Two instructions that turn pasted text into a source:

1. "Using only the text below"     (stay inside it)
2. "Quote the sentence you used"   (show your work)

Together they make "not in the source" a GOOD answer instead of a gap AI fills for you.

Privacy rules BEFORE pasting:

• Remove names, phone numbers, account numbers, anything you wouldn't pin to a public notice board
• For sensitive work: use temporary chat (ChatGPT) or incognito chat (Claude)
• Neither makes the paste disappear -- provider still holds a copy for a time

When a long chat goes stale: Ask AI to summarize what's been decided + what remains, copy it, open fresh chat, paste as first message. The summary is a source too.

For short text: Paste inline. For long documents: Attach the file (plus/paperclip button).

CONCEPT 6: THE DELIVERABLECopy Anchor Link

Answer vs Deliverable:

Ask for the FORM: table with named columns, document with sections, spreadsheet with formulas, file to download.

Critical rule: Add this line to every brief:

Keep every "unverified" mark exactly where it is.

Why: Deliverables LOSE caveats. An "unverified" mark that survived three chat replies can vanish when poured into a tidy document. Count caveats in chat vs document. If the document has fewer, find what was dropped.

CONCEPT 7: AUTHORITY (Two Faces)Copy Anchor Link

Face 1: How much of the METHOD AI decidesCopy Anchor Link

Plan-first = ask AI to list steps and wait for go-ahead BEFORE starting.

Before you start, list the steps you plan to take, in order,
and wait for my go-ahead. Do not begin until I say go.

Why: The cheapest moment in any job is BEFORE it runs. A plan costs a minute to read. The same mistake found AFTER costs the whole run.

An approved plan is NOT a cage. AI may find a better route. What it may NOT do: cross a line you drew, or change something that matters WITHOUT TELLING YOU. The failure = silent change, not the change itself.

Face 2: What AI may DO without asking (Boundaries)Copy Anchor Link

Trust grows with a record. First run: read the plan AND every step. Fifth clean run: read the plan. This is how General Agents turns these into permission modes for unattended work.

Four Questions Before AI Touches Anything RealCopy Anchor Link

1. What may it read?
2. What may it do?
3. Do I trust where its information comes from?
4. Am I allowed to give it this access at all?

CONCEPT 8: THE JOB AGAIN (Repetition)Copy Anchor Link

1st time: Run the brief. Check results.
2nd time: Save the brief outside the chat. Run again. Compare.
3rd time: Put what doesn't change WHERE AI READS IT BY ITSELF.

Where to Put Recurring ContextCopy Anchor Link

Project = a space that holds files + standing instructions. Every chat inside it starts already informed.

Standing instruction = a rule written once, followed in every chat inside the project. Example: "mark anything you could not confirm as unverified."

Memory vs Project: Both assistants keep memory about you (if enabled). Once a month, check BOTH what you stored and what AI remembered. Remove what's no longer true. A wrong line in a Project = a wrong line in EVERY result.

Scheduling (optional):

• ChatGPT: scheduled tasks on paid plans (ordinary feature)
• Claude: scheduled tasks on paid plans via Cowork
• Rule: A brief you saved today is what you schedule. But General Agents teaches what must be true before a job runs unattended.

When what repeats is the METHOD (not just context): That's a Skill, taught in Skills & Connectors. This course stops at Projects.

CONCEPT 9: INTERVENE (Six Moves)Copy Anchor Link

When AI drifts, diagnose by SYMPTOM:

SYMPTOM                          MOVE                     SIZE
+-------------------------------+------------------------+-----------+
| Answered a different question | STOP: restate outcome  | 1 line    |
| One fact/line is wrong        | CORRECT: name error +  | 1 sentence|
|                               |   what should replace  |           |
| Heading wrong way, not wrong  | REDIRECT: name the     | 1 sentence|
|   yet                         |   direction            |           |
| The job itself was wrong      | CHANGE the requirement | 1 sentence|
|                               |   (say you changed it) |           |
| Can't do it at this level     | ESCALATE: turn on      | 1 action  |
|                               |   thinking / stronger  |           |
|                               |   model                |           |
| Keeps failing after escalate  | TAKE IT BACK: some     | --        |
|                               |   jobs are yours       |           |
+-------------------------------+------------------------+-----------+

The move people MISS: Escalate. Sometimes "AI failed" means "I used the wrong level." Start fast. When the job needs careful comparison or must be right, move up.

