diff --git a/.replit b/.replit
new file mode 100644
index 0000000..056af0b
--- /dev/null
+++ b/.replit
@@ -0,0 +1,40 @@
+modules = ["python-3.11", "nodejs-20", "go-1.25"]
+[agent]
+expertMode = true
+stack = "BEST_EFFORT_FALLBACK"
+integrations = ["github:1.0.0"]
+
+[nix]
+channel = "stable-25_05"
+
+[workflows]
+runButton = "Project"
+
+[[workflows.workflow]]
+name = "Project"
+mode = "parallel"
+author = "agent"
+
+[[workflows.workflow.tasks]]
+task = "workflow.run"
+args = "Start application"
+
+[[workflows.workflow]]
+name = "Start application"
+author = "agent"
+
+[[workflows.workflow.tasks]]
+task = "shell.exec"
+args = "python3 serve.py"
+waitForPort = 5000
+
+[workflows.workflow.metadata]
+outputType = "webview"
+
+[[ports]]
+localPort = 5000
+externalPort = 80
+
+[deployment]
+deploymentTarget = "static"
+publicDir = "docs"
diff --git a/TRAINING_CURRICULA.pdf b/TRAINING_CURRICULA.pdf
deleted file mode 100644
index 3305738..0000000
Binary files a/TRAINING_CURRICULA.pdf and /dev/null differ
diff --git a/TRAINING_CURRICULA_v5.md b/TRAINING_CURRICULA_v5.md
index 253163f..4236d95 100644
--- a/TRAINING_CURRICULA_v5.md
+++ b/TRAINING_CURRICULA_v5.md
@@ -1,11 +1,13 @@
-# Expert-Driven Development — Training Curricula v5.0
-## Revised February 2026
+# Expert-Driven Development — Training Curricula v5.1
+## Revised February 2026 · Course 6 Bonus capstone added April 2026
### What Changed From v4.0
The v4.0 curriculum restructured training around six 201-level skills. That was the right move. But it crammed two different audiences into the same opening course: people who need to *use* AI effectively and people who want to *build* tools with AI. Those are different training objectives for different populations.
-This revision adds a fifth course — AI Fluency Fundamentals — as a standalone 2-hour universal requirement. Every Marine, sailor, and civilian at MCCES takes this course. It teaches the six 201-level skills, the jagged frontier, centaur/cyborg work patterns, and quality judgment. No tool-building required. No Power Platform. Just the applied judgment skills that research shows actually predict sustained AI adoption.
+The v5.0 revision added a fifth course — AI Fluency Fundamentals — as a standalone 2-hour universal requirement. Every Marine, sailor, and civilian at MCCES takes this course. It teaches the six 201-level skills, the jagged frontier, centaur/cyborg work patterns, and quality judgment. No tool-building required. No Power Platform. Just the applied judgment skills that research shows actually predict sustained AI adoption.
+
+The v5.1 revision adds a sixth course — **Full-Stack AI-Assisted Development** — as a Bonus capstone for builders whose problems exceed the Power Platform envelope. It is elective, runs after Advanced Workshop, and teaches Marines to direct AI to write a complete Go + React + Docker application. The full per-module instructor script lives on the EDD site (`docs/courses/fullstack.html`); only the headline summary is reproduced in the table below.
The builder courses (Trainings 1–3) now assume students arrive with the 201 framework already internalized. They spend zero time on theory and 100% of time building.
@@ -115,15 +117,18 @@ This means Frontier Recognition can't be learned once and memorized. It has to b
## Training Overview
-### Five Courses, One Goal
+### Six Courses, One Goal
+
+Five core courses plus one Bonus capstone. The first five form the standard EDD program; Course 6 is an elective for builders whose problems exceed the Power Platform envelope.
| Course | Duration | Audience | Prerequisite | Outcome |
|--------|----------|----------|--------------|---------|
-| **AI Fluency Fundamentals** | 2 hours | All personnel | None | Understand the six 201 skills, recognize the jagged frontier, know when to trust AI output, map AI into your workflow |
-| **Builder Orientation** | 2 hours | Aspiring builders | AI Fluency Fundamentals | Build a working prototype, apply task decomposition and iterative refinement in practice |
-| **Platform Training** | 4 hours | Builders | Builder Orientation | Build 3 complete tools on Power Platform using centaur and cyborg work patterns |
-| **Advanced Workshop** | 4 hours | Experienced builders | At least one deployed tool | Map the frontier for your domain, build verification protocols, teach others |
-| **Supervisor Orientation** | 30 minutes | Leadership | None | Evaluate proposals, create permission culture, understand the apprentice problem |
+| **1. AI Fluency Fundamentals** | 2 hours | All personnel | None | Understand the six 201 skills, recognize the jagged frontier, know when to trust AI output, map AI into your workflow |
+| **2. Builder Orientation** | 2 hours | Aspiring builders | AI Fluency Fundamentals | Build a working prototype, apply task decomposition and iterative refinement in practice |
+| **3. Platform Training** | 4 hours | Builders | Builder Orientation | Build 3 complete tools on Power Platform using centaur and cyborg work patterns |
+| **4. Advanced Workshop** | 4 hours | Experienced builders | At least one deployed tool | Map the frontier for your domain, build verification protocols, teach others |
+| **5. Supervisor Orientation** | 30 minutes | Leadership | None | Evaluate proposals, create permission culture, understand the apprentice problem |
+| **6. Full-Stack AI-Assisted Development** *(Bonus, elective)* | 8 hours | Advanced builders | Advanced Workshop + at least one deployed tool | Direct AI to build and deploy a complete Go + React + Docker application; apply the Cyborg pattern at full-stack scale |
### Prerequisites
@@ -658,7 +663,7 @@ Mollick's research is explicit: workers are already using AI and hiding it. They
| Time | Module | Focus |
|------|--------|-------|
| 0:00–0:05 | Why This Matters Now | The 80% problem, DoW AI Strategy, rapid adoption mandate |
-| 0:05–0:12 | What EDD Is | Five courses, six 201 skills, research foundation (headline level) |
+| 0:05–0:12 | What EDD Is | Six courses (five core plus one Bonus capstone), six 201 skills, research foundation (headline level) |
| 0:12–0:20 | Your Role: Creating Permission | What "yes" looks like, what kills adoption, guard rails |
| 0:20–0:27 | Evaluating Proposals and Output | Four questions to ask, what quality looks like |
| 0:27–0:30 | The Apprentice Problem | Preserving junior development while gaining efficiency |
@@ -676,7 +681,7 @@ Mollick's research is explicit: workers are already using AI and hiding it. They
**The four-layer framework:**
1. SOP — Governance: How tools are proposed, reviewed, approved, and maintained
-2. Training — Education: Five courses from universal fluency to advanced building
+2. Training — Education: Six courses (five core plus one Bonus capstone) from universal fluency to advanced building
3. QA — Quality: Peer review, security assessment, and user verification
4. Community — Sustainability: Shared frontier maps, workflow playbooks, cross-unit mentoring
@@ -828,6 +833,7 @@ Mollick's research is explicit: workers are already using AI and hiding it. They
| v3.0 | Jan 2026 | Restructured around EDD SOP, added QA framework |
| v4.0 | Feb 2026 | Restructured around 201-level skills, jagged frontier, centaur/cyborg patterns, quality judgment exercises, frontier mapping |
| v5.0 | Feb 2026 | Added AI Fluency Fundamentals as universal 5th course; expanded research foundation with GDPval, Mollick delegation equation, Brynjolfsson skill-leveling, UK Government study; added delegation equation module; strengthened apprentice problem protocols; added junior development field to workflow playbooks; full research source appendix |
+| v5.1 | Apr 2026 | Added Course 6 (Full-Stack AI-Assisted Development) as a Bonus elective capstone; updated headline counts from "five courses" to "six courses (five core plus one Bonus capstone)"; per-module instructor script for Course 6 lives on the EDD site rather than in the master MD |
---
diff --git a/docs/courses/advanced.html b/docs/courses/advanced.html
index 54cd92c..e0bee88 100644
--- a/docs/courses/advanced.html
+++ b/docs/courses/advanced.html
@@ -70,7 +70,13 @@
- This course requires an instructor who has personally built and deployed at least 3 tools using AI assistance. The complex build in Module 2 requires real-time troubleshooting ability. You cannot effectively teach this course from a script alone. Students will encounter unexpected errors, frontier limitations, and integration failures. The instructor must be able to diagnose problems on the fly, guide students through iterative refinement, and recognize when a problem is a frontier issue versus a context/prompting issue.
