# Just Delegate It: Complete Mind Map (/docs/courses/just-delegate-it/study-guide) --- type: Document title: "Just Delegate It: Complete Mind Map" description: "Exam-focused study guide covering the Delegation Loop, Delegation Brief, all 11 concepts, intervention moves, and verification method with domain mappings." status: stable generated: by: human:shazilk-dev at: "2026-09-29T03:22:49Z" ksor: audience: [public] owner: human:shazilk-dev approval: by: human:shazilk-dev at: "2026-09-29T03:22:50Z" order: 10 trust_tier: unverified build_id: sha256:c9386a58484b06cf62301b7977bd8e4dcf2d43d61a4bc8c9deea6107a3646da3 source_commit: 4323a20bccded0a914b7c2fb65128db995e45dce dirty: true ksor_version: 0.0.60 --- > **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 SENTENCE > **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 MAP ``` 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 GLANCE | # | Concept | One-Liner | Part | | --- | ---------------------- | ----------------------------------------------------------------------------------------- | ------ | | 1 | Asking vs Giving | Asking gets an answer (work stays with you). Giving a job gets a result (you just check). | Part 1 | | 2 | Outcome, Not Clicks | Describe the result. Leave the steps to AI. | Part 1 | | 3 | The Delegation Brief | 6 questions BEFORE AI starts = something to check against. | Part 1 | | 4 | Research | When the job needs current facts, make AI search, check that it did. | Part 2 | | 5 | The Source | Give it, say "use only this", check that it stayed inside. | Part 2 | | 6 | The Deliverable | Ask for the finished THING (table, doc, file), not an answer. | Part 2 | | 7 | Authority | Say what AI may decide alone, what it must ask about. Plan-first. | Part 3 | | 8 | The Job Again | Save what doesn't change. Third time = put it where AI reads it. | Part 3 | | 9 | Intervene | Fix drift with the smallest message. Write the lesson into the brief. | Part 3 | | 10 | Verify | Identify, Trace, Challenge, Decide. Accountability stays with YOU. | Part 4 | | 11 | Same Job, Different AI | A good brief runs in a second AI without a rewrite. The model is replaceable. | Part 4 | --- ## CONCEPT 1: ASKING vs GIVING THE JOB ``` 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 CLICKS Two ways to hand over the same job: | Approach | What Happens | Problem | | ----------------------------- | ------------------------------------------- | ------------------------------------------------------------------------------------------ | | **Recipe** (dictate steps) | AI follows YOUR plan, even where it's weak | Your plan is the LIMIT. If your chosen sites aren't the best, the result is neat but poor. | | **Outcome** (describe result) | AI chooses the method, you judge the result | AI might find better sources you didn't think of. | **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) 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 Example ``` 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 Competencies | Brief Line | Competency | | ------------ | ------------------------ | | Outcome | Delegation + Description | | Context | Description | | Constraints | Description + Diligence | | Authority | Delegation + Diligence | | Deliverable | Description | | Verification | Discernment + Diligence | --- ## CONCEPT 4: RESEARCH (Current Information) AI has two ways of knowing things: | Source | Characteristics | When to Use | | ------------------------------- | --------------------------------------------------------------------- | ---------------------------------------------------- | | **Remembering** (training data) | Large, out of date. Fact arrives ALONE, sounds just as sure. | Background knowledge, concepts | | **Searching** (web) | Current, with sources. Fact should arrive WITH a source you can open. | Prices, dates, links, whether something still exists | **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 SOURCE 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 DELIVERABLE **Answer vs Deliverable:** | Answer | Deliverable | | ----------------- | ------------------------------------------- | | Replies to YOU | Shaped for someone to USE, save, hand on | | Lives in the chat | Can leave the chat (doc, file, spreadsheet) | | Read once, done | Kept, passed on, acted on | **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) ### Face 1: How much of the METHOD AI decides **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) | Boundary | Why It Exists | | ------------------------------ | ------------------------------------------- | | Research, but do not purchase | Money that moves is hard to move back | | Draft, but do not send | Sending creates a promise someone else sees | | Inspect, but do not delete | Deleting destroys evidence needed to check | | Choose, but say when you chose | A silent decision can't be reviewed | **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 Real 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) ``` 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 Context | What It Is | Where It Goes | | ------------------------- | ----------------------------------------------------------------------------- | | Sources that don't change | **Project knowledge** (files AI reads in every chat inside the project) | | Rules