Overview & Learning Objectives
What You Will Learn
By the end of this course you will be able to apply six skills that research identifies as the actual predictors of sustained AI adoption:
- Context Assembly — giving AI the right information to produce useful output
- Quality Judgment — knowing whether AI output is correct, complete, and appropriate
- Task Decomposition — breaking large tasks into AI-appropriate subtasks
- Iterative Refinement — improving AI output through structured feedback loops
- Workflow Integration — embedding AI into recurring processes, not one-off experiments
- Frontier Recognition — knowing what AI can and cannot do, and when that boundary shifts
How to Use This Page
Two Ways to Use This Course
Self-paced: Work through each module on your own. All exercises, examples, and answer keys are included inline. Budget about two hours.
Companion to live instruction: Follow along during a training session. Your instructor will guide the discussion, but everything you need is on this page for reference during and after class.
Looking for the instructor guide? View the Instructor Version.
Module 1: Why 80% Quit
The Pattern You Need to Recognize
Microsoft tracked hundreds of thousands of employees using AI tools. Excitement peaked for about three weeks. Then came a crater of disappointment. The vast majority quietly stopped. The 20% who survived didn’t write better prompts — they learned to manage AI like a capable but inexperienced team member.
Why People Fail
Here is the typical failure arc. See if it sounds familiar:
- You type “help me with this report” and get something generic.
- You try again and get something confident and wrong.
- You try a third time and decide it’s faster to do it yourself.
- You never come back.
That failure is not your fault. It is a training problem. Nobody showed you the right skills because most organizations do not know what they are yet.
The Evidence That Training Works
The UK government gave AI tools to 20,000 workers across 12 departments with proper training and support. After three months, over 80% didn’t want to give them up. Workers reported saving 25 minutes a day — nearly two weeks a year. Nine of twelve departments kept their licenses. The difference wasn’t the tool. It was the training.
Why This Training Is Different
This course is not teaching you how to use a tool. It is teaching you how to manage one. The same skills that make you a good leader — breaking work into pieces, knowing what good looks like, giving constructive feedback, verifying quality — are the skills that make you effective with AI.
The Management Framing
Would you hand a 100-page RFP to a brand new Marine and say “handle this”? No. You would break it into pieces. You would tell them which parts to tackle first. You would explain what good looks like. You would review their work and give feedback. That is exactly how you should work with AI.
Self-Check: What separates the 20% from the 80%?
The key difference is not prompt quality, tool selection, or technical aptitude. The 20% who persist treat AI as a team member to be managed — they break work into pieces, provide context, review output critically, and refine iteratively. The 80% who quit expected magic on the first try, got disappointed, and never developed the management skills that make AI useful.
Module 2: The Six Skills That Actually Matter
These are the six 201-level skills. For each one, you will see the difference between how most people use AI (101 behavior) and how effective users work (201 behavior). Study the examples carefully — the gap between the two is where the value lives.
Skill 1 — Context Assembly
AI is extremely sensitive to context quality. The difference between mediocre and useful output is usually the information you provide.
101 Behavior
Write me a counseling statement.201 Behavior
Write a counseling statement for a Lance Corporal who was late to
formation twice this month. The tone should be corrective but not
adversarial. The Marine has otherwise been a solid performer. Use
the format from NAVMC 10274. This is a page 11 entry, not an
adverse 6105.Key Insight
The expert prompt provides five pieces of context the vague prompt lacks: rank, situation, tone, format reference, and document type. Each one steers the AI away from generic output and toward something you could actually use.
Skill 2 — Quality Judgment
AI can produce two completely different outputs from the same prompt. Your job is knowing which one is actually ready to submit. Read both outputs below — they came from the exact same prompt. Which one gets approved by the awards board?
Write a summary action paragraph for an end-of-tour Navy and
Marine Corps Achievement Medal for a Staff Sergeant who led a
maintenance team of 4, managed a $2.1M equipment account with
zero discrepancies, and trained 12 new Marines over 18 months.Output A
Output B
Take a minute. Read both carefully. Which paragraph would survive an awards board review? What specifically makes one better than the other?
Reveal Analysis
Output A — Returned by Awards Board
Zero verifiable facts. “Exceptional,” “unparalleled,” “tremendous” are filler words that say nothing specific. “Multi-million dollar” instead of the actual figure. “Numerous Marines” instead of 12. This paragraph describes every Staff Sergeant who ever lived — it distinguishes no one. Any awards board would return it for rewrite.
Output B — Gets Approved
Every claim is specific and verifiable: 87 end items, $2.1M, 98% readiness, zero discrepancies across three inspections, 40-hour syllabus, 12 Marines qualified, 22% reduction in completion time, perfect SMI score. Actions are tied directly to measurable outcomes. Written in proper award narrative style.
Why This Matters
Both outputs came from the same AI with the same prompt. Output A is what AI produces when no one pushes back. Output B is what you get when an expert reviews the draft and demands specifics. The AI had the information — the prompt included the numbers — but it defaulted to generic praise language because that pattern dominates its training data. The expert’s job is to reject “sounds good” and demand “is accurate.”
Key Insight
AI defaults to language that sounds impressive over language that is accurate. The Brynjolfsson study found AI helped novice workers improve by 34% — but only because experts caught what the novices missed. Your domain knowledge is the filter. “Would this survive review?” is the only question that matters.