-
-
-
-
Timing Breakdown
@@ -167,7 +164,7 @@
Timing Breakdown
- Total: 230 minutes (3 hours 50 minutes). The 10-minute buffer accounts for technical issues, extended Q&A, or students who need additional troubleshooting time during the complex build.
+ Total: 4 hours (230 min content + 10-min break + 20-min buffer). The buffer accounts for technical issues, extended Q&A, or students who need additional troubleshooting time during the complex build.
@@ -227,9 +224,6 @@
Agenda
-
- Total: 4 hours (includes 10-minute break and 10-minute buffer built into timing)
-
@@ -534,7 +528,8 @@
Final Verification
-
Debrief Questions (10 minutes)
+
Debrief (10 minutes)
+
Debrief Questions
Where did you switch modes? Why?
Where did the AI fail? Was it a frontier issue or a context issue?
@@ -625,8 +620,8 @@
Scenario 2: Power App Form Saves But Doesn't Validate
What is the DisplayMode property of the submit button? (This is where the actual problem usually is)
Does the text input show a red error indicator when you type an invalid number? (If yes, validation works but button logic is wrong; if no, validation formula is wrong)
-
Solution Approach: Set the submit button's DisplayMode to: If(TextInput_Phone.Valid, DisplayMode.Edit, DisplayMode.Disabled). This checks the Valid property of the input field, which reflects the validation formula result. The validation formula itself should be: IsMatch(TextInput_Phone.Text, "^\d{3}-\d{3}-\d{4}$")
-
Frontier Classification: This is a context issue. AI generated a validation formula, but the student didn't specify that the button's DisplayMode must respect the validation state. This is a common prompting gap: students ask for "validation" but don't specify all the integration points where validation must be enforced.
+
Solution Approach: Standalone Power Apps text input controls do not expose a .Valid property (that property only exists on data card controls inside a Form). Apply the validation pattern directly to the submit button's DisplayMode using regex pattern matching on the text: If(IsMatch(TextInput_Phone.Text, "^\d{3}-\d{3}-\d{4}$"), DisplayMode.Edit, DisplayMode.Disabled). This evaluates the regex on each keystroke and disables the submit button until the phone number matches the required XXX-XXX-XXXX format.
+
Frontier Classification: This is a context issue. AI generated a validation pattern but referenced a .Valid property that does not exist on standalone text inputs. The student didn't specify that the button's DisplayMode must enforce the validation directly. This is a common prompting gap: students ask for "validation" but don't specify all the integration points where validation must be enforced, and AI fills in plausible-but-wrong property references.
@@ -902,7 +897,12 @@
Group Debrief (5 minutes)
Who in your unit could benefit from learning these concepts?
-
Module 6: Workflow Playbook (20 minutes)
+
+
+
+
+
Module 6: Workflow Playbook 30 min
+
Duration: 30 minutes
Each participant produces a one-page playbook for one AI-integrated workflow
from their actual job. This is the final deliverable of the Advanced Workshop.
@@ -1029,6 +1029,11 @@
Completion Criteria: What a Finished Playbook Looks L
A completed playbook entry should have: a clear task name, realistic frequency, the correct mode (Centaur or Cyborg), 4-8 concrete steps with Human/AI labels, a verification checklist with 3-5 items, at least one known frontier issue, and a specific time savings estimate. If your playbook has fewer than 4 steps or no verification checklist, it is not detailed enough.
+
+
Wrap-Up
+
+ Close the workshop by reviewing the deliverables: each participant should leave with (1) a domain-specific frontier map, (2) a completed Unit Readiness Dashboard or equivalent multi-component build, (3) a 60-second teach-back rehearsed in front of peers, and (4) a one-page workflow playbook. Confirm that participants know how to submit their playbooks and frontier maps to the EDD GitHub Discussions for the broader community to learn from. Reinforce that they are now equipped to teach Platform Training and to mentor junior personnel using the protocols from Module 5.
+
@@ -1112,6 +1117,19 @@
Certification Recommendation
+
+
+
Next Steps
+
+
After This Course
+
+
Submit your frontier map and workflow playbook to the EDD GitHub Discussions: Discussions
+ Pick one example from the table and ask the room what verdict they’d give before revealing yours. Watch for: students defaulting to “AI Could Help” for everything — push back, force a real verdict.
+
+
@@ -779,6 +749,13 @@
Key Teaching Point
Module 4: The Trust Problem — Quality Judgment
Duration: 25 minutes
+
+
Time-Saver: Minimum Set
+
+ If running short on time, use the minimum set: Documents 1 and 3 only. Skip Document 2 (the SOP excerpt).
+
+
+
The Red Pen Review Exercise
This is the highest-value activity in the entire course. You will review three
@@ -1102,6 +1079,9 @@
Facilitation Tip
5-7), skipping the “verification needed” column, being unrealistic about
AI capabilities.
+
+ Failure mode: students will label everything “AI Could Help” to feel inclusive. Force them to pick one: which step would benefit MOST? Where would AI HURT?
+
@@ -1211,6 +1191,17 @@
Assessment Protocol
+
+ For self-paced students, the equivalent assessment is the Knowledge Check on the student page. Both cover the same six skills.
+
+
+
+
Closing the Session
+
+ Round-the-room close: 30 seconds per person on what they’ll attempt before next session. Drives accountability and seeds Module 1 of the next course.
+
+
+
Closing
You are now equipped with the same judgment framework that separates sustained AI
@@ -1233,7 +1224,7 @@
This course is budgeted at 8 hours, but the realistic target audience (one deployed tool, no full-stack background) needs two sittings. Default delivery: Day 1 — Modules 1-5 (frontier shift, env, backend, data, frontend; ends at “I can fetch JSON from my own server”). Day 2 — Modules 6-10 (pages, AI chat, auth, integration, deploy). The 1-day variant is an intensive offering for experienced builders only — most students will not complete in one sitting.
+
+
Courses 1–5 taught you to build tools on Power Platform — low-code applications
within the Microsoft ecosystem. That’s powerful, and most Marines will never need more.