that don't change | **Standing instructions** (rules AI follows in every chat inside the project) | | Today's specific job | The chat message | **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) 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) "Check AI's work" is advice. This is the METHOD. ### Four Steps Inside Verify ``` 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) 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: TRACE - 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: CHALLENGE - 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) ``` +----------+ +----------+ +-------------+ +----------+ | 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 AI **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" MAP The rule: a control is learned when a job NEEDS it, not before. | The Job Needs | So You | Concept | | ---------------------------------------------- | ---------------------------------------------------------- | ----------------- | | Current information | Ask for it, turn on web search | 4 | | A source only you have | Paste it or attach file | 5 | | Something private in the job | Remove private parts, use temp/incognito chat | 5 | | A finished thing, not an answer | Ask for the document/table/file itself | 6 | | Room to choose its own method | Say what it may decide alone + what it must ask | 7 | | The same context every time | Put sources in Project as knowledge, rules as instructions | 8 | | The same job on a clock | Scheduled task (where plan has one) | 8 | | More capability than it showed | Turn on thinking / stronger model | 9 | | A place for the result / source in another app | Connector | 7 + Skills course | --- ## THE DELEGATION RECORD (Template) ``` 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 COURSES ``` 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 LADDER | Step | What You Say to AI | Where Taught | | ----------------- | ------------------------------------------ | -------------------------------- | | Prompt | "Tell me something." | AI Prompting | | Collaboration | "Help me do something." | AI Fluency | | **Delegated job** | **"Here is the outcome. Do the work."** | **THIS COURSE** | | Workflow | "This work happens again and again." | General Agents + Workflow Design | | AI Worker | "Own this responsibility." | Building a Digital FTE | | Agent Factory | "Build and run those workers as a method." | Thesis | **The 10-80-10 Rule:** First 10% is yours (define). Middle 80% is AI's (execute). Last 10% is yours (verify). --- ## EXAM POWER RULES 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 VOCABULARY | Term | Definition | | ---------------------------------------- | ------------------------------------------------------------------------------------------------- | | **Delegation Loop** | 6 steps: Define, Delegate, Observe, Intervene, Verify, Accept | | **Delegation Brief** | 6 questions before AI starts: Outcome, Context, Constraints, Authority, Deliverable, Verification | | **Outcome** | What should exist when work is complete | | **Context** | What AI needs to know that it can't know on its own | | **Constraints** | Requirements and limits on the job | | **Authority** | What AI may decide without asking. In chat = decisions. Later = permissions. | | **Deliverable** | The FORM the result takes, shaped so someone can use it | | **Verification** | What you'll check before accepting, written BEFORE the work starts | | **Plan-first** | Ask AI to list steps and wait for go-ahead before starting | | **Project** | Space holding files + standing instructions, loaded in every chat inside it | | **Knowledge** | The files a Project holds (what AI may read) | | **Standing instruction** | A rule written once, followed in every chat inside the Project | | **Temporary chat** | Chat not saved to history, no memory added (ChatGPT name). Claude = incognito chat. | | **Running job** | One job carried through the course, re-delegated at each concept | | **Need first, then tool** | A control is learned when a job needs it, not before | | **Identify, Trace, Challenge, Decide** | The 4 steps inside Verify | | **Accept, Correct, Investigate, Reject** | The 4 decisions verification ends in | | **Delegation Record** | One-page record: brief, run, check, decision | | **Fresh chat** | New chat with no carryover from previous conversation | | **Escalate** | Turn on thinking or stronger model when AI can't handle the level | | **Operating layer** | AI that acts on your real work, not just a box you type into | --- ## QUICK SELF-TEST (10 Questions) **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 Minutes ``` 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 MAPPING | Exam Domain | Concepts That Map Here | | ------------------------------------------- | ------------------------------------------- | | Domain 1: Prompting & Task Execution | Concepts 1, 2, 3, 6 | | Domain 2: Output Evaluation & Validation | Concept 10 (verification) | | Domain 3: Product & Model Selection | Concepts 4, 5, 11 | | Domain 5: Configuration & Knowledge Mgmt | Concept 8 (Projects, standing instructions) | | Domain 6: Governance, Risk, Responsible Use | Concepts 7, 10 (authority, accountability) | | Domain 7: Troubleshooting & Optimization | Concept 9 (intervention moves) |