Skill 3 — Task Decomposition
Breaking a large task into smaller pieces improves AI’s success probability on each piece. Mollick calls this the Equation of Agentic Work. Compare these two approaches:
101 Behavior
Write me a training plan for the platoon.201 Behavior
First, list the required annual training events for FY26. Then map
them against the calendar, avoiding known conflicts. Then identify
which events need range time and put those first.Key Insight
The 201 prompt breaks the task into three sequential steps: list, map, prioritize. Each step has a clear input and a clear output. The AI can succeed at each piece even if it would struggle with the whole thing at once.
Skill 4 — Iterative Refinement
The first output is never the final product. What follows is a complete walkthrough showing how three specific refinement passes take a generic 70% draft to a 95% finished product. Pay attention to what changes at each step and why.
First Draft — AI Output (70% There)
Weekend Safety Brief
All personnel are reminded to exercise caution and good judgment
during the upcoming liberty period. The forecast calls for variable
weather conditions in the local area. All Marines and Sailors should
plan accordingly and ensure they have appropriate clothing and
supplies for any outdoor activities.
Personnel are expected to adhere to all applicable regulations
regarding alcohol consumption, vehicle operation, and personal
conduct. Designated drivers should be arranged in advance for any
social gatherings involving alcohol. Report any incidents to the
chain of command immediately.
All hands are reminded to maintain situational awareness at all
times. Avoid high-risk activities that could result in injury or
harm. Take care of your fellow Marines and do not allow them to
make poor decisions.
The chain of command remains available throughout the weekend for
any issues that arise. Enjoy the time off and return to duty
prepared for the training week ahead.What Is Wrong With This?
- Too generic — could be any unit, any weekend, anywhere
- No specific weather, dates, or local hazards
- Missing POC and emergency contact information
- Overly formal tone — no one talks like this at a formation
- No liberty radius or boundary information
Pass 1 — Fix Tone and Add Weather
The tone is too formal. Make it sound like a Gunny talking to his
Marines. Add this weekend's weather: high of 62, rain Saturday,
clear Sunday.Weekend Safety Brief
Listen up. We made it through another week. Before I cut you
loose, here's what you need to know.
Weather this weekend: Saturday is going to be wet — rain most of
the day, high of 62. Sunday clears up and should be decent. Plan
accordingly. If you're doing anything outdoors Saturday, bring rain
gear and don't be the guy who ends up hypothermic because he wanted
to tough it out in a t-shirt.
If you're drinking, have a plan. I don't care if it's two beers
or ten — you are not driving. Set up a designated driver before
you go out, not after your fourth round. Ride shares exist. Use
them.
Look out for each other. If your buddy is about to do something
stupid, stop him. That's not being soft — that's being a Marine.
If something goes sideways this weekend, do not try to hide it.
Call up the chain immediately. We can deal with almost anything
if we know about it early. What we can't deal with is finding
out Monday morning that something happened Friday night.
Enjoy the weekend. Come back ready to work.Better. The tone is right and the weather is in there. But it is still missing key information. Notice that the refinement prompt was specific — not “make it better” but exact instructions on what to fix.
Pass 2 — Add POC, Liberty Radius, Remove Redundancy
Add the duty officer contact info: Capt Rodriguez, (831) 555-0147.
Add the liberty radius: 250 miles. Remove the third paragraph, it
repeats the first.Weekend Safety Brief
Listen up. We made it through another week. Before I cut you
loose, here's what you need to know.
Weather this weekend: Saturday is going to be wet — rain most of
the day, high of 62. Sunday clears up and should be decent. Plan
accordingly. If you're doing anything outdoors Saturday, bring rain
gear and don't be the guy who ends up hypothermic because he wanted
to tough it out in a t-shirt.
If you're drinking, have a plan. I don't care if it's two beers
or ten — you are not driving. Set up a designated driver before
you go out, not after your fourth round. Ride shares exist. Use
them.
Liberty radius is 250 miles. If you need to go beyond that, you
come see me before you leave. No exceptions.
If something goes sideways this weekend, do not try to hide it.
Call up the chain immediately. Duty officer this weekend is Capt
Rodriguez: (831) 555-0147. Save that number in your phone right
now. We can deal with almost anything if we know about it early.
Enjoy the weekend. Come back ready to work.Getting close. Two passes in, the brief has the right tone, real weather, POC info, and a liberty boundary. One more pass to tighten it and add local hazards.
Pass 3 — Tighten and Add Local Hazards
Tighten to one page. Add: no swimming at Point Lobos due to rip
current advisory. Remind them about the Monday 0630 formation.Final Version — Pass 3 (95%)
Weekend Safety Brief
Listen up. Before I cut you loose — weather: Saturday is rain all
day, high of 62. Sunday clears up. Bring rain gear if you're going
out Saturday.
Drinking — have a plan before you go out. Designated driver or ride
share, set it up in advance. You are not driving.
Liberty radius is 250 miles. Beyond that, see me first. No
exceptions.
Local hazard: Point Lobos has a rip current advisory this weekend.
No swimming there. I don't care how good a swimmer you think you
are.
Duty officer this weekend is Capt Rodriguez: (831) 555-0147. Save
it in your phone right now. If something happens, call immediately.
Don't wait until Monday.
Monday formation is 0630. Not 0631. Be there, be ready, be sober.