@@ -316,6 +326,9 @@
Pacing Note
Students who fall behind can clone the checkpoint and continue from there
Prioritize: if time is short, skip Module 9 (External Integrations) — it’s the most independent module
+
+ Modules 7 and 9 are the heaviest — for time-pressed deliveries, demo Module 7 live and have students complete the prompts as homework; Module 9 (External Integrations) is the most independent and can be skipped without breaking the rest of the course.
+
@@ -352,7 +365,7 @@
Key Reveal
“This application has 8,400 lines of Go backend code, 4,400 lines of React frontend code,
35 API endpoints, 5 database backends, Microsoft Graph integration, CAC authentication,
and FIPS 140-3 compliance. It was built in days, not months, by one person
- working with AI. No team of contractors. No six-month timeline. No $2M budget.”
+ working with AI. Comparable contractor estimates run multi-month timelines and six-figure budgets. The point is not the savings claim — it’s that the work is now possible at a different scale of person-hours.”
@@ -869,6 +882,15 @@
Instructor Checkpoint: Data Layer Working
+
+
Common Issues
+
+
SQLite file permission errors on Windows — run as admin or specify an absolute path.
+
Connection-pool errors — close DB handles in defer db.Close().
+
ORM “foreign key constraint” errors — check that referenced tables exist before dependent tables.
+
+
+
Module 4 Complete When
@@ -1062,6 +1084,15 @@
Instructor Checkpoint: Multi-Page App
+
+
Common Issues
+
+
CORS errors — frontend running on :3000, backend on :8080. Add cors.AllowAll() middleware in dev only.
+
404 on refresh of SPA routes — frontend dev server needs historyApiFallback: true.
+
State lost between routes — verify React Router is wrapping the app, not nested.
+
+
+
Module 6 Complete When
@@ -1087,6 +1118,11 @@
Module 7: AI Chat Integration
and can query the application’s own data through tool use.
+
+
DoD network note
+
This module’s example uses api.openai.com for simplicity. For DoD networks, use Azure OpenAI Government or GenAI.mil endpoints, not the commercial OpenAI API. The code pattern is identical — only the base URL and auth header change. When you’re ready to deploy on a DoD network, ask the AI: “Convert this OpenAI client to Azure OpenAI Government with API key auth. Base URL is https://[your-resource].openai.azure.us/openai/deployments/[deployment-name].” Per MARADMIN 018/26, GenAI.mil is the default platform for AI tools handling DoD data.
+
+
API Key Required
@@ -1281,6 +1317,15 @@
Instructor Checkpoint: Auth Working
+
+
Common Issues
+
+
Cookies not sending on cross-origin requests — set credentials: 'include' on fetch AND Access-Control-Allow-Credentials: true on backend.
+
Role middleware fires on /login itself — exclude the login endpoint.
+
JWT expires immediately — check exp field is in seconds, not milliseconds.
+
+
+
Module 8 Complete When
@@ -1374,6 +1419,10 @@
Path B: Microsoft Graph
GET /api/v1/mail/summary — returns recent email subjects and senders
+
+
+ After Path B implementation, the calendar endpoint should return JSON like { "events": [...] } for the test tenant. Common failure mode: tenant-not-authorized 401 from the Graph API — the Marine’s account does not have calendar.read consent yet. Have a TA pre-grant consent on the test tenant before class, or skip the live test and review the prompt only.
+
@@ -1382,6 +1431,15 @@
Instructor Checkpoint: Integration Working
Path B: Calendar and mail endpoints return real data from Outlook.
+
+
Common Issues
+
+
API rate limits hit during testing — back off with exponential retry, don’t hammer.
+
Secrets accidentally committed — gitignore .env BEFORE first commit, not after.
+
Webhook receiver not reachable — use ngrok or similar for local dev; webhooks need a public URL.
+
+
+
Module 9 Complete When
@@ -1582,7 +1640,7 @@
Time Expectations
structure your conversations with AI.
- For comparison: a contractor would quote 6–12 months and $500K+ for the same application.
+ For comparison: a contractor would typically quote multi-month timelines and six-figure budgets — but verify against your unit’s actual procurement experience, not this number.
Student Build + Peer Review: When Something Breaks
-
40 min
-
-
-
Module 4
-
Decomposition Framework: Your Problem
-
20 min
-
-
-
Module 5
-
Wrap-Up & Assignment
-
10 min
-
-
-
Buffer Time
-
10 min
-
-
-
Total
-
120 min
-
-
-
-
-
-
-
-
Contingency: Technical Failures
-
If Power Platform, AI chat, or internet access is unavailable:
-
-
Option 1 (Preferred): If AI tools are unavailable during the live build, use the step-by-step walkthrough below as a guided demonstration. The instructor reads each prompt aloud, describes the expected output, and leads discussion on what the AI produced.
-
Option 2: Conduct paper-based decomposition exercise using whiteboard scenarios. Provide 3 sample problems (equipment tracker, leave request form, inventory dashboard) and have students work through full decomposition in pairs.
-
Option 3: Use offline debugging exercise (see Module 2 fallback below). Review pre-built broken code on paper, identify errors as a group, discuss how to frame debugging prompts.
-
Do NOT cancel the course. The decomposition skills are platform-independent and can be taught without live tools.
-
-
-
Agenda
@@ -215,6 +159,18 @@
Agenda
+
+
+
Contingency: Technical Failures
+
If Power Platform, AI chat, or internet access is unavailable:
+
+
Option 1 (Preferred): If AI tools are unavailable during the live build, use the step-by-step walkthrough below as a guided demonstration. The instructor reads each prompt aloud, describes the expected output, and leads discussion on what the AI produced.
+
Option 2: Conduct paper-based decomposition exercise using whiteboard scenarios. Provide 3 sample problems (equipment tracker, leave request form, inventory dashboard) and have students work through full decomposition in pairs.
+
Option 3: Use offline debugging exercise (see Module 2 fallback below). Review pre-built broken code on paper, identify errors as a group, discuss how to frame debugging prompts.
+
Do NOT cancel the course. The decomposition skills are platform-independent and can be taught without live tools.
+
+
+
Module 1: From User to Builder
@@ -242,6 +198,11 @@
Instructor Note
refresher.
+
+
+
Opening Check-In & Builder Mindset Framing
+
Open with a 2-minute check-in: “Who tried something with AI between Course 1 and today? What worked, what surprised you?” This surfaces real student usage and seeds the Builder Mindset framing. Common misconception: students think “builder” means “coder” — push back. Builder = anyone who turns recurring work into a tool, regardless of code complexity.
+
@@ -311,6 +272,8 @@
Step 2: Build with AI (20 min)
Narrate each prompt before typing it. Follow this sequence:
+
Click each prompt step to expand. Open them in order during live build.
+
When AI Output Doesn't Match
@@ -324,7 +287,7 @@
When AI Output Doesn't Match
If those three are right, the output is usable even if it doesn't match word-for-word. If the AI produces JavaScript syntax (forEach, =>), re-prompt: "Use Power Fx only, not JavaScript."
-
+ Prompt 1: Define the Problem (2 min)
@@ -342,7 +305,7 @@
When AI Output Doesn't Match
-
+ Prompt 2: Create the Data Structure (3 min)
@@ -394,7 +357,7 @@
When AI Output Doesn't Match
This is your starting point. From here, use AI to refine the auto-generated app to match your needs. The remaining prompts show how.
