# AI-Augmented IT Project Manager

> Infrastructure Transformations & Transitions — plan better, manage risk smarter, communicate clearer, and deliver with less firefighting. 8 weeks, artifact-driven.

- **Audience:** IT/Infra Project Managers · Transition & Transformation PMs · Program Managers · Delivery Leads
- **Level:** Intermediate (PM experience assumed; no AI background)
- **Duration:** 8 weeks · 4–6 h/week
- **Modules:** 8
- **Pass mark:** 70%
- **Interactive version:** https://ragentic.netlify.app/#/courses/ai-it-pm

**This file is generated from the course data by `scripts/build-notes.mjs`. Edit the course data, not this file.**

## The world of this course

**Organisation:** Northgate Retail Group

Your employer for the next 8 weeks: a retailer — 12,000 employees, 300 stores, HQ in Leeds.
You are the PM of Program Horizon, which has two workstreams: (1) EUC TRANSFORMATION — 8,000 endpoints
from Windows 10/SCCM to Windows 11/Intune, persona-based rings; (2) DATACENTER TRANSITION — exit the
aging Manchester DC (lease expires in 11 months, immovably) to colocation + Azure, 214 servers in waves.
Two workstreams, one budget, one steering committee, one you.

---

## Phase 1 — Foundations & Prompt Craft (weeks 1–2)

### Module 1 — AI for Infrastructure PMs: The New Operating System

**Guiding question:** What can AI actually take off my plate on a complex program — and what must never leave it?

**Outcome:** Calibrate what AI does well and badly across the project lifecycle, set up your toolkit, and map AI capability onto YOUR program's workload — with the co-pilot/autopilot line drawn from day one.

**Delivery lens:** You run programs where a slipped dependency costs weeks and a bad cutover makes the news. AI won't attend the steering committee for you — but it can draft, analyse, correlate, and rehearse at a speed that changes what one PM can carry. The judgment, the relationships, and the accountability stay yours.

**Apply-at-work mission — Tool up and map your load:** Set up your AI toolkit (approved work tool + learning sandbox). List your 10 most time-consuming recurring PM tasks (status, minutes, RAID upkeep, plan revisions…), estimate hours/week each, and mark which are AI-assistable. That map is your course focus — and your Week 8 before/after baseline.

**Reflection:** Which parts of my PM week are judgment and relationships, and which are actually document manufacturing? What did the split reveal about where my time really goes?

#### Resources

- [Andrej Karpathy — Intro to Large Language Models](https://www.youtube.com/watch?v=zjkBMFhNj_g) — The one-hour mental model — watch it as a PM: what does "plausible text generator" mean for a risk register draft?
  - Explain what "plausible text generator" implies for an AI-drafted risk register
  - Say why an AI draft is a starting point to verify, never a plan to trust
  - Carry the two-file mental model into governance conversations about "the AI"
- [Microsoft Learn — Fundamentals of Generative AI](https://learn.microsoft.com/en-us/training/modules/fundamentals-generative-ai/) — Enterprise-flavoured grounding in the vocabulary your governance discussions will use.
  - Speak the enterprise generative-AI vocabulary your governance discussions use
  - Separate what the model does from what the platform wraps around it
  - Place AI against the PM tooling your program already runs
- [Microsoft — M365 Copilot overview](https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-overview) — The tool most enterprise PMs will actually be handed: what it sees (your tenant) and why that matters for project data.
  - State what M365 Copilot sees inside your tenant and why that matters for project data
  - Explain to sponsors why tenant boundary changes the data-handling calculus
  - Pick the PM tasks Copilot is licensed and positioned to accelerate
- [PMI — AI in project management (reports & resources)](https://www.pmi.org/learning/thought-leadership/ai-impact) — The professional body's view: where AI lands across the PM discipline — useful framing, sceptical reading encouraged.
  - Read the profession's own view of where AI lands across PM — sceptically
  - Separate PMI's evidence from its aspiration
  - Form a defensible line on what AI does and does not change about the role
- [Claude / ChatGPT / Perplexity — toolkit setup](https://claude.ai/) — Set up your kit: approved work tool, learning sandbox, and a search-grounded tool for current-fact questions.
  - Set up an approved work tool, a sandbox, and a search-grounded tool for current facts
  - Know which tool to reach for when a question needs today's facts
  - Have your kit ready so later weeks are hands-on, not setup

#### In-world ticket queue

> Day one as Program Horizon's PM. The previous PM left "for personal reasons" — along with a plan
> last updated 9 weeks ago, a RAID log with 4 entries (all closed), and 200 unread emails. The Manchester
> lease clock is already running: 11 months. Diane's welcome: "I don't need a hero. I need to never be surprised."

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0001 | P1 | you, to yourself | Inherited plan review — 9 weeks stale, both workstreams |
| HZN-0002 | P1 | Diane Moreau | Steering committee in 10 days — first impression status |
| HZN-0003 | P2 | Sam Kowalski | "Quick sync?" — Sam wants to talk ring 1 scope "informally" |
| HZN-0004 | P2 | Victor Alonso | Vendor kickoff minutes from Nexon — 14 pages, unread |

#### Project — AI Capability Map for Your Program

Set up your toolkit, then map your real workload: list your 10 most time-consuming recurring PM tasks with hours/week estimates (status reporting, minutes, RAID upkeep, plan revisions, stakeholder updates, CR paperwork…). For each, test AI on a sanitised sample and grade the assist: transformative / helpful / marginal / unsafe. Include at least one EUC-type and one datacenter-type artifact in your tests. The result is your personal AI Capability Map — and your Week 8 baseline.

**Deliverable:** playbook/w01-capability-map.md — the toolkit decision, the 10-task map with hours + grades, and one caught AI mistake.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Honest workload map | 30% | Ten real tasks with defensible hour estimates — the actual shape of your week, not the job description. |
| Tested, not guessed | 30% | Each grade backed by an actual trial on a sanitised sample; at least one surprise in either direction documented. |
| Both domains sampled | 20% | At least one endpoint/EUC artifact and one datacenter/migration artifact among the trials. |
| Critical catch | 20% | One confidently-wrong AI output caught and verified wrong — the habit that anchors everything else. |

#### Scenario drills

**Drill 1.** Day one on Program Horizon. Diane's standard is "never be surprised." Four things landed at once: HZN-0001 (an inherited plan, 9 weeks stale), HZN-0002 (steering committee in 10 days, first-impression status), HZN-0003 (Sam wants an "informal" chat about ring 1 scope), HZN-0004 (14 pages of unread Nexon kickoff minutes).