Enjoy the weekend. Take care of each other.What to Notice
Each refinement prompt was specific — not “make it better” but exact instructions on what to fix. Three passes. Each one addressed concrete problems. Now it is 95% — good enough that you just give it a final read and send it. The GDPval study found that even simple improvements to prompts could boost AI win rates by another five percentage points.
Skill 5 — Workflow Integration
The difference between a one-time experiment and a real productivity gain is whether AI becomes part of a recurring process.
101 Behavior
"I'll try using AI for this one thing later."201 Behavior
"Every Friday, the duty NCO uses AI to draft the weekend safety
brief from the current weather, liberty boundaries, and recent
incidents."Key Insight
Microsoft research shows it takes up to 11 weeks to build the AI habit. The 201 behavior turns AI from something you try once into a standing step in a weekly or daily workflow. That is when the time savings compound.
Skill 6 — Frontier Recognition
Not every task is a good AI task. The capability boundary — what AI handles well versus what it handles poorly — is called the jagged frontier. It is not intuitive: AI may excel at a task you expect it to fail at, and fail at a task you expect it to handle easily. Worse, the frontier shifts as models improve.
Critical Research Finding
The BCG-Harvard study found that untrained users working outside the frontier performed 19 percentage points worse than those without AI at all. Using AI on the wrong task does not just waste time — it actively makes your work worse. Knowing where the frontier is for your domain is not optional.
The Full 201 Skills Reference
This table summarizes all six skills. Bookmark it. Come back to it. Use it to self-assess where you are strong and where you need practice.
| # | Skill | What It Means | 101 Behavior | 201 Behavior |
|---|---|---|---|---|
| 1 | Context Assembly | Giving AI the right information to produce useful output | “Write me a counseling statement.” | Provides rank, situation, tone, format, and regulatory reference |
| 2 | Quality Judgment | Knowing whether AI output is correct, complete, and appropriate | Accepts first output without review | Red-pens every output; verifies facts, format, and references |
| 3 | Task Decomposition | Breaking large tasks into AI-appropriate subtasks | “Write me a training plan.” | Breaks into sequential subtasks with clear inputs and outputs |
| 4 | Iterative Refinement | Improving AI output through structured feedback loops | Accepts or rejects first draft | Three revision passes with specific feedback each time |
| 5 | Workflow Integration | Embedding AI into recurring processes, not one-off experiments | “I’ll try AI sometime.” | AI is a standing step in a weekly or daily workflow |
| 6 | Frontier Recognition | Knowing what AI can and cannot do — and when that boundary shifts | Assumes AI can do everything or nothing | Maintains a personal map of AI strengths, weaknesses, and moving edges |
Self-Check: Can you name all six 201 skills?
Without looking at the table above, list all six skills and what each one means in your own words. Then scroll back up and compare. If you missed any or got the definitions wrong, re-read that skill's section before moving on.
- Context Assembly
- Quality Judgment
- Task Decomposition
- Iterative Refinement
- Workflow Integration
- Frontier Recognition
Module 3: The Delegation Equation
Three Questions Before You Delegate
Before handing any task to AI, run it through these three questions. This is Mollick’s framework, and it prevents the two most common mistakes: delegating tasks AI cannot do well, and refusing to delegate tasks AI would handle fine.
- Human Baseline Time — How long would this take you to do yourself?
- Probability of Success — How likely is AI to produce acceptable output?
- AI Process Time — How long to write the prompt, wait, and evaluate the result?
The math is simple: if (AI Process Time / Probability of Success) is less than Human Baseline Time, delegate. If not, do it yourself.
Example 1: Writing a 5-Paragraph Order
- Human baseline: 4 hours
- AI drafts: minutes
- Review time: 45 minutes
- Probability AI gets it right: ~70%
- Verdict: Delegate. Even at 70% accuracy, you save hours. The review catches what AI misses, and you are still done in under an hour instead of four.
Example 2: Calculating TIS/TIG
- Human baseline: 5 minutes
- AI process: Checking AI’s math takes as long as doing it yourself
- Verdict: Don’t delegate. The task is so fast that the overhead of prompting, waiting, and verifying eliminates any savings.
Example 3: Quarterly Training Schedule
- Human baseline: 6 hours
- AI produces: Good first-draft structure
- Review time: 30 minutes
- Verdict: Delegate the structure, fill in the specifics yourself. AI builds the skeleton; you provide the unit-specific knowledge that makes it real.
Key Insight
The more expertise you have, the better you are at all three variables. You estimate human baseline time more accurately. You judge AI success probability more accurately. You review AI output faster. AI does not replace expertise — it rewards it.
Practice: Apply the Delegation Equation to Your Own Work
Think of one recurring task from your own workflow. Estimate all three variables:
- Human Baseline Time: How long does this task take you now?
- Probability of Success: If you gave AI good context, how likely is it to produce something acceptable? (Be honest — 50%? 80%? 30%?)
- AI Process Time: How long would it take to write the prompt, wait for the output, and evaluate it?
Would you delegate it? If AI Process Time divided by Probability of Success is less than your Human Baseline Time, the answer is yes. If it is close, try it once and see. If it is clearly not worth it, move on to the next task.