Subtask 1: ___________________________________
+ Pattern: [Centaur / Cyborg / Neither]
+ Frontier risk: ___________________________________
+
+Subtask 2: ___________________________________
+ Pattern: [Centaur / Cyborg / Neither]
+ Frontier risk: ___________________________________
+
+(repeat for as many subtasks as the project requires)
4. Potential Frontier Issues (what might AI struggle with?):
-
-
Example: Integration with existing CAC-authenticated system
-
+
- ___________________________________
+- ___________________________________
+(Example: Integration with existing CAC-authenticated system)
5. Simplest Useful Version (what is the MVP?):
-
+
MVP: ___________________________________
@@ -943,6 +889,11 @@
Closing the Session
Logistics reminder: Confirm date/time of Platform Training. Remind students to bring their prototype (on laptop, thumb drive, or screenshots if system access is limited).
+
+
+
Round-the-Room Close
+
Round-the-room close: 30 seconds per person on one specific thing they'll attempt before next session. Two-week deadline mentioned in the Assignment box. This drives accountability and surfaces planning gaps before students leave the room.
Problem: A request routing workflow — submit request, goes to the
@@ -413,7 +357,9 @@
When AI Output Doesn't Match
If those three are right, the output is usable even if it doesn't match word-for-word. If the AI produces JavaScript syntax (forEach, =>) or non-existent functions, re-prompt: "Use Power Fx only, not JavaScript."
-
+
Click each step to expand. Open in sequence during live build.
+
+ Step 1: SharePoint List Setup
@@ -502,7 +448,7 @@
Instructor Checkpoint: Verify Column Types
This is your starting point. From here, use AI to customize the auto-generated app. The prompt below shows how.
Steps 5-6 are extensions. The core build (Steps 1-4) is complete. Complete Steps 5-6 if time permits in class, or assign as post-class practice.
-
+ Step 5 (Extension): Requester Notification
@@ -847,7 +793,7 @@
Instructor Checkpoint: Test End-to-End Flow
-
+ Step 6 (Extension): Error Handling
@@ -984,26 +930,11 @@
Build #1 Success Indicators
-
Phase 3 — Human Verifies
- Test every path. Is the right person notified? What about edge cases? Submit
- test requests through every branch of your approval chain and verify the
- results against your whiteboard design.
-
-
-
Phase 4 — AI Refines
-
- Back in GenAI.mil, paste your test results: “The flow works for the first
- approver but skips the second. Here’s the flow logic: [paste]. The business
- rule is [explain]. Fix the routing logic.” Describe exactly what failed,
- what you expected, and what happened instead. Copy the corrected code back into
- Power Platform.
-
-
-
Phase 5 — Human Accepts
-
- Final review. Walk through the complete workflow one more time. Does it match
- the original design? Is it ready for users?
+ Phases 3-5 of the Centaur pattern (Verify, Refine, Approve) are embedded in
+ the instructor checkpoints above — Test Form Submission and Test Routing
+ exercise verification; the iteration prompts exercise refinement; the final
+ "Test End-to-End Flow" is the approval gate.
Compile all failure cases from the session into a shared frontier map.
@@ -1945,10 +1876,12 @@
Module 6: Frontier Map Update and Wrap
-
-
Completing Your Frontier Map
+
+
Facilitator Framing
- The table above is a starter example. Each student should add at least 2-3 rows specific to the task types they encountered during today's builds. A useful frontier map row names a specific task type (not a generic category), gives concrete examples of what AI produced correctly in the "Handles Well" column, describes a specific failure with enough detail to recognize it next time in the "Handles Poorly" column, and states a testable verification action (not just "check it"). If your rows read like the example above, make them more specific to your actual experience.
+ Use the table to drive group debrief — capture 2-3 unit-specific rows on
+ the whiteboard during this module. Students add their own rows on the student
+ page after class.
@@ -1959,6 +1892,22 @@
Assignment Before Advanced Workshop
Document three failure cases with specifics: what failed, how you caught it, how you fixed it
Identify one area where AI capability surprised you — something it did better than you expected
+
+
+
Facilitator Scaffolding
+
+ Frontier-map debrief (10 min): each student adds 2-3
+ unit-specific rows to the table. Whiteboard the most-novel rows. Watch for:
+ students conflating "AI struggled" with "AI cannot do this" — those are
+ different. Many tasks just need better decomposition.
+
+
+ Assignment briefing (5 min): explicitly call out the deadline
+ and what the deliverable looks like. Common pushback: "I won't get to deploy
+ because I lack tenant access" — answer: "Build it in the simulator,
+ document the deployment plan, and submit both."
+
-
+ Prerequisite Check: Are you ready for this course?
This course assumes you have:
@@ -361,10 +365,10 @@
Phase 1: Data Backend (Centaur Mode)
Self-Check: Phase 1
- Before moving to the next phase, verify: Does the output match your design?
- Are the data types correct? Do the relationships make sense? If you proceed
- with a flawed schema, you will have to redo the input interface and dashboard
- as well.
+ If your data backend is a single “sessions” table without separate
+ Students or Tutors tables, what reporting query would you find slow or
+ impossible? (Hint: “Total hours per tutor across all students” would
+ require denormalizing every session row.)
+ Debugging is a core skill for anyone building with AI. In this module you will
+ work through three scenarios individually, diagnosing root causes and identifying
+ fixes. Each scenario represents a common failure pattern in AI-assisted builds.
+ These are the same scenarios used in the live debrief — review your
+ diagnosis against the answer key before you present.
+
+
+
+
Scenario 1: Power Automate Flow Triggers But Sends Wrong Data
+
+ Situation: A Power Automate flow is supposed to trigger when a
+ SharePoint list item is updated, then send an email notification to the item
+ creator with the updated information. The flow triggers correctly, but the email
+ always contains the OLD data, not the updated data.
+
+
+ What you have already tried: “I rebuilt the email body
+ three times. I checked the SharePoint permissions. I re-authenticated the
+ connection. Nothing works.”
+
+
+
+ Show diagnosis: Root Cause & Solution
+
+ Root cause: This is a classic Power Automate timing issue. The
+ trigger “When an item is created or modified” fires immediately when
+ the update is detected, but it captures the item data at the moment of detection,
+ which may be before all fields have finished saving. This is especially common
+ with calculated columns or cascading updates.
+
+
Diagnosis questions to ask yourself:
+
+
Does the email contain the old data or blank data? (Old data suggests timing issue; blank suggests permissions issue.)
+
Are any of the fields calculated or lookup columns? (Calculated columns update asynchronously.)
+
Does the flow work correctly if you manually trigger it 30 seconds after updating? (Confirms timing issue.)
+
+
+ Solution: Add a “Delay” action of 5–10 seconds
+ immediately after the trigger, then use “Get item” to retrieve the
+ updated data explicitly rather than relying on the trigger output. The corrected
+ flow structure: Trigger → Delay (5 seconds) → Get item (by ID) →
+ Send email (using Get item output).
+
+
+ Frontier classification: This is NOT a frontier issue. This is a
+ platform limitation (Power Automate's trigger timing) that requires domain
+ knowledge to diagnose. AI can generate the flow, but it won't know about this
+ timing quirk unless you tell it.
+
+
+
+
+
Scenario 2: Power App Form Saves But Doesn’t Validate
+
+ Situation: A Power Apps form is supposed to validate that a phone
+ number is in the format XXX-XXX-XXXX before saving to a SharePoint list. The form
+ has a text input field with a validation formula, and the submit button should be
+ disabled if the phone number is invalid. The form saves successfully, but it
+ accepts phone numbers in any format, even completely invalid entries like
+ “abc123”.