**Task:** Assign each item AI's honest role — summarise/extract, draft, suggest questions, or stay out (human judgment only) — and rank the four by which most threatens Diane's "no surprises" if you let it sit. Name the one AI must not touch.

**Drill 2.** HZN-0004 — 14 pages of Nexon kickoff minutes you have not read. You feed them to an LLM and ask for "commitments and dates".

**Task:** Write what you would check before trusting that extraction (whose commitments, stated vs implied, what is missing), and the one question you would send Victor to confirm the single most load-bearing date.

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. For PM work, the most accurate mental model of an LLM is…**

   a. A certified project assistant
   b. A pattern-matcher producing plausible drafts — including plausible-but-wrong risk items, dates, and dependencies
   c. A search engine over PMBOK
   d. A scheduling engine

**2. The PM tasks where AI helps most are characterised by…**

   a. High-politics, high-nuance judgement work
   b. Document manufacturing at volume
   c. The final decisions themselves, always
   d. Difficult vendor negotiations and talks

**3. What must NEVER be delegated to AI on a program?**

   a. Drafting all the meeting agendas
   b. Accountability and the go/no-go
   c. Formatting all of the data tables
   d. Summarising all the meeting minutes

**4. M365 Copilot differs from consumer ChatGPT for project data because…**

   a. It simply has a much better underlying model than that one
   b. It runs inside your tenant's boundary with controls
   c. It is free
   d. It writes better English

**5. A transformation project differs from a transition project in that transformation…**

   a. Is always bigger
   b. Changes how things work (e.g. SCCM→Intune modern management); transition moves where they run (e.g. DC→colo/cloud)
   c. Only involves software
   d. Has no risks

**6. PM data sensitivity differs from ticket data because it leaks…**

   a. Nothing at all really — it is just the plans involved
   b. Commercial secrets — pricing, terms, org changes
   c. Only the purely technical implementation details of it
   d. Public information

**7. The knowledge-cutoff problem bites PMs when asking about…**

   a. General scheduling techniques and their methods
   b. Current roadmaps and recent licensing changes
   c. The various standard risk categories people use
   d. RACI theory

**8. The Week 1 capability map exists because…**

   a. Courses generally need some kind of artifacts
   b. You can't measure gains without a baseline
   c. The PMO strictly requires you to have one of these
   d. It fills the toolkit

**9. AI's structural blind spot on YOUR program is…**

   a. The basic mathematics and arithmetic involved
   b. Local reality — the landmines only you know
   c. Terminology
   d. Templates

**10. The best FIRST artifact to trust AI with is…**

   a. The full steering committee decision itself
   b. A meeting summary or status draft
   c. The final cutover go/no-go decision
   d. The vendor contract itself, in full

**11. "AI will replace project managers" fails as a claim because…**

   a. AI can't use MS Project
   b. The core of the role is accountability, influence, and judgment under ambiguity — AI removes the document grind AROUND that core
   c. PMs are unionised
   d. Projects are unpredictable

**12. Grading each task "transformative / helpful / marginal / unsafe" builds…**

   a. A procurement business case to put to the PMO board
   b. Calibration: your tested map of where to lean in
   c. A slide for the PMO
   d. Nothing lasting

### Module 2 — Prompt Engineering Mastery for Project Managers

**Guiding question:** How do I brief an AI as precisely as I'd brief a workstream lead?

**Outcome:** Master structured prompting for project work: role + program context + constraints + artifact format, chain-of-thought for analysis, few-shot for consistent artifacts, critique prompts for quality — plus the commercial-confidentiality sanitisation discipline PM data demands.

**Delivery lens:** PM artifacts leak different secrets than tickets do: vendor pricing, contract terms, org politics, named underperformers. Your prompts carry program context to get program-quality answers — but the sanitisation pass (names, numbers, commercials) runs first, every time.

**Apply-at-work mission — Build your first 25 templates:** Convert your recurring PM asks into 25 tested prompt templates across planning, risk, comms, change, and reporting — each with role/context/constraints/format and a field note from real use. Run everything through the PM Sanitisation Checklist first.

**Reflection:** Which template surprised me most with its quality — and which PM task turned out to need more of MY thinking than I'd been giving it?

#### Resources

- [Anthropic — Prompt engineering overview](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview) — The core reference: roles, context, examples, chain-of-thought — read it with artifacts in mind.
  - Build a prompt with role, program context, constraints and an artifact format
  - Read the core techniques with real PM artifacts in mind
  - Diagnose why a generic prompt produced a generic charter, and fix it
- [OpenAI — Prompt engineering guide](https://platform.openai.com/docs/guides/prompt-engineering) — Second perspective on the same craft; transfers to Copilot and every other tool.
  - Transfer a second vendor's strategies to Copilot and other tools
  - Keep the prompting techniques that survive across tools
  - Compare two frameworks on a real planning artifact
- [Microsoft — Copilot prompt gallery](https://adoption.microsoft.com/en-us/copilot/prompt-gallery/) — Enterprise prompt examples — mine it for PM-adjacent patterns to adapt into your library.
  - Mine the enterprise gallery for PM-adjacent patterns to adapt
  - Turn a gallery prompt into a program-specific one rather than starting blank
  - Recognise the anatomy of a well-formed enterprise prompt
- [UK ICO — anonymisation guidance](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/) — The formal grounding for the sanitisation habit — extended in this course to commercial confidentiality.
  - Extend the sanitisation habit from personal data to commercial confidentiality
  - Strip client and vendor identifiers before AI sees a document
  - Justify the discipline to a sponsor worried about a leak

#### In-world ticket queue

> You've survived week one. Now the volume hits: three meetings a day, two workstreams' worth of artifacts, and Diane's standing request for "crisp" everything. Time to build the prompt library that carries the load.