Module 4: The Trust Problem — Red Pen Review
The Highest-Value Exercise in This Course
Below are three AI-generated documents for 1st Bn, 99th Marines. These are not documents with a single planted “gotcha” — they are realistic first drafts with the kinds of problems AI actually produces. Each one looks professional. Each one has multiple issues that would matter if you signed it and sent it up the chain.
Your job: red pen each document. Mark everything you would change, question, or verify before putting your name on it. Work through all three before checking the answer key. The point is not to find every issue on the first try — it is to build the habit of systematic review.
Digital learners: Open a notes app or document alongside this page. As you review each document below, write down every issue you find by paragraph number before checking the answer key.
Before You Start: The Review Checklist
Use this checklist as you review each document. Professional editors and quality reviewers do not rely on gut feeling — they use a systematic process.
AI Document Review Checklist
- References & Citations. Does every order number, publication, regulation, or directive actually exist? Can you look it up?
- Facts & Statistics. Are the numbers, dates, percentages, and claims verifiable? Do they match reality, or do they just sound good?
- Internal Consistency. Do all parts of the document agree with each other? Does the timeline work? Do names, units, and details stay consistent?
- Procedural Accuracy. Does the described process match how things actually work in your experience? Are critical steps missing?
- Substance vs. Style. Strip away the professional formatting and confident tone. Is there real, specific content underneath, or just polished filler?
- Would You Sign It? If your name goes on this document, are you confident in every claim it makes?
Document 1: Award Recommendation Narrative
Document 1: Award Recommendation Narrative (NAM)
RECOMMENDATION NARRATIVE
Corporal David R. Hernandez, USMC
Weapons Company, 1st Battalion, 99th Marines
Period of Service: March 2025 to January 2026
Corporal Hernandez demonstrated exceptional initiative and unwavering dedication to duty while serving as the Readiness Clerk for Weapons Company, 1st Battalion, 99th Marines. Through his unparalleled commitment to organizational excellence, he single-handedly transformed the company’s readiness reporting process, resulting in a 47% reduction in reporting errors and saving the command an estimated 156 man-hours per quarter.
Upon identifying critical inefficiencies in the company’s legacy readiness tracking system, Corporal Hernandez took it upon himself to design and implement a comprehensive digital solution utilizing the Microsoft Power Platform. His innovative tool streamlined the consolidation of training data from four subordinate sections, automating what had previously been a tedious manual process requiring extensive cross-referencing of multiple spreadsheets and paper records.
Corporal Hernandez’s tool was directly responsible for Weapons Company achieving a 98.7% readiness reporting accuracy rate during the Battalion’s Annual Training Assessment, the highest mark in the regiment. His system was subsequently adopted by two additional companies within the battalion, demonstrating the scalability and robustness of his solution. The Regimental Commander personally recognized his contribution during the quarterly awards ceremony, per MARADMIN 045/26.
Corporal Hernandez’s exceptional performance, tireless work ethic, and selfless dedication to mission accomplishment reflect great credit upon himself, Weapons Company, 1st Battalion, 99th Marines, and the United States Marine Corps.
Document 2: Unit SOP Excerpt
Document 2: Unit SOP Excerpt (Check-in/Check-out Procedures)
STANDARD OPERATING PROCEDURES (EXTRACT)
Chapter 4, Section 3: Personnel Check-in / Check-out Procedures
Effective Date: 01 January 2026
References: (a) MCO 1000.6A Personnel Assignment Policy
(b) MCO 1320.11H Permanent Change of Station
(c) BnO 1000.1 Personnel Administration Procedures
1. PURPOSE. To establish standardized procedures for the check-in and check-out of all personnel assigned to or departing from Weapons Company.
2. CHECK-IN PROCEDURES. All newly reporting personnel shall complete the following within 72 hours of reporting:
a. Report to the Company Office (Bldg 1284) with original orders, service record book, and three copies of PCS orders.
b. Complete Company Check-in Sheet (Appendix 4-3A).
c. Receive section assignment from the Company First Sergeant.
d. Report to assigned Section Leader within 24 hours of company check-in.
e. Complete gear inventory with Section Leader using CMC Form 4790/142.
f. Attend Company Indoctrination Brief, held every Monday at 0800 in the Company Classroom (Bldg 1284, Rm 203).
g. Complete initial counseling with Section Leader within 14 calendar days.
3. CHECK-OUT PROCEDURES. Personnel departing the company shall initiate checkout NLT 10 working days prior to departure date:
a. Obtain Company Check-out Sheet from the Company Office.
b. Clear all issued gear through the Section Leader and Company Armory.
c. Return all company-specific access badges and keys to the Company Gunnery Sergeant.
d. Complete a transfer of duties brief with the identified relief, if applicable.
e. Obtain Company Commander’s signature on checkout sheet.
f. Forward completed checkout sheet to Battalion S-1 NLT 5 working days prior to departure.
4. EXCEPTIONS. Personnel departing on emergency leave are exempt from the 10-working-day requirement but must complete telephonic checkout with the Company First Sergeant within 48 hours of departure. Checkout sheet will be completed by the Marine’s Section Leader in their absence and forwarded to Battalion S-1 per reference (b).
5. RESPONSIBILITIES. The Company First Sergeant is responsible for maintaining the check-in/check-out tracker and providing a monthly report to the Company Commander. Section Leaders will report any check-in discrepancies to the First Sergeant within 24 hours per reference (a).