+
+
+ What you have already tried: “I asked the AI to write a
+ validation formula three times. I tested different regex patterns. The formula
+ shows no errors in Power Apps, but it doesn't actually prevent invalid data from
+ being submitted.”
+
+
+
+ Show diagnosis: Root Cause & Solution
+
+ Root cause: The submit button's DisplayMode property isn't
+ checking the validation state correctly. A common AI-generated mistake is to
+ reference a .Valid property on a standalone text input, which does
+ not exist (only data card controls inside a Form expose .Valid).
+ The form allows submission because the button logic doesn't actually evaluate
+ the validation pattern.
+
+
Diagnosis questions to ask yourself:
+
+
What is the exact formula in the text input's validation? Is it referencing a property the control actually has?
+
What is the DisplayMode property of the submit button? (This is where the actual problem usually is.)
+
Does the text input show a red error indicator when you type an invalid number? (If yes, validation works but button logic is wrong; if no, validation formula is wrong.)
+
+
+ Solution: Apply the regex check directly on the button's
+ DisplayMode using IsMatch() against the input's text:
+ If(IsMatch(TextInput_Phone.Text, "^\d{3}-\d{3}-\d{4}$"), DisplayMode.Edit, DisplayMode.Disabled).
+ This evaluates on each keystroke and disables the submit button until the phone
+ number matches the required XXX-XXX-XXXX format.
+
+
+ Frontier classification: This is a context issue. AI generated
+ a validation pattern but referenced a property that doesn't exist on the
+ control. The student didn't specify all the integration points where validation
+ must be enforced, and AI filled in plausible-but-wrong property references.
+
+
+
+
+
Scenario 3: Dashboard Shows Stale Data
+
+ Situation: A Power BI dashboard pulls from an Excel file stored in
+ SharePoint. When the Excel file is updated, the dashboard doesn't show the new
+ data until you manually click “Refresh” in Power BI Desktop and
+ republish the report. You want the dashboard to automatically update when the
+ source file changes.
+
+
+ What you have already tried: “I set up a scheduled refresh
+ in the Power BI service. I checked the data source credentials. I even rebuilt
+ the data connection. The refresh runs successfully according to the logs, but
+ the dashboard still shows old data unless I manually republish from Power BI
+ Desktop.”
+
+
+
+ Show diagnosis: Root Cause & Solution
+
+ Root cause: This is a data connection configuration issue.
+ When the Power BI report was created, the data source was set to the local file
+ path (e.g., C:\Users\...\file.xlsx) instead of the SharePoint URL.
+ The scheduled refresh in Power BI Service is trying to refresh from the local
+ path, which doesn't exist in the cloud. The refresh “succeeds” but
+ retrieves no new data because it's looking in the wrong place.
+
+
Diagnosis questions to ask yourself:
+
+
When you created the data connection, did you use “Get Data → SharePoint Folder” or did you download the Excel file and use “Get Data → Excel”? (If the latter, this is the problem.)
+
In Power BI Desktop, go to Transform Data → Data source settings. What does the file path show? (If it shows a C:\ path instead of a SharePoint URL, this confirms the diagnosis.)
+
In the Power BI Service refresh history, does it show any warnings or errors, or does it show “Completed successfully”? (This scenario usually shows “Completed successfully” with zero rows refreshed.)
+
+
+ Solution: Rebuild the data connection using the SharePoint
+ connector. In Power BI Desktop: Get Data → SharePoint Folder → Enter
+ the SharePoint site URL → Navigate to the folder containing the Excel file
+ → Filter to your specific file → Load. Then configure scheduled
+ refresh in Power BI Service using the SharePoint Online credentials. This
+ establishes a cloud-to-cloud connection that can refresh automatically.
+
+
+ Frontier classification: This is a domain knowledge issue, not
+ a frontier issue. AI can generate the dashboard, but it can't know whether you
+ connected to a local file or a SharePoint URL unless you explicitly describe
+ your connection method. The AI assumes the “standard” approach but
+ doesn't know the nuances of your environment.
+
+
+
+
+
Key Insight
+
+ Every debugging session teaches you something about the frontier. When you find
+ a bug, ask: “Is this a frontier issue (AI cannot do this reliably) or a
+ context issue (I did not give AI enough information)?” Document the answer.
+ Over time, your frontier map becomes more precise and your prompts become more
+ effective.
+
+
+
+
+
-
Module 3: Verification Protocols and QA
+
Module 4: Verification Protocols and QA
Duration: 45 minutes
@@ -599,179 +770,6 @@
What This Exercise Teaches
-
-
-
Module 4: Debugging Practice
-
Duration: 45 minutes
-
-
- Debugging is a core skill for anyone building with AI. In this module you will
- work through three scenarios individually, diagnosing root causes and identifying
- fixes. Each scenario represents a common failure pattern in AI-assisted builds.
-
-
-
-
Scenario 1: Approval Flow Sends to the Wrong Person
-
- A Power Automate flow is supposed to route purchase requests based on dollar
- amount. Requests under $2,500 go to the Department Head for approval. Requests
- $2,500 and above go to the Commanding Officer.
-
-
-
-
Symptom
-
Request for $1,500 office supplies submitted by Cpl Torres.
-Expected: Routed to Department Head (Maj Williams).
-Actual: Routed to CO (Col Richardson).
-
-Request for $2,500 equipment purchase submitted by SSgt Park.
-Expected: Routed to CO (Col Richardson).
-Actual: Routed to CO (Col Richardson). [Correct]
-
-All requests at $2,500 exactly are handled correctly. Only requests BELOW
-$2,500 are being sent to the wrong approver.
-
-
-
- Try to diagnose this yourself first
-
Think about what could cause this symptom. What would you check? Write down your hypothesis before reading the diagnosis guide below.
-
-
-
Diagnosis guide:
-
-
What is the condition that controls routing?
-
If requests at exactly $2,500 go to the CO correctly, but requests below $2,500 also go to the CO, what comparison operator would cause this?
-
What should the condition be instead?
-
-
-
- Self-Check: Root Cause and Fix
-
- Root cause: The condition logic uses > (greater than)
- instead of >= (greater than or equal to) for the CO threshold, or
- equivalently, the Department Head condition uses < instead of
- <=. Because the threshold check for “send to Department Head”
- is amount < 2500 but was written as amount > 2500 routing
- to CO, every amount that is not > 2500 falls through — except the
- logic is inverted: the condition likely reads “if amount >= 0, send to CO”
- as a catch-all because the < 2500 branch was written as
- > 2500.
-
-
- Fix: Change the condition to: “If amount is greater than or equal
- to 2500, route to CO. Otherwise, route to Department Head.” Verify by testing with
- values at $2,499, $2,500, and $2,501.
-
-
-
-
-
Scenario 2: Gallery View Shows All Items Instead of Filtered
-
- A Power App has a gallery that should display only overdue items —
- items whose due date has passed and whose status is not “Complete.”
-
-
-
-
Symptom
-
The "Overdue Items" gallery shows ALL items from the task list,
-including items that are marked "Complete" and items whose due date
-is in the future.
-
-Total items in list: 47
-Items actually overdue: 12
-Items displayed in gallery: 47
-
-
-
- Try to diagnose this yourself first
-
Think about what could cause this symptom. What would you check? Write down your hypothesis before reading the diagnosis guide below.
