Week 1 · Course 1 · 2 Hours · All Personnel
The six 201-level skills that separate sustained adopters from the 80% who quit.
Expert-Driven Development · MCD-Monterey
Before we start
Agenda
Module One
Microsoft Work Trend Index
Microsoft tracked AI tool adoption across 300,000+ employees. Excitement peaked at three weeks. Then most people quietly stopped.
The failure loop
UK Government Digital Services
20,000 government workers, 12 departments, three months. After training, 80% refused to give the tools back — and 9 of 12 departments kept their licenses.
You are not learning to use a tool. You are learning to manage one.
The thesis of this course
The management framing
That's how you should work with AI.
What separates the 20% who stick with AI from the 80% who quit?
Not better prompts. Not a better tool. Not being more technical.
They manage AI — like a capable but inexperienced team member. They give it context, break work into pieces, evaluate the output, and iterate. The same skills you already use to lead Marines.
Module Two
The Six 201 Skills
Skill 1 · Context Assembly
Skill 2 · Quality Judgment
Skill 3 · Task Decomposition
Skill 4 · Iterative Refinement
Three specific prompts. Not "make it better." Specific.
Skill 5 · Workflow Integration
Skill 6 · Frontier Recognition
Drafting correspondence, summarising long documents, formatting, naming conventions, brainstorming, restructuring text, translating between styles.
Interpreting specific MCOs, calculating TIS/TIG accurately, anything requiring institutional knowledge that isn't in its training data, anything where the cost of being wrong is high.
BCG × Harvard, 2023
When 758 BCG consultants used AI on tasks outside the frontier, they were 19 percentage points less likely to produce correct work than the control group with no AI.
An AI draft of an awards write-up has dates, names, dollar figures, and reads beautifully. Sign it?
No. Not until Quality Judgment kicks in. AI presents fabricated facts with the same confidence as real ones. Verify the dollar figures against the property book. Verify the dates against the award period. Verify the reference numbers exist.
Reads beautifully ≠ is accurate. Your name goes on the bottom.
Module Three
Three questions, every task
Delegate when (1) is large, (2) is high, and (3) is much smaller than (1).
Example 1 · 5-paragraph order
Example 2 · TIS/TIG calculation for promotion
Short tasks with a high cost of being wrong are usually not worth delegating.
Example 3 · Quarterly training schedule
AI doesn't replace expertise.
It rewards it.
The expertise multiplier — Mollick, 2026
A task takes you 10 minutes. AI does it in 30 seconds — but you spend 12 minutes verifying. Delegate?
No. The third number in the equation — AI process time — includes evaluation. If checking takes longer than doing, the AI cost you time, not saved you time.
The fast tasks where AI looks efficient are often the ones where the math doesn't actually work. Always count evaluation time.
Halfway point
Ten minutes. Stretch, refill, check messages. We're back at [TIME] for the Red Pen Review.
Module 4 starts at the top of the hour
Module Four
AI is confidently right and confidently wrong — in the same paragraph.
The fundamental Quality Judgment problem
The three documents
These are not "gotcha" documents with one planted error. They're realistic first drafts — the kind of output AI actually produces, with the kinds of problems it actually creates.
Debrief · Document 1
Debrief · Document 2
Debrief · Document 3
The Verification Hierarchy
An AI document includes a confident reference to "MCO 5215.1J". You don't recognize it. Action?
Look it up. Always. AI invents Marine Corps order numbers in confident, plausible formats — correct prefix, correct format, completely fabricated content.
If a reference can't be verified in MCO Library / EPME / your reference library, it's fabricated. Strike it. Verifiable references go in. Made-up ones don't.
Module Five
The two patterns — BCG × Harvard, 2023
Clear division of labour. Human does Phase 1, hands off to AI for Phase 2, human reviews in Phase 3. Distinct responsibilities, clear handoff points, verification at every boundary.
Best for: high-stakes work that needs accountability. Legal review, admin separations, security packages, fitrep drafts.
Continuous integration. Constant back-and-forth between human and AI. The boundary is fluid — sometimes you write, sometimes AI does, sometimes you edit each other.
Best for: creative and iterative work. Building Power Apps, writing SOPs, drafting training materials, exploratory data analysis.
Centaur, in practice
Verification at every boundary. This is the pattern for anything you sign.
Cyborg, in practice
Use this for creative work, prototyping, and exploration — not for documents you sign.
You're prototyping a Power App. Centaur or cyborg?
Cyborg. Prototyping is creative, iterative, exploratory — the boundary between you and AI is fluid. You're not signing anything; you're discovering what the right tool looks like.
Once the prototype works and you're documenting it for production — switch back to centaur. Verify everything before your name goes on the user guide.
Module Six
The Jagged Frontier
Drafting correspondence · summarizing long documents · restructuring text · brainstorming · formatting · explaining concepts · translating between styles.
Specific MCO interpretation · TIS/TIG calculations · current personnel data · anything classified · anything where institutional knowledge isn't in its training set.
Your assignment — before next week
If you're continuing to Builder Orientation: bring a problem you want to build a tool to solve.
Course recap
Where you go from here
2 hours. Aspiring builders. From user to builder. Live builds. Debugging with AI. Decomposition exercises on a real problem you bring.
Bring: a problem in your workflow you want to solve with a tool.
Keep applying the six skills this week. Re-take this course on the EDD site any time. Add to your section's frontier map. Recommend the course to one person who hasn't taken it.
Thank you
You now have the framework that separates the 20% who stick from the 80% who quit. The tools will keep changing. The six skills won't.
Expert-Driven Development · MCD-Monterey · UNCLASSIFIED