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0012 | task | this week's milestone | Turn your recurring asks into 25 tested templates |
| HZN-0013 | P2 | Victor Alonso | Nexon's 14-page minutes → decisions & actions with owners |
| HZN-0014 | P2 | Fatima Khan | Fatima wants "a store manager version" of the program overview |

#### Project — 25 PM Prompt Templates (Capstone Milestone 1)

Build your first 25 reusable prompt templates across the five PM domains: planning (charter, WBS, dependencies), risk (identification sweeps, scoring, responses), communication (status, escalation, exec brief), change (CR, impact assessment), and reporting (weekly, dashboard narrative, minutes→actions). Each: role + program context slots + constraints + artifact format, tested on a sanitised real sample with a field note. Include the critique pattern ("review this draft as a sceptical steering member") in at least 3 templates. 🎯 Starts Capstone Milestone 1.

**Deliverable:** playbook/w02-prompt-templates.md — 25 tested templates by domain (also saved to My Prompt Library), plus your completed PM Sanitisation Checklist run-through.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Structure discipline | 25% | All 25 carry role, context slots, constraints, and output format — artifact-shaped, not chat-shaped. |
| Domain coverage | 25% | All five PM domains covered in proportion to YOUR capability map — heaviest where your hours are. |
| Tested with field notes | 30% | Every template ran on a sanitised real sample; notes capture what needed human correction. |
| Sanitisation evidence | 20% | The checklist visibly applied: no client names, commercials, or personnel specifics in any tested sample. |

#### Scenario drills

**Drill 1.** Sam's "quick informal sync" (HZN-0003) turns out to contain three commitments he's already made to store managers about ring 1 timing.

**Task:** Draft the follow-up note with AI: confirm what was actually discussed, convert informal commitments into "subject to plan confirmation" language without embarrassing Sam, and log the items for the plan review.

**Drill 2.** Fatima asks for "a store manager version" of the program overview (HZN-0014). Your current overview is 6 pages of workstream jargon.

**Task:** Produce it with AI: one page, tills-and-trading language, what changes when, what never changes, who to call. Then have AI critique its own draft as a sceptical store manager with 4 minutes.

**Drill 3.** You realise your prompt templates all assume you have complete context. Half your Program Horizon asks start with fragmentary inherited information.

**Task:** Build the "incomplete-info" template: it must instruct the AI to answer with what's known, list what's missing, and never fill gaps silently. Test it on the stale inherited plan.

#### Prompt clinic — The 14-page vendor minutes (HZN-0013)

- **Weak:** "Summarise these minutes."
- **Average:** "Summarise these vendor kickoff minutes and list the action items."
- **Good:** "From these (sanitised) vendor kickoff minutes, extract: decisions made, actions with owner + due date, open questions, and anything that sounds like a scope assumption we haven't agreed to. Table format."
- **Excellent:** "Act as a program PM reviewing vendor kickoff minutes for risk. From the (sanitised) minutes: (1) decisions table (decision | who committed | binding on whom); (2) actions with owner + date, flagging any assigned to 'TBD'; (3) open questions ranked by schedule risk; (4) SCOPE WATCH: quote any sentence that assumes scope, responsibility, or timeline we have not formally agreed — these become CR candidates or clarification emails. If a name/date is ambiguous in the source, mark ⚠ rather than guessing."

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. The four load-bearing ingredients of a PM prompt are…**

   a. Greeting, the question, thanks, a signature
   b. Role, context, constraints, format
   c. Model choice, temperature, length, tone
   d. Urgency, seniority, deadline, and threat

**2. Program context transforms AI output because…**

   a. It flatters the model
   b. It collapses generic-project answers onto YOUR situation: a DC exit with a lease deadline plans differently than an open-ended migration
   c. Longer prompts always win
   d. It prevents hallucination

**3. Few-shot prompting matters most for PM work when…**

   a. Brainstorming a broad range of possible project risks upfront
   b. Producing artifacts that must match your house style
   c. Asking questions
   d. Summarising

**4. The critique pattern ("review this as a sceptical steering member") is powerful because…**

   a. It's entertaining
   b. It weaponises AI against your own drafts
   c. It entirely replaces the need for any human review
   d. Steering members like it

**5. Chain-of-thought helps PM analysis tasks (e.g. dependency reasoning) because…**

   a. It sounds rigorous
   b. Visible step-by-step reasoning lets you spot the wrong assumption mid-chain instead of distrusting the conclusion blindly
   c. It shortens answers
   d. It saves cost

**6. PM sanitisation extends beyond names to…**

   a. Nothing else at all beyond just that
   b. Commercials and personnel details
   c. Just the various key dates involved
   d. Technical details only

**7. Prompting for TECHNICAL accuracy vs BUSINESS communication differs in that…**

   a. Only wording changes
   b. Technical asks need environment specifics and verification hooks; business asks need audience, message, and length constraints
   c. Business prompts are longer
   d. There is no difference

**8. "Already decided / already tried" context in PM prompts prevents…**

   a. Overly long, rambling answers from it
   b. Re-litigating settled decisions
   c. Outright hallucination of the facts
   d. Excessive politeness all throughout

**9. A prompt TEMPLATE's field note ("worked, but invented two stakeholders") exists because…**

   a. Notes are simply a long-standing tradition
   b. It captures the template's failure modes
   c. It earns you some extra XP points somehow
   d. Templates expire

**10. Your charter template produces great transformation charters but weak transition ones. Likely cause…**

   a. Transitions inherently resist being chartered
   b. The template encodes transformation bias
   c. The model simply dislikes datacenters somehow
   d. Charters are generic

**11. The meta-prompt "what information would you need to make this plan realistic?" is useful because…**

   a. It simply stalls the whole thing for time
   b. It surfaces context you forgot
   c. Models genuinely enjoy being asked questions
   d. It quietly resets the whole chat

**12. 25 templates is the right Week 2 target because…**

   a. It's a round number
   b. It forces coverage of your real recurring work
   c. More templates is always simply better than fewer
   d. The rubric says so

---

## Phase 2 — Plan & De-Risk (weeks 3–4)

### Module 3 — AI-Powered Planning & Initiation

**Guiding question:** How do I produce initiation artifacts in days that used to take weeks — without inheriting AI's blind spots?

**Outcome:** Build charters, WBS, schedules, RACI matrices, and dependency maps with AI acceleration — for both transformation work (endpoint modernisation) and transition work (DC exit) — while catching the dependencies AI can't know.

**Delivery lens:** An AI can draft a WBS for "Windows 11 migration, 8,000 endpoints" in 40 seconds — a generic one. Your program's reality (the stores that can't take downtime in December, the app nobody owns, the lease that expires in March) is what turns a draft into a plan. AI supplies structure and completeness; you supply the landmines.