Document 3: Information Paper
Document 3: Information Paper
Subject: Feasibility of Implementing Digital Training Tracker for 1st Battalion, 99th Marines
1. PURPOSE. To provide the Commanding Officer with an assessment of implementing a digital training tracker using the Microsoft Power Platform to replace the current spreadsheet-based system.
2. BACKGROUND. 1st Battalion currently tracks individual and collective training completion using a combination of Excel spreadsheets maintained at the company level and manual reports consolidated by the S-3. This process requires approximately 12 hours per week of staff time across the battalion. According to a 2024 Marine Corps Training and Education Command (TECOM) study, units using digital tracking systems reported a 34% improvement in training visibility and a 28% reduction in lapsed individual training requirements.
3. CURRENT SITUATION.
a. The S-3 section consolidates training data from four companies weekly. Average turnaround time from company submission to battalion roll-up is 3.2 working days.
b. Common errors include duplicate entries, outdated completion dates, and missing course identifiers. An internal audit in November 2025 found a 23% error rate in the consolidated tracker.
c. Three Marines in the S-3 shop hold Power Platform development experience. MSgt K. R. Odom completed the Microsoft PL-900 certification in October 2025.
4. ANALYSIS.
a. A Power Apps-based tracker would provide real-time visibility at all echelons, eliminating the weekly consolidation cycle.
b. Estimated development time: 60–80 hours using AI-assisted development methods per MARADMIN 612/24, which authorizes the use of AI tools for administrative process improvement across the Marine Corps.
c. Recurring cost: $0. Power Apps is included in the battalion’s existing M365 E3 licensing. No additional licenses or procurement actions required.
d. Risk: Low. The Power Platform operates within the DoD M365 security boundary and has been granted an Authority to Operate (ATO) at the DoD enterprise level.
5. RECOMMENDATION. Approve development of a Power Apps training tracker as a 90-day pilot program within Headquarters Company. Success criteria: reduction in consolidation time to same-day, error rate below 5%, and positive user feedback from a minimum of 10 users.
Prepared by: Capt M. J. Sullivan, S-3A
Date: 03 February 2026
Answer Key
Did you review all three? Do not open this until you have gone through each document with the checklist above.
Reveal All Findings — Combined Answer Key
Document 1: Award Recommendation — 5 Issues Found
Issue 1 (Fabricated Statistics — HIGH): “47% reduction in reporting errors” and “156 man-hours per quarter.” These are suspiciously precise figures with no supporting data. Where did these numbers come from? An award narrative must be backed by verifiable accomplishments. If the CO asks “where did you get 47%?” there is no answer.
Issue 2 (Fabricated Reference — HIGH): “per MARADMIN 045/26” — this MARADMIN does not exist. MARADMINs do not typically direct individual recognition at quarterly award ceremonies. AI fabricated a plausible-looking reference number to add authority to the claim.
Issue 3 (Inflated Language — MEDIUM): “Unparalleled commitment,” “single-handedly transformed,” “tireless work ethic.” These are generic superlatives that could describe anyone. Strong award narratives use specific accomplishments, not adjective stacking. What did the Marine actually do?
Issue 4 (Unverifiable Claim — MEDIUM): “98.7% readiness reporting accuracy rate” and “highest mark in the regiment.” The percentage is suspiciously precise, and claiming the highest mark in the regiment requires regimental-level data that a company clerk would not normally have access to.
Issue 5 (Style Over Substance — LOW): The narrative is heavy on adjectives and light on specifics. How many sections used the tool? What was the actual before/after workflow? How was it tested? The writing is polished but hollow. A strong award narrative replaces “unwavering dedication” with “volunteered 40 off-duty hours over 6 weeks.”
Document 2: Unit SOP — 5 Issues Found
Issue 1 (Fabricated References — HIGH): “MCO 1000.6A Personnel Assignment Policy” and “MCO 1320.11H Permanent Change of Station” do not exist. AI generates plausible MCO numbers and titles. Every reference in an AI-generated document must be verified. If you cannot find it in the MCO library, it is fabricated.
Issue 2 (Fabricated Form Number — HIGH): “CMC Form 4790/142” is not a real form. The 4790 series relates to equipment maintenance (MIMMS/GCSS-MC), not personnel gear inventories. AI mixed up form number series.
Issue 3 (Missing Critical Step — HIGH): No mention of medical/dental records check-in. Every Marine who has ever checked into a unit knows that medical and dental are on the checklist. This is a glaring omission that reveals the document was not written by someone who has actually processed a check-in.
Issue 4 (Procedural Gap — MEDIUM): Section 4 addresses emergency leave telephonic checkout but does not address gear accountability. The Marine departed with issued gear. Who inventories it? Who signs for it? The First Sergeant’s biggest concern is property accountability, and this SOP has no procedure for it.
Issue 5 (Timeline Gap — LOW): Section 2.d requires reporting to Section Leader “within 24 hours” but 2.g allows “14 calendar days” for initial counseling. In practice, a Section Leader conducts initial counseling at the same time as the first meeting. The 14-day window is unrealistically long and suggests the author did not understand the actual workflow.
Document 3: Information Paper — 5 Issues Found
Issue 1 (Fabricated Study — HIGH): “According to a 2024 Marine Corps Training and Education Command (TECOM) study” with specific statistics (34% improvement, 28% reduction). This study almost certainly does not exist. AI invents authoritative-sounding studies with precise statistics to support its recommendations. If you cannot find the study, do not cite it.