If free plan has no stronger model: Make the job SMALLER. One step at a time, each checked.

After every intervention:

1. Check the fix IN THE RESULT (not in AI's description of the fix)
2. Write the lesson into the brief as a Constraint or Authority line
3. "A claimed fix that did not happen is the failure this step exists to catch."

Golden rule: The SMALLEST message that fixes it is the intervention.

CONCEPT 10: VERIFY BEFORE YOU ACCEPT (The Method)Copy Anchor Link

"Check AI's work" is advice. This is the METHOD.

Four Steps Inside VerifyCopy Anchor Link

IDENTIFY ──> TRACE ──> CHALLENGE ──> DECIDE
   |            |          |            |
Which claims  Open the   Ask how it   One of four
matter enough source,    could be     decisions
to check?     find the   wrong
              sentence,
              check the
              date

Step 1: IDENTIFY (What to Check)Copy Anchor Link

Check where an error would COST you:

• Prices, dates, figures, calculations
• Quotations (exact words)
• Legal or safety claims
• The recommendation you'll act on
• Anything hard to reverse

Step 2: TRACECopy Anchor Link

• Open the source
• Find the SENTENCE that supports the claim
• Check the DATE on the page
• Ask: is this source good enough? (provider's own page > blog about provider)

Step 3: CHALLENGECopy Anchor Link

• Could this have changed since the page was written?
• Is AI repeating a secondary source?
• Is an assumption being stated as a fact?
• When two credible sources disagree: primary usually wins, unless it's older

Step 4: DECIDE (Four Exits)Copy Anchor Link

+----------+  +----------+  +-------------+  +----------+
|  ACCEPT  |  | CORRECT  |  | INVESTIGATE |  |  REJECT  |
| Evidence |  | Specific |  | Evidence is |  | Not      |
| is enough|  | error,   |  | incomplete  |  | reliable |
|          |  | fixable  |  | or contra-  |  | enough   |
| Use the  |  | (= Inter-|  | dictory     |  | to use   |
| result   |  |  vene)   |  | (trace      |  |          |
|          |  |          |  |  again)     |  | Take job |
|          |  |          |  |             |  | back     |
+----------+  +----------+  +-------------+  +----------+

AI doing the work does NOT transfer accountability to AI. The name on the result is yours.

CONCEPT 11: SAME JOB, DIFFERENT AICopy Anchor Link

The test of a brief: It works somewhere else without a rewrite.

Why it works: The brief describes the JOB, not the TOOL. Nothing in the six lines names a button. Buttons differ between assistants and change every few months.

What to record (NOT a ranking):

• Did it finish the job?
• How many messages needed?
• Did it honor the Authority line?
• Did it mark anything unverified?
• How did the deliverable compare?

Critical distinction: Two runs on one day ≠ a ranking of two assistants. Models change too fast.

The job belongs to you. The model is replaceable.

THE "NEED FIRST, THEN TOOL" MAPCopy Anchor Link

The rule: a control is learned when a job NEEDS it, not before.

THE DELEGATION RECORD (Template)Copy Anchor Link

DELEGATION RECORD
Job: ______________________  Date: __________

THE BRIEF
  Outcome:
  Context:
  Constraints:
  Authority:
  Deliverable:
  Verification:

THE RUN
  Assistant used:             Messages needed:
  Plan asked for first?       Plan changed by me?
  Where it drifted, and the move that fixed it:

THE CHECK
  Claim 1 that mattered:     Source opened:     Finding:
  Claim 2 that mattered:     Source opened:     Finding:
  Anything AI marked "unverified":

THE DECISION
  Accept / Correct / Investigate / Reject
  The one sentence behind it:

THE SECOND RUN (Concept 11)
  Assistant:     Finished?     Brief changed?
  Where the results differed:

WHAT I WILL SAVE FOR NEXT TIME (Concept 8)
  Lines that did not change:

If a line is empty = a step of the loop hasn't happened yet.

HOW THE BRIEF GROWS ACROSS COURSESCopy Anchor Link

THIS COURSE (watched job):
  Outcome, Context, Constraints, Authority, Deliverable, Verification

GENERAL AGENTS (unwatched job) adds:
  + Sources it may use
  + Destination for finished file (Tier 3)
  + Rule for what should stop the run and come back to you

AI WORKER OUTCOME CONTRACT adds:
  + Trigger that starts the work
  + Evidence that must support the result
  + Acceptance criteria
  + Main number and guardrails
  + Forbidden actions
  + Named human with final authority
  + Record to keep

The same questions, asked at larger scale.