**Apply-at-work mission — Charter and decompose a real initiative:** Take a real (or realistic) initiative from your world and produce with AI assist: a one-page charter, a two-level WBS, and a dependency list flagging the critical path candidates. Then mark every item AI missed that only local knowledge could supply — that margin is your value, made visible.

**Reflection:** What did the AI's draft plan get structurally right, and which of my program's real landmines did it have no way to see? How does that change how I'll use it at initiation?

#### Resources

- [PMI — PMBOK Guide (standards hub)](https://www.pmi.org/standards/pmbok) — The reference frame for charters, WBS, and baselines — AI drafts against structures like these.
  - Draft charters, WBS and baselines against a recognised standard structure
  - Hold an AI-drafted plan to PMBOK-shaped expectations
  - Spot where an AI draft skipped a structural element the standard requires
- [Microsoft — Azure Cloud Adoption Framework: Migrate](https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/migrate/) — The canonical structure for datacenter transition planning: assess, migrate waves, optimise — mine it for WBS shape.
  - Borrow the assess / migrate-waves / optimise shape for a transition WBS
  - Structure datacenter-exit planning against the canonical framework
  - Catch the migration step an AI draft glossed over
- [Microsoft — Windows 11 deployment planning](https://learn.microsoft.com/en-us/windows/deployment/) — The EUC transformation side: rings, app compat, servicing — the other workstream's planning skeleton.
  - Borrow the rings / app-compat / servicing shape for an EUC-transformation WBS
  - Structure Windows 11 deployment planning against the reference
  - Hold the EUC workstream's plan to a documented skeleton
- [Atlassian — work breakdown structure guide](https://www.atlassian.com/work-management/project-management/work-breakdown-structure) — A clean practical WBS refresher to hold AI's drafts against.
  - Hold an AI-drafted WBS against a clean practical reference
  - Tell a genuine work breakdown from a flat task list
  - Fix the level of decomposition an AI draft got wrong

#### In-world ticket queue

> Steering wants a proper re-baseline: charter refresh and a credible plan for both workstreams. Sam says "two weeks per ring, easy". Priya quietly hands you a list titled "things the old plan forgot".

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0021 | P1 | Diane Moreau (steering) | Charter refresh + two-level WBS, both workstreams |
| HZN-0022 | P1 | Priya Nair | Priya's list: 31 "forgotten" items incl. 12 unowned apps in Manchester |
| HZN-0023 | P2 | plan review | Cross-workstream dependency: ring 3 stores need the new network first |

#### Project — Charter + WBS + Dependencies (Capstone Milestone 2)

Take a real or realistic initiative — ideally one EUC transformation (e.g. Win11/Intune modernisation) OR one datacenter transition (e.g. DC exit to colo/cloud). Produce with AI assist: (1) a one-page charter (objectives, scope boundary, success criteria, governance); (2) a two-level WBS; (3) a dependency list with critical-path candidates flagged; (4) a draft RACI for the top 10 activities. Then the key step: annotate everything AI missed that only local knowledge supplies — frozen change windows, unowned apps, lease dates, store trading calendars. 🎯 Completes Capstone Milestone 2.

**Deliverable:** playbook/w03-charter-wbs.md — all four artifacts plus the "what AI couldn't know" annotation layer.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Charter quality | 25% | One page a sponsor would sign: measurable objectives, explicit scope boundary, named governance — not a vision statement. |
| WBS completeness | 25% | Two clean levels covering the real work including the unglamorous parts (comms, training, decommissioning, hypercare). |
| Dependency realism | 25% | Dependencies include the cross-workstream and external ones (network before migration, vendor lead times, business calendars). |
| Local-knowledge layer | 25% | The annotations prove the point: a visible margin of landmines no AI could have listed. |

#### Scenario drills

**Drill 1.** Priya's list of 31 forgotten items (HZN-0022) includes 12 applications with no identified owner, all hosted in Manchester.

**Task:** Design the discovery mini-plan with AI: how to find owners (or confirm orphan status) for 12 apps in 3 weeks — methods, owners of the discovery itself, and the decision rule for true orphans before wave planning.

**Drill 2.** Sam estimates "two weeks per ring, easy". The inherited plan says three. Ring 1's actuals aren't in yet.

**Task:** Use AI to build the estimate interrogation: what would have to be true for 2 weeks/ring, what evidence exists for each condition, and what ring 1 must be measured for so ring 2's estimate is data, not vibes.

**Drill 3.** The cross-workstream dependency (HZN-0023): ring 3 stores need the new network, which is wave 2's deliverable, which is owned by Nexon, whose plan you've never seen aligned to Sam's.

**Task:** Feed both (sanitised) plan extracts to AI and ask for collision candidates: date conflicts, resource conflicts, and undefined handoffs. Then draft the three-way alignment agenda.

#### Prompt clinic — The WBS request (HZN-0021)

- **Weak:** "Create a WBS for a datacenter migration."
- **Average:** "Create a work breakdown structure for migrating 214 servers from an on-prem datacenter to colocation and Azure."
- **Good:** "Create a two-level WBS for a datacenter EXIT: 214 servers, VMware on-prem → colo + Azure in waves, lease expires in 11 months (immovable). Include discovery, dependency mapping, network prerequisites, wave migrations, decommissioning, and contract exit."
- **Excellent:** "Act as an infrastructure transition planner. Create a two-level WBS for a DC exit: 214 servers (VMware → colo + Azure), waves, lease expiry in 11 months IMMOVABLE — plan backwards from that date. Context: 12 applications have no identified owner; store systems freeze mid-Nov to early Jan (retail trading); a parallel EUC workstream needs the new store network before its ring 3. Requirements: include the endings people forget (decommissioning, data destruction evidence, contract exit, straggler servers); flag which work packages are critical-path candidates given the deadline; mark cross-workstream dependencies explicitly. Then list 10 questions whose answers would most change this WBS."