Issue 2 (Fabricated MARADMIN — HIGH): “MARADMIN 612/24, which authorizes the use of AI tools for administrative process improvement across the Marine Corps.” This MARADMIN does not exist. Even if a similar policy existed, AI fabricated both the number and the characterization of its content. Citing a nonexistent directive in a paper for the CO is a career-level error.
Issue 3 (Misleading Claim — MEDIUM): “Recurring cost: $0.” This ignores ongoing maintenance hours, the developer’s time investment (60–80 hours is not free), and the possibility that premium connectors may be needed. Technically true about licensing, but misleading about total cost of ownership.
Issue 4 (Unverifiable Internal Data — MEDIUM): “3.2 working days” turnaround and “23% error rate” from “an internal audit in November 2025.” These figures sound authoritative but were generated by the AI. If the CO asks “who conducted this audit?” there is no answer because no audit occurred.
Issue 5 (Oversimplified Authority Claim — LOW): “Granted an Authority to Operate (ATO) at the DoD enterprise level” oversimplifies the ATO process. While Power Platform operates within DoD M365, characterizing it as having a blanket enterprise ATO may not be accurate for all use cases and data types. The CO’s information assurance officer would flag this.
The Verification Hierarchy
Not every document needs the same level of review. Use this table to calibrate your verification effort to the stakes:
| Stakes Level | Examples | Review Method | Mode |
|---|---|---|---|
| High | Legal, personnel, financial documents | Line-by-line verification | Centaur mode |
| Medium | Correspondence, reports, briefings | Spot-check key facts and references | Mixed mode |
| Low | First drafts, brainstorming, formatting | Quick review for obvious errors | Cyborg mode |
The Lesson
AI-generated documents will look professional. The formatting will be clean. The tone will be appropriate. That is exactly what makes errors dangerous — they hide inside polished output. AI does not know what it does not know. It will never say “I made up this MCO number” or “this statistic has no source.” It will present fabricated information with the same confidence as verified facts. Your domain expertise is the only quality gate. The Red Pen Review is not a one-time exercise — it is a permanent habit. Every AI output you sign gets the red pen treatment. Every time.
Module 5: Centaur, Cyborg, or Neither
Two Patterns for Human-AI Work
Research identifies two productive patterns for how humans and AI work together. Understanding which pattern fits your task is key to using AI effectively.
- Centaur: Clear division of labor. Human does Phase 1, AI does Phase 2, human reviews Phase 3. Best for high-stakes work where you need a clean handoff and thorough review. Think of it as sequential — you take turns.
- Cyborg: Continuous integration. Fluid boundary between human and AI contributions. Best for creative and iterative work where you refine together in real time. Think of it as simultaneous — you work in tandem.
When to Use Each Pattern
Centaur works best when the stakes are high and you need an audit trail — legal documents, personnel actions, financial reports. You know exactly where the human work ends and the AI work begins.
Cyborg works best when you are creating something iteratively — drafting correspondence, brainstorming plans, building briefing outlines. The boundary between your input and AI output blurs, and that is fine because the stakes are lower.
Exercise: Workflow Mapping
Pick one recurring task from your actual work — something you do weekly or monthly. Then complete the following five steps:
- Name the task. Be specific. Not “admin work” but “weekly training schedule update.”
- Break it into 3–5 subtasks. What are the actual steps you go through?
- Mark each subtask:
- Human Only — requires judgment, context, or authority AI cannot provide
- AI Could Help — AI drafts, human refines
- AI Should Do This — routine, repeatable, low-stakes
- Identify the pattern. Is this a centaur workflow (clear handoffs) or a cyborg workflow (continuous collaboration)?
- Estimate time savings per iteration if you used AI for the marked subtasks.
Use this template to map your workflow:
Workflow Map Template
Current Time to Complete: ________
Subtask Breakdown:
1. ________________________________ [ Human Only / AI Could Help / AI Should Do This ]
2. ________________________________ [ Human Only / AI Could Help / AI Should Do This ]
3. ________________________________ [ Human Only / AI Could Help / AI Should Do This ]
4. ________________________________ [ Human Only / AI Could Help / AI Should Do This ]
5. ________________________________ [ Human Only / AI Could Help / AI Should Do This ]
Pattern: [ Centaur / Cyborg ]
Estimated Time With AI: ________
Estimated Time Saved Per Iteration: ________
Example: Completed Workflow Map
Task Name: Weekly Training Schedule Update
Current Time to Complete: 2 hours
Subtask Breakdown:
- Collect inputs from platoon sergeants — Human Only (requires conversations, judgment about priorities)
- Draft the schedule format with time blocks — AI Should Do This (routine formatting)
- Map training events to available ranges and facilities — AI Could Help (AI drafts, you verify availability)
- Resolve conflicts and make final decisions — Human Only (requires command judgment)
- Format and distribute the final product — AI Should Do This (routine formatting and distribution prep)
Pattern: Centaur (clear handoffs between human and AI steps)
Estimated Time With AI: 50 minutes
Estimated Time Saved Per Iteration: 70 minutes
Self-Check: Is your workflow map complete?
Check your map against these criteria:
- Does it have at least 3 subtasks?