-
-
-
Diagnosis guide:
-
-
What filter formula is the gallery using? Check the Items property of the gallery.
-
Is the filter comparing dates correctly? (Common issue: comparing a date value to a text string.)
-
Is the status check using the correct column name and value? (Common issue: column display name vs. internal name.)
-
-
-
- Self-Check: Root Cause and Fix
-
- Root cause: The filter formula is likely malformed or not applied at all.
- Common causes include: (1) the gallery Items property points to the raw data
- source without a Filter() function, (2) the date comparison uses a text
- string like "Today" instead of the Today() function, or (3) the
- status column is referenced by its display name (e.g., “Status”) when Power
- Apps requires the internal name (e.g., “OData_Status”).
-
-
- Fix: Set the gallery Items property to:
- Filter(TaskList, DueDate < Today() && Status.Value <> "Complete").
- Verify by checking the displayed count against a manual count of overdue, incomplete items.
-
-
-
-
-
Scenario 3: Dashboard Numbers Don’t Match Source Data
-
- A Power BI dashboard reports completion percentage for a training tracker.
- Leadership questions the numbers because they do not match a manual count.
-
-
-
-
Symptom
-
Dashboard shows: 85% training completion rate.
-Manual count shows: 67% training completion rate.
-
-Source data: 47 Marines, 12 training events each = 564 total slots.
-Dashboard appears to be counting something differently than
-"completed slots / total slots."
-
-
-
- Try to diagnose this yourself first
-
Think about what could cause this symptom. What would you check? Write down your hypothesis before reading the diagnosis guide below.
-
-
-
Diagnosis guide:
-
-
How is the dashboard calculating “completion rate”? Is it counting rows vs. distinct values?
-
If a Marine completed 10 of 12 events, does the dashboard count that Marine as “85% complete” or as “1 complete Marine”?
-
Are there duplicate rows in the source data inflating the count?
-
-
-
- Self-Check: Root Cause and Fix
-
- Root cause: The counting methodology differs between the dashboard and
- the manual count. The most likely cause is that the dashboard is counting
- rows (each training completion record) rather than
- distinct values (unique Marine-event combinations). If some Marines have
- duplicate completion records (e.g., re-certifications, data entry errors), the row count
- is inflated. Alternatively, the dashboard may be calculating the average completion rate
- per Marine (averaging individual percentages) rather than the overall rate (total completed
- slots divided by total slots), which produces a different number.
-
-
- Fix: Align the calculation method. Use
- DISTINCTCOUNT instead of COUNT to avoid duplicates. Ensure the
- formula is: total completed distinct slots / total expected slots. Verify by manually
- calculating the rate for a small subset (e.g., one platoon) and comparing.
-
-
-
-
-
Key Insight
-
- Every debugging session teaches you something about the frontier. When you find
- a bug, ask: “Is this a frontier issue (AI cannot do this reliably) or a
- context issue (I did not give AI enough information)?” Document the answer.
- Over time, your frontier map becomes more precise and your prompts become more
- effective.
-
-
-
-
Module 5: Teaching Others — The 201 Multiplier
@@ -883,7 +881,12 @@
60-Second Teach Template
-
Deliverable: Workflow Playbook
+
+
+
+
+
Module 6: Workflow Playbook 30 min
+
Duration: 30 minutes
Your final deliverable is a one-page playbook for one AI-integrated workflow
from your actual job. Study the completed example below, then create your own
@@ -894,7 +897,7 @@
Deliverable: Workflow Playbook
Start simple. Your first playbook does not need to be this detailed. Begin with: Task, Mode, 3–5 steps with Human/AI labels, and one known frontier issue. Add detail as you use the playbook over time and discover what matters.
-
Completed Example: Weekly Training Schedule Publication
+
Completed Example: Weekly Training Schedule Publication
@@ -970,7 +973,7 @@
Workflow Playbook Completion Criteria
-
Blank Workflow Playbook Template
+
Blank Workflow Playbook Template
Your Workflow Playbook
@@ -1034,22 +1037,32 @@
Assignment
Capstone Deliverable
-
Documentation Package
+
Submission
- Complete the full documentation package (User Guide, Replication Guide,
- Adaptation Guide, Maintenance Guide) for your tool. The package must be
- thorough enough that another developer could rebuild your tool from the
- Replication Guide alone.
+ Submit your completed Workflow Playbook (Module 6 deliverable) along with at
+ least one frontier-map row that captures a real lesson from your build. Post
+ both to EDD Discussions.
+
+
+
+
What's next: Course 6 — Full-Stack AI-Assisted Development
+
You've moved from individual builder to teaching others. Full-Stack AI-Assisted Development is the capstone — build and deploy a complete web app from scratch using AI as your development partner.
+
+
+
Knowledge Check
Knowledge Check
+
+ Module 1: Frontier Mapping — Questions
+
The BCG-Harvard study found that workers who applied AI beyond the frontier without knowing it performed how much worse than those without AI?
@@ -1128,6 +1141,11 @@
Knowledge Check
+
+
+
+ Module 2: Complex Build — Questions
+
In the Tutoring Management System example, the data backend uses centaur mode while the input interface uses cyborg mode. What principle drives this mode selection?
@@ -1206,162 +1224,177 @@
Knowledge Check
+
+
+
+ Module 3: Group Debugging — Questions
+
-
The QA checklist lists "Source verification" as item #1. According to the timed QA exercise, why is this the most important step?
+
In Scenario 1 (Power Automate Flow Triggers But Sends Wrong Data), the flow fires on item creation but the email body sometimes contains stale or empty fields. What is the root cause?
-
Source verification is item #1 because it is the easiest step to skip and the most dangerous when you do. AI generates plausible-sounding regulation numbers, form numbers, and citations that do not correspond to real documents. Most people catch formatting errors quickly but miss fabricated references unless they explicitly verify every citation against authoritative sources.
+
This is a classic Power Automate timing issue. The trigger captures the item the moment a row is created, but downstream fields populated by other flows or by SharePoint defaults have not yet been written. The fix is either to add a delay action before reading the fields, switch to a "When an item is modified" trigger with a "Status = Ready" guard, or fetch the item fresh with a Get item action immediately before the email step. This is platform timing, not a logic bug — recognizing it saves hours.
-
-
In the timed QA exercise, the AI-generated SOP contains steps that are out of logical order (receiving an S-1 orientation brief before in-processing at S-1). What type of QA check catches this error?
+
+
In Scenario 2 (Power App Form Saves But Doesn't Validate), the AI suggested If(TextInput_Phone.Valid, DisplayMode.Edit, DisplayMode.Disabled). Why won't this work?
-
This is a logic check error. The question is whether the reasoning and sequence make sense: logically, you would in-process at S-1 before receiving an orientation brief from S-1. Domain knowledge tells you the steps are reversed. Format compliance would catch the inconsistent numbering (a separate error in the same document), but the logical ordering issue requires understanding the actual workflow.
+
The AI confidently invoked a .Valid property that does not exist on standalone text inputs. The working fix uses regex pattern matching directly on the text: If(IsMatch(TextInput_Phone.Text, "^\d{3}-\d{3}-\d{4}$"), DisplayMode.Edit, DisplayMode.Disabled). This is a frontier issue, not a context issue — the AI has been trained on enough Power Apps code to invent plausible-looking properties that do not exist. Verify any unfamiliar property reference by searching Microsoft Learn before trusting it.