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. AI's genuine value at initiation is…**

   a. Knowing your program
   b. Structural completeness at speed: a full-skeleton charter/WBS draft in minutes, for you to correct and localise
   c. Political judgment
   d. Approving the charter

**2. A WBS for a datacenter transition that AI typically UNDER-represents is…**

   a. The bulk of the core server migration tasks themselves
   b. The endings — decommissioning, exits, stragglers
   c. Kickoff meetings
   d. Tooling selection

**3. For an 8,000-endpoint Win11 transformation, the planning structure AI should be steered toward is…**

   a. A single big-bang deployment
   b. Ring/wave-based
   c. Ordered alphabetically by user
   d. Simply whatever is fastest to do

**4. The critical path in a DC-exit with a fixed lease expiry is special because…**

   a. It isn't
   b. The end date is immovable: float is consumed backwards from the deadline, so late discovery of dependencies is existential
   c. Lease dates always move
   d. Critical paths only matter in construction

**5. A dependency AI can't know without being told is…**

   a. Doing the network work before the migration
   b. Your own business-calendar constraints
   c. Doing all the testing before the go-live
   d. Backup before change

**6. A RACI draft from AI most often errs by…**

   a. Missing the required matrix format entirely
   b. Over-assigning the Accountable role
   c. Including far too few of the actual roles
   d. Wrong colours

**7. Schedule estimates from AI should be treated as…**

   a. Firm commitments to be made to the whole board
   b. Starting points to calibrate to your reality
   c. Padding that should simply be cut straight out
   d. Vendor quotes

**8. The "what would you need to know?" prompt at initiation doubles as…**

   a. A deliberate stalling tactic on your part
   b. Your stakeholder discovery agenda
   c. A full project risk register document
   d. A standard charter document section

**9. Cross-workstream dependencies (EUC ring 2 needs the new store network from the DC workstream) are dangerous because…**

   a. They are really quite rare in practice
   b. They're owned by nobody by default
   c. They are always perfectly clearly visible
   d. The tools always reliably flag them all

**10. Governance in the charter (boards, cadence, decision rights) matters at AI-draft time because…**

   a. It fills the page
   b. AI writes plausible governance theatre
   c. Sponsors nearly always skip over that whole section
   d. Governance never changes

**11. Success criteria in an AI-drafted charter typically need…**

   a. The addition of a good many more descriptive adjectives
   b. Quantification with baselines, not aspirations
   c. Removal
   d. Legal review

**12. The two-workstream frame (transformation + transition) teaches planning because…**

   a. Two is more than one
   b. They fail differently: transformations die on adoption/compat; transitions die on dependencies/deadlines — plans must reflect which game they're in
   c. Auditors require both
   d. It doubles the artifacts

### Module 4 — Risk Assessment & Management with AI

**Guiding question:** How do I find the risks I'd normally discover the hard way — in week one instead of month four?

**Outcome:** Run AI-assisted risk identification against both workstream types, qualify and rank with probability × impact, build response strategies worth the paper, keep RAID logs alive instead of decorative, and rehearse "what-if" scenarios before reality runs them.

**Delivery lens:** Every infra PM has a scar list: the app-compat surprise, the network change that broke the stores, the vendor who slipped. AI has read ten thousand projects' worth of scar lists. It won't know YOUR program's risks — but it will name the categories you haven't checked yet, at 9am on day one.

**Apply-at-work mission — Build the register that would have saved you:** Pick a project you know well (running or past). Build its risk register with AI assist: identification sweep by category, probability × impact scoring, response strategy per top-10 risk. Then run one "what-if" scenario (e.g. "wave 1 cutover fails at 02:00") and document the response plan. Compare against what actually happened, if past.

**Reflection:** Which risk did the AI sweep surface that I genuinely hadn't listed — and which of my known risks could no AI have found? What does that teach me about where the sweep fits in my practice?

#### Resources

- [PMI — risk management standards & practice guides](https://www.pmi.org/standards/risk-management) — The formal frame: identification, qualitative/quantitative analysis, responses — AI accelerates inside this structure.
  - Run AI-assisted risk identification inside the formal identify / analyse / respond frame
  - Qualify and rank risks rather than just listing them
  - Keep AI acceleration inside a structure a PMO recognises
- [Microsoft — Cloud Adoption Framework: risk sections](https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/) — Migration-specific risk thinking from the transition side; skim assess/migrate for risk categories.
  - Mine migration-specific risk categories from the transition framework
  - Apply cloud-adoption risk thinking to a datacenter exit
  - Name the transition risks an AI generic list would miss
- [NIST — risk management resources](https://www.nist.gov/risk-management) — The security/compliance risk vocabulary your infra programs intersect with.
  - Bring security and compliance risk vocabulary your infra program intersects with
  - Frame a risk conversation in terms audit and security accept
  - Cite recognised risk functions in a register
- [Anthropic — chain-of-thought prompting](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/chain-of-thought) — Re-read for risk analysis: make the model reason about mechanism and leading indicators, not just list.
  - Prompt the model to reason about mechanism and leading indicators, not just list risks
  - Use chain-of-thought to surface a risk's early-warning signal
  - Turn a flat risk list into ranked, reasoned entries

#### In-world ticket queue

> Risk week, involuntarily: Nexon's lead migration engineer resigned, ring 1 pilot shows 7% app-compat failures (estimate was 3%), and someone remembers the Manchester DC also hosts the store CCTV archive. The register you're about to build isn't theoretical.

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0031 | P1 | Victor Alonso | Nexon lead engineer resignation — wave 1 impact unknown |
| HZN-0032 | P1 | Sam Kowalski | Ring 1 compat failures at 7% vs 3% planned — extrapolate! |
| HZN-0033 | P2 | Priya Nair | Surprise: CCTV archive dependency on Manchester DC discovered |

#### Project — Risk Register + Response Playbook (Capstone Milestone 3)

For a project you know well (running or past): (1) run AI identification sweeps by category — technical, vendor, resource, business, compliance — for BOTH workstream types if possible; (2) score probability × impact and rank; (3) write response strategies (avoid/mitigate/transfer/accept) for the top 10, each with owner, trigger, and leading indicator; (4) run one full what-if scenario ("wave 1 cutover fails at 02:00" or "app-compat failure rate double the estimate") into a response plan. If a past project: compare the register against what actually happened. 🎯 Completes Capstone Milestone 3.

**Deliverable:** playbook/w04-risk-register.md — the register (top 10 fully developed), the scenario plan, and the reality comparison if applicable.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Sweep breadth | 25% | Category-by-category sweeps done; at least two risks surfaced that weren't on your instinctive list. |
| Scoring honesty | 20% | Probability × impact scored with your program knowledge, not accepted from AI defaults; ranking defensible. |
| Response quality | 30% | Top-10 responses are actionable: named owner, concrete trigger, leading indicator — not "monitor closely". |
| Scenario depth | 25% | The what-if plan covers decision points, comms, rollback, and the first 24 hours — rehearsal-grade, not gesture-grade. |

#### Scenario drills

**Drill 1.** The Nexon lead-engineer resignation (HZN-0031): Victor says "no impact, we have bench strength". Your program's wave 1 says otherwise if he's wrong.