- Did you mark each subtask as Human Only, AI Could Help, or AI Should Do This?
- Did you identify the pattern (centaur or cyborg)?
- Did you estimate time savings?
- Did you identify at least one frontier risk?
If any of these are missing, go back and fill them in. A complete workflow map is the foundation for everything in Builder Orientation.
Module 6: Frontier Mapping & Your Assignment
Your Personal Frontier Map
The jagged frontier is different for every role. A supply clerk’s frontier looks nothing like an intelligence analyst’s. Your job is to start mapping yours. Use three columns: what AI handles well for your work, what it handles poorly, and what is on the moving edge — tasks you should re-test periodically as models improve.
| AI Handles Well | AI Handles Poorly | Moving Frontier (Check Periodically) |
|---|---|---|
| Drafting correspondence | Regulatory interpretation | Data analysis and summarization |
| Formatting documents | Classified information handling | Technical writing for specialized fields |
| Brainstorming and outlining | Real-time operational decisions | Complex planning and scheduling |
These are example entries. Your frontier map will have different items based on your MOS, billet, and daily tasks. Start with three entries in each column and update it as you gain experience.
Your Assignment
This Week: Try It
- Pick one recurring task from your workflow map (Module 5)
- Try using AI on it this week
- Note what worked and what did not
- Document your results: what worked, what didn't, and which 201 skill you used most. If you are in a live session, be ready to share. If self-paced, keep these notes for reference when you move to Builder Orientation.
The goal is not perfection. The goal is one real attempt with a real task so you have concrete experience to build on. The 80% who quit never got past the generic first try. You will.
For Those Continuing to Builder Orientation
If you are moving on to the Builder Orientation course, come prepared with a specific problem you want to solve. Be ready to decompose it into subtasks. Builder Orientation assumes you can already apply all six 201 skills — there you will put them to work building actual tools.
What You Now Know
You are now equipped with the same judgment framework that separates sustained AI adopters from the 80% who quit. The skills you practiced today — Context Assembly, Quality Judgment, Task Decomposition, Iterative Refinement, Workflow Integration, and Frontier Recognition — are not about any particular tool. They are about how you think. That is what makes them durable.
The research is clear: trained users with these skills save meaningful time, produce better work, and keep using AI months after the initial excitement fades. You have the skills. Now go use them.
Research Sources
The statistics and studies cited throughout this course are fully referenced in the SOP Section 11: References — Research Foundation. Key sources include:
- Microsoft Work Trend Index — 300,000+ employees, 80% abandonment rate
- UK Government Digital Services (2025) — 20,000 employees, 25 min/day savings
- Dell'Acqua et al. (2023), Harvard/BCG — Jagged frontier, 19-point performance drop
- Brynjolfsson, Li, & Raymond (2025) — 34% improvement for novice workers
- Mollick (2026), Wharton — Delegation equation, management as AI superpower
- OpenAI GDPval (2025) — 1,320 tasks, expert parity benchmarks
Capstone Deliverable
Personal Frontier Map
Identify 5 real tasks from your current role. For each task, classify it as “Delegate to AI,” “Collaborate with AI,” or “Do Yourself.” Include a one-sentence justification for each classification based on what you learned about the jagged frontier.
Knowledge Check
Module 1: Why 80% Quit
According to the research cited in Module 1, what is the primary reason most people stop using AI tools within a few weeks?
The 80% who quit expected magic on the first try. The 20% who persisted learned to treat AI as a capable but inexperienced team member that needs context, broken-down tasks, and review. The failure is a training gap, not a technology gap.
The UK government study gave AI tools to 20,000 workers with proper training. What was the key outcome that demonstrates training matters?
With proper training and support, over 80% of the UK government workers wanted to keep using AI, saving an average of 25 minutes per day. The difference was not the tool but the training, which flips the 80% failure rate seen in organizations without structured support.
The course uses a "management framing" for working with AI. Which analogy best captures this framing?
The management framing says you should treat AI the same way you would manage a capable but inexperienced team member. You break work into pieces, provide context, explain standards, review output, and give specific feedback. The same leadership skills that make you effective with people make you effective with AI.
Module 2: The Six 201-Level Skills
A prompt reads: "Write me a counseling statement." An improved version provides rank, situation, tone, format reference, and document type. Which 201 skill does this improvement demonstrate?
Context Assembly is giving AI the right information to produce useful output. The expert prompt provides five specific pieces of context (rank, situation, tone, format, document type) that steer the AI away from generic output. The quality of AI output is directly proportional to the quality of the context you provide.
During the weekend safety brief exercise, the first AI draft was labeled "70% there." After three specific refinement passes, it reached "95%." What made each pass effective?
Iterative Refinement works when each pass addresses concrete, specific problems. Pass 1 fixed tone and added weather. Pass 2 added POC info, liberty radius, and removed redundancy. Pass 3 tightened length and added local hazards. "Make it better" is vague and ineffective; specific instructions produce specific improvements.
What distinguishes Workflow Integration (201 behavior) from one-off AI use (101 behavior)?
Workflow Integration means AI becomes part of a recurring process, not a one-time experiment. For example, "Every Friday, the duty NCO uses AI to draft the weekend safety brief." Microsoft research shows it takes up to 11 weeks to build the AI habit, and the time savings only compound when AI is integrated into standing workflows.