-
-
The timed QA exercise demonstrates that AI can generate a draft SOP in seconds, but finding the five errors takes several minutes of careful review. What does this teach about the value proposition of AI-assisted work?
+
+
After resolving a debugging scenario, the course suggests asking: "Is this a frontier issue or a context issue?" What is the practical difference between the two?
-
The exercise demonstrates the core value proposition: AI handles the generation (fast, repetitive work) while humans handle the verification (expert judgment). Even with QA time included, the total time is less than writing from scratch. But the QA step is where the actual value is created -- it is what separates a useful draft from a document with fabricated references and logical errors.
+
The distinction is critical for how you respond. A context issue means you can fix the problem by providing better information in your prompt -- the AI is capable, it just lacked the right input. A frontier issue means AI cannot reliably handle this task regardless of how you prompt it, so you need to plan for manual work or extra verification. Documenting which category each bug falls into makes your frontier map more precise and your future prompts more effective.
In Debugging Scenario 1, a $1,500 request routes to the CO instead of the Department Head. The key diagnostic question is: "If requests at exactly $2,500 go to the CO correctly, but requests below $2,500 also go to the CO, what comparison operator would cause this?" What is the root cause?
+
+
The QA checklist lists "Source verification" as item #1. According to the timed QA exercise, why is this the most important step?
-
The root cause is inverted or incorrect comparison operators in the condition logic. For example, if the condition reads "if amount >= 0, send to CO" as a catch-all because the Department Head branch was incorrectly written, every request falls through to the CO. The fix is to ensure the condition properly routes amounts below $2,500 to the Department Head and amounts at or above $2,500 to the CO, then verify with test values at $2,499, $2,500, and $2,501.
+
Source verification is item #1 because it is the easiest step to skip and the most dangerous when you do. AI generates plausible-sounding regulation numbers, form numbers, and citations that do not correspond to real documents. Most people catch formatting errors quickly but miss fabricated references unless they explicitly verify every citation against authoritative sources.
-
-
In Debugging Scenario 2, a gallery view that should show only overdue items displays all 47 items instead of the expected 12. Which of the following is the most systematic first step in diagnosing this issue?
+
+
In the timed QA exercise, the AI-generated SOP contains steps that are out of logical order (receiving an S-1 orientation brief before in-processing at S-1). What type of QA check catches this error?
-
The systematic first step is to inspect the gallery's Items property. If the gallery shows all items, the most likely cause is that the filter is missing, malformed, or referencing incorrect column names. Common AI errors include: pointing to the raw data source without a Filter() function, using a text string "Today" instead of the Today() function, or using a column's display name when Power Apps requires the internal name.
+
This is a logic check error. The question is whether the reasoning and sequence make sense: logically, you would in-process at S-1 before receiving an orientation brief from S-1. Domain knowledge tells you the steps are reversed. Format compliance would catch the inconsistent numbering (a separate error in the same document), but the logical ordering issue requires understanding the actual workflow.
-
-
After resolving a debugging scenario, the course suggests asking: "Is this a frontier issue or a context issue?" What is the practical difference between the two?
+
+
The timed QA exercise demonstrates that AI can generate a draft SOP in seconds, but finding the five errors takes several minutes of careful review. What does this teach about the value proposition of AI-assisted work?
-
The distinction is critical for how you respond. A context issue means you can fix the problem by providing better information in your prompt -- the AI is capable, it just lacked the right input. A frontier issue means AI cannot reliably handle this task regardless of how you prompt it, so you need to plan for manual work or extra verification. Documenting which category each bug falls into makes your frontier map more precise and your future prompts more effective.
+
The exercise demonstrates the core value proposition: AI handles the generation (fast, repetitive work) while humans handle the verification (expert judgment). Even with QA time included, the total time is less than writing from scratch. But the QA step is where the actual value is created -- it is what separates a useful draft from a document with fabricated references and logical errors.
+
+
+
+ Module 5: Teaching Others — Questions
+
Mollick's research shows workers are already using AI but hiding it. According to the course, what is the organizational fix for this "shadow AI culture"?
“make it better” but exact instructions on what to fix.
-
Pass 2 — Add POC, Liberty Radius, Remove Redundancy
-
- Refinement Prompt
-
Add the duty officer contact info: Capt Rodriguez, (831) 555-0147.
+
+ See Pass 2 (intermediate iteration)
+
Pass 2 — Add POC, Liberty Radius, Remove Redundancy
+
+ Refinement Prompt
+
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.
-
-
- AI Output — Pass 2
-
Weekend Safety Brief
+
+
+ AI Output — Pass 2
+
Weekend Safety Brief
Listen up. We made it through another week. Before I cut you
loose, here's what you need to know.
@@ -473,12 +475,13 @@
Pass 2 — Add POC, Liberty Radius, Remove Redundancy
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.
-
+
+ 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
@@ -1008,6 +1011,7 @@
Exercise: Workflow Mapping
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.
+
Identify one frontier risk for your task — what could go wrong if you delegate this to AI?
Use this template to map your workflow:
@@ -1123,7 +1127,12 @@
Your Personal Frontier Map
Your Assignment
-
This Week: Try It
+
Optional: Try It This Week
+
+ Recommended next step before the capstone — a low-stakes warm-up to surface real
+ friction before you build the formal frontier map. The capstone (below) is the explicit
+ deliverable for this course.
+
Pick one recurring task from your workflow map (Module 5)
Try using AI on it this week
@@ -1145,6 +1154,11 @@
For Those Continuing to Builder Orientation
201 skills — there you will put them to work building actual tools.
+
+
What's next: Course 2 — Builder Orientation
+
You've finished AI Fluency Fundamentals. The next step on the Builder Path is Builder Orientation, where you'll build your first tool from start to finish.
+
+
What You Now Know
You are now equipped with the same judgment framework that separates sustained AI
@@ -1190,8 +1204,10 @@
Personal Frontier Map
Knowledge Check
+
Click each module to expand its questions. Your answers save automatically.
-
+
+ Module 1 Questions (3 questions)
Module 1: Why 80% Quit
@@ -1268,9 +1284,10 @@
Module 1: Why 80% Quit
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 Questions (3 questions)
Module 2: The Six 201-Level Skills
@@ -1347,9 +1364,10 @@
Module 2: The Six 201-Level Skills
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 Questions (3 questions)
Module 3: The Delegation Equation
@@ -1426,58 +1444,59 @@
Module 3: The Delegation Equation
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 Questions (3 questions)
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?
+
+
In the award narrative document, the AI claimed the Marine "increased mission readiness by 47%." What error category 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.
+
Hallucinated metrics are the most common category of AI error in military writing. AI fabricates precise-sounding numbers to make claims more authoritative, but there is no audit, no baseline measurement, and no methodology behind them. Strong award narratives use verifiable accomplishments, not AI-generated percentages.
-
-
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?
+
+
What category of error is "MARADMIN 045/26" when the document references it as a citation?
-
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?"
+
Fabricated references — fake MARADMINs, made-up policy numbers, invented sources — are the single most dangerous category of AI error because they look authoritative but cannot be verified.
@@ -1505,9 +1524,10 @@
Module 4: Red Pen Review
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 Questions (3 questions)
Module 5: Centaur and Cyborg Modes
@@ -1584,9 +1604,10 @@
Module 5: Centaur and Cyborg Modes
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 Questions (3 questions)
Module 6: Frontier Mapping
@@ -1663,7 +1684,7 @@
Module 6: Frontier Mapping
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.