**Task:** Build the trust-but-verify response with AI: what evidence would substantiate "no impact" (named replacement, handover plan, wave 1 task coverage), the risk register entry with trigger and indicator, and the polite-but-specific email requesting the evidence.

**Drill 2.** The CCTV archive discovery (HZN-0033): a compliance-relevant system nobody scoped, hosted on hardware that must die in 11 months.

**Task:** Run the full new-risk workflow: AI-assisted impact sweep (compliance, storage, migration options), register entry with scoring, response options, and the one-paragraph steering notification that neither panics nor buries.

**Drill 3.** Diane asks: "What are the three risks that would actually kill this program? Not the register top-10 — the killers."

**Task:** Use AI as a sparring partner: feed the (sanitised) register and program context, ask for existential-risk candidates with kill mechanisms, then argue against each until you own a three-item answer.

#### Prompt clinic — The compat extrapolation (HZN-0032)

- **Weak:** "Is 7% app failure bad?"
- **Average:** "Our Windows 11 pilot shows 7% app compatibility failures vs 3% planned. What should we do?"
- **Good:** "Ring 1 pilot (800 endpoints, office personas) shows 7% app-compat failure vs 3% planned. Remaining rings cover 7,200 endpoints including store and warehouse personas. Analyse: what does 7% in ring 1 imply for later rings, what factors could make it better or worse, and what are our response options?"
- **Excellent:** "Act as a deployment risk analyst. Data: ring 1 = 800 office-persona endpoints, 7% app-compat failure (planned 3%); ring 2–4 = 7,200 endpoints across office/store/warehouse personas; store persona apps are OLDER on average; each failure costs ~2 engineer-hours to remediate. Reason step by step: (1) is ring 1 likely representative — argue both directions; (2) scenario table for 5%/7%/10% failure rates across remaining rings: remediation hours, schedule impact at current capacity, cost; (3) response options (expand app-testing before ring 2, add remediation capacity, re-sequence rings by persona risk) with trade-offs; (4) what data from ring 1 would sharpen this analysis most. State every assumption explicitly."

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. AI's core contribution to risk identification is…**

   a. Actually knowing your specific risks
   b. Category completeness at speed
   c. High scoring accuracy on all of them
   d. Owning responses

**2. Probability × impact scores from AI should be…**

   a. Accepted, since it's read more projects than you
   b. Replaced with your own judgment
   c. Roughly doubled just for safety's sake
   d. Ignored entirely

**3. A "leading indicator" on a risk means…**

   a. The name of the risk's currently assigned owner
   b. The early signal before the risk fires
   c. The overall calculated impact score of it
   d. A single steering committee summary slide

**4. The classic transformation-side risk (Win11/Intune) is…**

   a. A hard, immovable lease expiry deadline
   b. App compatibility and user adoption
   c. Insufficient physical rack space available
   d. Limited bandwidth between the two DCs

**5. The classic transition-side risk (DC exit) is…**

   a. Insufficient end-user training
   b. The undiscovered dependency
   c. Cosmetic icon changes on the desktop
   d. Team morale

**6. "Monitor closely" as a risk response is…**

   a. A valid mitigation
   b. A non-response — no action, owner, or trigger named
   c. Perfectly sufficient for all of the lower risks at least
   d. Required by PMBOK

**7. What-if scenario planning with AI is valuable because…**

   a. Scenarios are fun
   b. You rehearse the response cheaply, before reality does
   c. Steering committees will always demand it of you anyway now
   d. It replaces the register

**8. A RAID log stays alive (vs decorative) when…**

   a. It is formally reviewed once every year annually
   b. Updates are cheap and scheduled each week
   c. It's in SharePoint
   d. It has more columns

**9. Feeding meeting minutes to AI and asking "what new risks appear here?" catches…**

   a. Nothing — minutes are noise
   b. The risks that announce themselves in passing ("vendor mentioned their lead engineer is leaving") but never reach the register
   c. Only formatting issues
   d. Attendance problems

**10. Transfer as a risk response in infra programs typically means…**

   a. Emailing it off to someone else
   b. Contractual vendor placement
   c. Simply deleting the risk entirely
   d. Escalating it straight to steering

**11. Comparing a fresh AI-built register against a PAST project's reality teaches…**

   a. Nothing — the past is past
   b. Calibration: which risk classes the sweep catches, which only experience catches — evidence for how to run the next one
   c. That registers are futile
   d. Vendor selection

**12. Quantitative support (e.g. "expected schedule impact if compat failure rate is 8% vs 3%") from AI requires…**

   a. Blind trust
   b. Your inputs and assumptions made explicit — AI arithmetic on stated assumptions is useful; invented numbers are not
   c. A statistician
   d. Monte Carlo software

---

## Phase 3 — Communicate & Control (weeks 5–6)

### Module 5 — Stakeholder Communication & Influence

**Guiding question:** How do I make every audience feel the program was explained just for them — without writing five versions by hand?

**Outcome:** Map stakeholders by power and interest, generate audience-tuned communications (exec brief, business update, technical detail, vendor coordination), draft escalations that get decisions instead of resentment, and run the meeting-to-actions workflow.

**Delivery lens:** A DC exit reads as "risk reduction and cost avoidance" to the CFO, "will my tills stop working?" to store operations, and "wave 3 network prerequisites" to the infra team. Same truth, three renderings — exactly what AI does instantly once YOU decide what each audience needs. Influence stays human; production gets help.

**Apply-at-work mission — Ship the comms pack:** Build your stakeholder map (power × interest) for a real program, then produce the pack with AI assist: one exec briefing, one business-facing update, one escalation draft (real or realistic), and one decision-request memo. Run one real meeting through the minutes→actions→owners workflow (with recording/consent rules observed).

**Reflection:** Where did AI's audience-tuning genuinely land, and where did only my knowledge of the person (not the role) make the difference? What does that boundary tell me?