Module 3: The Delegation Equation
You have a task that takes 5 minutes to do yourself. Writing the prompt, waiting, and verifying the AI output would also take about 5 minutes. What does the Delegation Equation recommend?
The Delegation Equation says: if (AI Process Time / Probability of Success) is less than Human Baseline Time, delegate. When the task is already fast (5 minutes), the overhead of writing the prompt, waiting, and verifying the result wipes out any savings. This is exactly the TIS/TIG calculation example from the module.
A 5-paragraph order takes 4 hours to write manually. AI can draft it in minutes, but the output is only about 70% accurate. According to the Delegation Equation, what should you do?
Even at 70% accuracy, delegating a 4-hour task to AI and spending 45 minutes reviewing saves you over 3 hours. The review catches what AI misses, and you are done in under an hour. The equation works because your expert review is the quality gate, not AI's accuracy alone.
The course states that "the more expertise you have, the better you are at all three variables" in the Delegation Equation. Why does expertise make AI more useful, not less?
Domain expertise improves all three variables in the Delegation Equation: you estimate human baseline time more accurately, you judge AI success probability more accurately, and you review AI output faster because you know what right looks like. AI does not replace expertise; it rewards it.
Module 4: Red Pen Review
In the naval message review exercise, the AI-generated DTG read "R 06 FEB 2026 1430Z." What category of error does this represent?
The correct DTG format is "R 061430Z FEB 2026" where day and time run together followed by the Zulu designator. AI frequently defaults to civilian date conventions because they are more common in its training data. A communications clerk would catch this immediately, which illustrates why domain expertise is the quality gate.
The Red Pen Review exercise describes fabricated references (like fake MARADMIN numbers) as the "single most dangerous category of AI error." Why are fabricated references more dangerous than other errors?
Fabricated references are dangerous precisely because they look real. AI generates plausible-looking MCO numbers, MARADMIN numbers, and study citations with full confidence. It will never flag its own fabrication. The only way to catch them is to verify every reference against the actual source. This is why the review checklist starts with "Does every reference actually exist?"
The award recommendation narrative for Cpl Hernandez included "a 47% reduction in reporting errors" and "saving 156 man-hours per quarter." What should make you suspicious of these figures?
AI fabricates precise-sounding statistics to make claims more authoritative. "47%" and "156 man-hours" sound specific and credible, but there is no audit, no baseline measurement, and no methodology behind them. If the CO asks "where did you get 47%?" there is no answer. Strong award narratives use verifiable accomplishments, not AI-generated numbers.
Module 5: Centaur and Cyborg Modes
You are drafting a legal personnel action that will go into a Marine's permanent record. Which collaboration pattern should you use, and why?
Centaur mode uses a clear division of labor with sequential handoffs: human does one phase, AI does the next, human reviews. This is essential for high-stakes documents like legal actions, personnel records, and financial reports where you need to know exactly where the human work ends and the AI work begins, and where thorough review is non-negotiable.
In the workflow mapping exercise, each subtask is labeled "Human Only," "AI Could Help," or "AI Should Do This." What determines which label a subtask gets?
"Human Only" subtasks require judgment, context, or authority that AI cannot provide (like gathering inputs from platoon sergeants or making final decisions). "AI Should Do This" subtasks are routine, repeatable, and low-stakes (like formatting). "AI Could Help" falls in between, where AI drafts and the human refines. The key factor is whether the subtask needs human expertise, not whether it involves text or numbers.
A completed workflow map shows that using AI could reduce a weekly training schedule update from 2 hours to 50 minutes. What makes this estimate credible?
The workflow map example breaks the task into five specific subtasks, labels each one, and identifies where AI contributes. The time savings estimate comes from subtracting the time AI handles (routine formatting, initial drafting) while keeping human-only tasks at their original duration. A subtask-level breakdown makes the estimate testable rather than aspirational.
Module 6: Frontier Mapping
The BCG-Harvard study found that people who used AI on tasks outside the "jagged frontier" performed 19 percentage points worse than those without AI. What does this finding mean for your daily work?
The BCG-Harvard finding is clear: using AI outside its capability boundary does not just waste time, it makes your work actively worse. People trusted AI's confident-sounding but incorrect output instead of relying on their own judgment. This is why Frontier Recognition is one of the six essential skills: you must know where the boundary is for your specific domain.
The frontier map template uses three columns: "AI Handles Well," "AI Handles Poorly," and "Moving Frontier (Check Periodically)." Why is the third column necessary?
The jagged frontier is not static. As AI models improve, tasks that were beyond AI's capability move inside the boundary. The "Moving Frontier" column captures tasks you should re-test periodically. Data analysis, technical writing, and complex planning were examples given as areas where the boundary is actively shifting. A frontier map is only useful if you update it.
The capstone deliverable asks you to classify 5 tasks as "Delegate to AI," "Collaborate with AI," or "Do Yourself." This exercise applies which combination of 201 skills?
Classifying tasks requires Frontier Recognition (understanding what AI can and cannot handle in your domain) combined with Task Decomposition (breaking your work into pieces small enough to evaluate individually). You cannot classify a task like "do admin work" because it is too broad. You must decompose it into specific subtasks and then map each one against the frontier.
Course Completion Checklist
Course Complete!
You have completed AI Fluency Fundamentals. View your progress dashboard or generate your certificate.