This course is budgeted at 8 hours, but the realistic target audience (one deployed tool, no full-stack background) needs two sittings. Default delivery: Day 1 — Modules 1-5 (frontier shift, env, backend, data, frontend; ends at “I can fetch JSON from my own server”). Day 2 — Modules 6-10 (pages, AI chat, auth, integration, deploy). The 1-day variant is an intensive offering for experienced builders only — most students will not complete in one sitting.
+
+
Courses 1–5 taught you to build tools on Power Platform — low-code applications
within the Microsoft ecosystem. That’s powerful, and most Marines will never need more.
@@ -185,15 +195,6 @@
Six Principles of AI-Assisted Full-Stack Development
Don’t say “it doesn’t work” — say “I get this error when I run go build.”
-
-
-
Module 1 Complete When
-
-
Students can articulate the four layers of a web application
-
Students understand they won’t “learn to code” — they’ll learn to direct AI to code for them
-
Students have seen the live Heywood demo and understand the target
-
-
@@ -301,14 +302,8 @@
Agenda & Timing
Pacing Note
- This is an aggressive pace. Some students will fall behind, especially in Modules 3–4 (backend)
- and Module 7 (AI chat). That’s expected. The instructor should:
+ This is an aggressive pace. If you fall behind in Modules 3–4 (backend) or Module 7 (AI chat), that’s normal. If you’re tight on time, skip Module 9 (External Integrations) — it’s the most independent and can be done later. Plan two sittings minimum.
-
-
Have a pre-built “checkpoint” version of the app at each module boundary
-
Students who fall behind can clone the checkpoint and continue from there
-
Prioritize: if time is short, skip Module 9 (External Integrations) — it’s the most independent module
-
@@ -343,7 +338,7 @@
Key Reveal
“This application has 8,400 lines of Go backend code, 4,400 lines of React frontend code,
35 API endpoints, 5 database backends, Microsoft Graph integration, CAC authentication,
and FIPS 140-3 compliance. It was built in days, not months, by one person
- working with AI. No team of contractors. No six-month timeline. No $2M budget.”
+ working with AI. Comparable contractor estimates run multi-month timelines and six-figure budgets. The point is not the savings claim — it’s that the work is now possible at a different scale of person-hours.”
@@ -476,9 +471,7 @@
Module 2: Environment Setup
Pre-Class Preparation
- Ideally, send setup instructions to students before class so they arrive
- with tools installed. In practice, expect 30–50% to need help during this module.
- Have the instructor and any TAs circulate during this time.
+ Install the tools in Section 2.1 before starting Module 3. Plan for 30–60 minutes if this is your first time installing developer tools. Expect at least one tool to need a system restart or PATH fix — that is normal. Have the EDD Discussions tab open in another window if you get stuck.
@@ -1003,12 +996,16 @@
Module 7: AI Chat Integration
and can query the application’s own data through tool use.
+
+
DoD network note
+
This module’s example uses api.openai.com for simplicity. For DoD networks, use Azure OpenAI Government or GenAI.mil endpoints, not the commercial OpenAI API. The code pattern is identical — only the base URL and auth header change. When you’re ready to deploy on a DoD network, ask the AI: “Convert this OpenAI client to Azure OpenAI Government with API key auth. Base URL is https://[your-resource].openai.azure.us/openai/deployments/[deployment-name].” Per MARADMIN 018/26, GenAI.mil is the default platform for AI tools handling DoD data.
+
+
API Key Required
You need an OpenAI API key (OPENAI_API_KEY) or Azure OpenAI credentials
- for this module. If keys aren’t available, the instructor should demo this module live
- and you can add it later. Alternatively, have one shared API key for the class.
+ for this module. If you don’t have an OpenAI key, skip the live calls in this module — read the prompts, examine the expected code, and complete this module against a mock response. Wire in real keys later when access is approved. The Module 7 lessons (chat backend, tool use, system-prompt design) work without live calls.
@@ -1475,6 +1472,271 @@
Course 6 Complete
+
+
+
+
+
+
Knowledge Check
+
+ Six questions covering the Six Principles, the 3-Minute Rule, and the Cyborg pattern at
+ full-stack scale. Score 80% or higher to pass. There is a 2-minute minimum reading time
+ before you can submit.
+
+
+
Click each module to expand its questions. Your answers save automatically.
Foundations: How AI-Assisted Full-Stack Work Differs
+
+
+
In Principle 1 (“The Conversation is the IDE”), what role does your code editor play during a full-stack build?
+
+
+
+
+
+
+
+
The editor is for verification and execution, not authoring. The chat is where you direct work; the editor is where you confirm it.
+
+
+
+
+
You are following Principle 2 (Scaffold → Flesh Out → Integrate). The AI just produced 600 lines that wire up the backend, the database, and three frontend pages in one shot. What should you do?
+
+
+
+
+
+
+
+
The point of Scaffold → Flesh Out → Integrate is to never try to build everything at once. Start with the smallest thing that runs end-to-end, then add one capability at a time.
+
+
+
+
+
You hit a runtime error. You spend three minutes reading the stack trace and trying small tweaks in the editor. The 3-Minute Rule says: what next?
+
+
+
+
+
+
+
+
The 3-Minute Rule exists because manual debugging of AI-generated code is slow and tends to introduce a second bug. The AI wrote the code; the AI knows the most recent context; let it fix the error.
Principle 4 says “Incremental Deployment — deploy after every major feature.” Why is waiting until the end to deploy actually slower?
+
+
+
+
+
+
+
+
Environment issues (paths, secrets, permissions, container quirks) are harder to debug when they all surface at once. Catch them one at a time after each major feature.
+
+
+
+
+
Principle 5 (Interface-First Design) tells you to specify what a component does before how. Which prompt below best follows that principle?
+
+
+
+
+
+
+
+
Interface-first prompts describe inputs, outputs, and behavior — not implementation details. The AI will pick a reasonable implementation; you keep authority over the contract.
Course 3 introduced the Cyborg pattern (fluid, continuous human/AI integration). Course 6 applies it at full-stack scale. Which sentence best describes the difference at this scale?
+
+
+
+
+
+
+
+
Cyborg with denser verification is the right framing. The 3-Minute Rule, incremental deployment, and the “paste the error back” reflex are all ways of keeping the loop tight enough to catch problems before they compound.
You’ve shipped a complete full-stack application. Three suggested next moves:
+
+
Rebuild from scratch. Repeat the course without the prompts. The fastest path to fluency is doing it twice.
+
Post your repo to EDD Discussions. Get feedback from other builders. Inspire someone in their first week.
+
Start a real project for your unit. Use the Course 3 Problem Definition Worksheet to scope it. Apply the Course 4 Workflow Playbook to plan it. Ship something.
Start here: AI Fluency Fundamentals (required for everyone). After completing it, continue with Builder Orientation if you want to build tools, or stop here if you only need daily AI skills.
Builder Path: AI Fluency Fundamentals → Builder Orientation → Platform Training → Advanced Workshop → Full-Stack AI-Assisted Development (bonus)
+
Note: Course 5 (Supervisor Orientation) is a 30-minute leadership briefing for those who supervise builders. It is not self-paced material — see the instructor catalog if you supervise builders.