#### Resources

- [MindTools — stakeholder analysis (power/interest grid)](https://www.mindtools.com/aol0rms/stakeholder-analysis) — The classic mapping frame your engagement plan builds on.
  - Map stakeholders on a power/interest grid and plan engagement from it
  - Decide comms cadence and depth per quadrant
  - Turn the map into an actual engagement plan, not a diagram
- [Prosci — ADKAR change management model](https://www.prosci.com/methodology/adkar) — The adoption side of transformation comms: awareness → desire → knowledge → ability → reinforcement.
  - Frame transformation comms with ADKAR — awareness through reinforcement
  - Diagnose which ADKAR stage a resistant stakeholder is stuck at
  - Tune an AI-drafted comms plan to the adoption stage
- [HBR — communication (collection)](https://hbr.org/topic/communication) — Pick one piece on executive communication; read it as a standard for your AI-drafted briefs.
  - Hold your AI-drafted exec briefs to a real executive-communication standard
  - Lead a brief with the decision, not the background
  - Cut a brief to what a busy sponsor will actually read
- [Atlassian — incident & stakeholder communication practices](https://www.atlassian.com/incident-management/incident-communication) — Status cadence and stakeholder comms discipline — transferable directly to cutover comms.
  - Transfer status-cadence discipline directly to cutover communications
  - Run stakeholder comms during a planned change like a controlled incident
  - Set the comms rhythm before the cutover weekend, not during it

#### In-world ticket queue

> The rumour mill beat your comms plan: a store manager heard "the IT change" will hit during Christmas trading and called Fatima. Fatima called Diane. Diane called you. Communication week begins under fire.

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0041 | P1 | Fatima Khan via Diane | Fatima escalation: "Christmas freeze — confirm in writing, TODAY" |
| HZN-0042 | P1 | Diane Moreau | Stakeholder map + comms pack — steering wants comms "professionalised" |
| HZN-0043 | P2 | steering feedback | Exec brief: CFO asking why the program needs the contingency it has |

#### Project — Stakeholder Plan + Comms Pack (Capstone Milestone 4)

For a real program: (1) build the stakeholder map — power × interest, with per-stakeholder concerns (the CFO's, store operations', the infra team's, the vendor's); (2) produce the comms pack with AI assist: one exec briefing (1 page), one business-facing update (plain language), one escalation draft that requests a specific decision, one decision-request memo with options; (3) run one real meeting through the minutes → decisions → actions-with-owners workflow, observing recording/consent rules. Every artifact audience-tuned and sanitised. 🎯 Completes Capstone Milestone 4.

**Deliverable:** playbook/w05-comms-pack.md — the map, all four communications, and the meeting workflow output.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Map insight | 25% | Stakeholders placed with named concerns (not just titles); the map changes what you'd send whom — visibly. |
| Audience tuning | 30% | The four artifacts genuinely differ in content, register, and length; the CFO version answers CFO questions. |
| Escalation quality | 25% | The escalation states impact, options, and the specific decision requested by when — decidable, not just alarming. |
| Meeting workflow | 20% | A real meeting produced verified minutes, decisions, and owned actions — with consent handled properly. |

#### Scenario drills

**Drill 1.** The CFO's contingency question (HZN-0043) is really a trust question: "does this PM know what they're doing with my money?"

**Task:** Draft the exec brief with AI: contingency as a function of quantified risk exposure (use your Week 4 register), what consumed contingency so far bought, and the release conditions. One page, CFO dialect.

**Drill 2.** Priya's private risk list — she'll discuss it verbally but won't write it down ("last program, the risk log got weaponised").

**Task:** Design your approach (AI as rehearsal partner): the conversation that gets her risks into the register safely — framing, anonymisation options, what you'll change about register culture. Rehearse her likely objections.

**Drill 3.** Your weekly status goes to Diane, Fatima, the CFO, and two workstream leads — five audiences, one report, everyone reads a different paragraph first.

**Task:** Restructure with AI: one master status with audience-tagged sections, generated from one verified input set. Ship this week's real(istic) report in the new structure.

#### Prompt clinic — The Fatima escalation (HZN-0041)

- **Weak:** "Write an email saying the stores won't be affected at Christmas."
- **Average:** "Write a reassuring email to our Head of Store Operations confirming no IT changes during the Christmas period."
- **Good:** "Write a confirmation to our Head of Store Operations (plain language, no jargon): store-affecting changes freeze from Nov 15 to Jan 5; what IS still happening in that window (non-store work); who to contact if anything looks wrong. Tone: confident, respectful of her concern, 150 words max."
- **Excellent:** "Context: a rumour reached store operations that our program will disrupt Christmas trading; the Head of Store Ops escalated to the program director; the truth is store-affecting changes freeze Nov 15–Jan 5, but datacenter waves (not store-visible) continue. Draft TWO artifacts: (1) a written confirmation to her — plain language, acknowledges the escalation was reasonable, states the freeze dates and what continues (with why it cannot affect tills), names the escalation contact, 150 words; (2) a 5-line brief she can forward to 300 store managers verbatim. Do not over-promise: if a P1 incident forced an emergency change in the window, our change process still allows it — say so honestly in one clause."

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. The power × interest grid drives communication by…**

   a. Ranking importance for the org chart
   b. Determining engagement strategy: manage closely / keep satisfied / keep informed / monitor — different effort per quadrant
   c. Choosing meeting rooms
   d. Setting salaries

**2. AI's strongest contribution to stakeholder comms is…**

   a. Knowing your stakeholders
   b. Instant audience re-rendering: one verified truth → exec brief, business update, technical detail — once YOU define each audience's needs
   c. Replacing relationships
   d. Attending meetings

**3. An escalation that WORKS contains…**

   a. The maximum possible amount of alarm and urgency it can raise
   b. Impact, options with trade-offs, one decision by a date
   c. Blame allocation
   d. Every detail

**4. For store operations facing a DC transition, the communication that lands is…**

   a. The full migration architecture diagram
   b. Impact in their world, not yours
   c. The complete full project plan itself
   d. The vendor contract summary

**5. ADKAR matters to a transformation PM because…**

   a. It is really just another compliance framework of some kind
   b. Adoption is the transformation's real finish line
   c. It replaces the WBS
   d. Auditors expect it

**6. The minutes → actions workflow needs a consent check because…**

   a. The meeting minutes are copyrighted
   b. Recording has policy rules
   c. The AI itself strictly requires it
   d. The actions are all private