THE BOOK'S GROWTH LADDERCopy Anchor Link

The 10-80-10 Rule: First 10% is yours (define). Middle 80% is AI's (execute). Last 10% is yours (verify).

EXAM POWER RULESCopy Anchor Link

1. Asking = answer (work stays with you). Job = result (you just check).
2. Describe the OUTCOME, not the steps. Steps limit AI to the quality of YOUR plan.
3. Six lines: Outcome, Context, Constraints, Authority, Deliverable, Verification. Authority + Verification are the two people skip.
4. A citation is not verification. CHECKING the citation is verification.
5. "Using only the text below" + "quote the sentence you used" = two instructions that make pasted text a real source.
6. Deliverable ≠ answer. A deliverable can be handed on, saved, acted on.
7. Plan-first = cheapest moment to fix a job (before it runs).
8. Authority has two faces: what AI may DECIDE (method) + what AI may DO (actions/permissions).
9. Six intervention moves: Stop, Correct, Redirect, Change requirement, Escalate, Take it back. Escalate is the one people miss.
10. Verify = Identify, Trace, Challenge, Decide. Four exits: Accept, Correct, Investigate, Reject.
11. AI doing the work does NOT transfer accountability. Your name, your responsibility.
12. A good brief runs in a second AI without rewriting. It describes the job, not the tool.
13. Third time you paste the same context = put it in a Project.
14. The smallest message that fixes the drift IS the intervention.
15. Need first, then tool. Every button is learned when a job asks for it, not before.

KEY VOCABULARYCopy Anchor Link

QUICK SELF-TEST (10 Questions)Copy Anchor Link

Q1: What's the difference between asking AI and giving AI a job?

Asking gets an answer and keeps the next steps with you. A job asks for a thing to exist and leaves you the checking.

Q2: Name the 6 steps of the Delegation Loop. Which stay yours?

Define, Delegate, Observe, Intervene, Verify, Accept. Both ENDS stay yours: Define + Delegate at the start, Verify + Accept at the end.

Q3: Why does describing the outcome beat dictating the steps?

A recipe limits the result to the quality of YOUR plan. An outcome lets AI choose the method and lets you judge the result against what you asked for.

Q4: Name the 6 lines of the Delegation Brief. Which two do people skip?

Outcome, Context, Constraints, Authority, Deliverable, Verification. People skip Authority and Verification.

Q5: How can you tell whether AI searched or answered from memory?

A searched fact should arrive WITH a source you can open. A remembered fact arrives alone and sounds just as sure. No source = don't assume it looked.

Q6: What two instructions turn pasted text into a source?

"Using only the text below" + "quote the sentence you used." Together they make "not in the source" a good answer.

Q7: What's the cheapest moment to fix a job, and how do you get there?

BEFORE the run. Ask for the plan and wait for go-ahead (plan-first), then change one step.

Q8: Third time you paste the same background -- where does it go?

Into a Project: sources as knowledge, rules as standing instructions. The chat carries only today's specific job.

Q9: Name the 6 intervention moves. Which one do people miss?

Stop, Correct, Redirect, Change the requirement, Escalate, Take it back. Escalate is the one people miss ("AI failed" sometimes means "I used the wrong level").

Q10: What are the 4 steps of verification and the 4 decisions it ends in?

Steps: Identify, Trace, Challenge, Decide. Decisions: Accept, Correct, Investigate, Reject.

PRACTICAL CHEAT SHEET: Writing a Brief in 2 MinutesCopy Anchor Link

Step 1: "When this is done, I will have ___"          --> OUTCOME
Step 2: "What does AI not know?"                       --> CONTEXT
Step 3: "What limits matter?"                          --> CONSTRAINTS
Step 4: "What may it decide? What must it ask about?"  --> AUTHORITY
Step 5: "What form should the result take?"            --> DELIVERABLE
Step 6: "What will I check before trusting this?"      --> VERIFICATION

Then add:
  "Mark anything you could not confirm as 'unverified'."
  "If evidence is missing, say so and stop rather than inventing it."

EXAM DOMAIN MAPPINGCopy Anchor Link