**7. AI-drafted difficult-conversation prep ("the vendor is slipping; rehearse my talking points") is best used to…**

   a. Script the confrontation verbatim
   b. Rehearse: anticipate responses, stress-test your framing, find the non-inflammatory phrasing — then talk like a human
   c. Send instead of meeting
   d. Avoid the conversation

**8. Status updates fail most often because…**

   a. They are simply far too short to be useful
   b. They report activity instead of meaning
   c. They use entirely the wrong font throughout
   d. Too infrequent

**9. The vendor-coordination communication challenge in multi-vendor programs is…**

   a. Persistent language barriers between them
   b. Each vendor optimises only its own scope
   c. Time zones
   d. Invoice formats

**10. A decision-request memo beats a status report for getting decisions because…**

   a. It is simply a good deal shorter overall
   b. It's built for the decision
   c. Executives generally prefer memos
   d. It CCs a good many more people

**11. Anonymising stakeholder analysis before AI processing matters because…**

   a. Stakeholder names are often really quite long
   b. It encodes sensitive political reality
   c. The AI actively dislikes handling real names
   d. It doesn't

**12. The sign your comms pack is working is…**

   a. Compliments on the nice formatting
   b. Changed stakeholder behaviour
   c. Noticeably higher email open rates
   d. Consistently much longer meetings

### Module 6 — Scope, Change Control & Governance

**Guiding question:** How do I catch scope creep in week two instead of at the steering committee post-mortem?

**Outcome:** Baseline scope explicitly, write change requests and impact assessments that boards can decide on, use AI to detect drift between baseline and reality (status language, change logs, meeting actions), and prep CCB decisions with balanced options.

**Delivery lens:** "While you're migrating the stores anyway, could you also…" — creep never announces itself; it accumulates in meeting minutes and hallway agreements. AI can compare what you're reporting against what you baselined and flag the drift — if you baselined sharply enough to compare against. Baseline discipline is what makes creep detectable.

**Apply-at-work mission — Baseline, then hunt your own creep:** Write (or sharpen) the scope baseline for a real project: in-scope, out-of-scope, assumptions. Then feed your last month of status reports/minutes to AI against that baseline and ask for drift candidates. Write one full change request + impact assessment for the most real one found.

**Reflection:** What did the drift analysis surface that I'd absorbed without noticing — and what does my out-of-scope list's quality say about my baselining habits?

#### Resources

- [PMI — scope management (PMBOK area)](https://www.pmi.org/standards/pmbok) — The formal frame: baseline, verify, control — your CR discipline hangs off this.
  - Baseline scope explicitly against the formal verify / control frame
  - Hang your change-request discipline off a recognised scope-management structure
  - Tell a real scope baseline from a vague statement of intent
- [ITIL 4 — change enablement (Axelos)](https://www.axelos.com/certifications/itil-service-management) — The change-control vocabulary your CAB/CCB and infra teams share.
  - Speak the change-enablement vocabulary your CAB and infra teams share
  - Write a change request that a change board can actually decide on
  - Align PM change control with ITIL change enablement
- [Atlassian — scope creep guide](https://www.atlassian.com/work-management/project-management/scope-creep) — A practical read on how creep actually arrives — matches this week's detection exercise.
  - Recognise how scope creep actually arrives, matching this week's detection exercise
  - Catch creep in the wording of an "informal" request
  - Name the creep before it is baked into a plan

#### In-world ticket queue

> Creep season: stores want new peripherals "while you're visiting anyway", Sam agreed to "a small pilot" of self-service kiosks in ring 2 minutes, and Nexon's CR-007 quietly widens their scope. Your baseline is about to earn its keep — if it's sharp enough.

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0051 | P1 | Victor Alonso | CR-007 from Nexon: "clarification" that adds 40 servers to managed scope |
| HZN-0052 | P2 | minutes review | Kiosk pilot appeared in ring 2 minutes — nobody approved it |
| HZN-0053 | P2 | Fatima Khan (sympathetically) | Store peripherals request — 300 stores × "just a docking station" |

#### Project — Scope Baseline + Creep Hunt + Change Request

Three artifacts: (1) write or sharpen a real project's scope baseline — in-scope, out-of-scope (the underrated half), assumptions, acceptance criteria; (2) run the creep hunt: feed your last month of status reports, minutes, and action logs to AI against that baseline, asking for drift candidates with evidence quotes; (3) take the most real drift found and write the full change request + impact assessment (schedule, cost, risk, resources) as it would go to your CCB, with balanced options.

**Deliverable:** playbook/w06-scope-control.md — the baseline, the drift analysis with your verdicts on each candidate, and the CCB-ready change request.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Baseline sharpness | 30% | Out-of-scope list is explicit and program-specific ("store hardware refresh excluded"); assumptions testable; a drift comparison is actually possible against it. |
| Creep hunt rigour | 25% | Real artifacts fed (sanitised); each AI drift candidate verified by you with a verdict: real creep / approved change / false positive. |
| CR quality | 30% | The change request is decidable: quantified impacts, options including "reject", a recommendation, and rollback of the change itself considered. |
| Governance fit | 15% | The CR fits your actual CCB's format and authority levels — not a textbook's. |

#### Scenario drills

**Drill 1.** The kiosk pilot (HZN-0052) appeared in ring 2 minutes as "agreed". Sam says it's "basically free — the devices are already there".

**Task:** Run the drift protocol: baseline comparison (is it in scope?), impact quantification with AI ("basically free" audited: support, imaging, training, security review), and the CR-or-kill conversation plan with Sam.

**Drill 2.** The 300 × docking station request (HZN-0053): individually tiny, collectively a six-figure scope change wearing a friendly face.

**Task:** Build the response with AI: quantified impact at 300-store scale, the options memo (fold into scope via CR / separate mini-project / decline with rationale), and the reply to Fatima that preserves the relationship whatever the answer.

**Drill 3.** Month-end: you run the creep hunt over four weeks of status reports and minutes against the baseline.

**Task:** Execute it for real (sanitised artifacts): AI flags drift candidates with quotes; you verdict each (creep / approved / false positive); the real ones get a CR or a kill decision this week.

#### Prompt clinic — CR-007, the quiet scope widener (HZN-0051)

- **Weak:** "Is this change request OK?"
- **Average:** "Review this vendor change request and tell me if there are any problems with it."
- **Good:** "Review this (sanitised) vendor CR against our scope baseline (attached). Identify: what exactly changes, quantified impact on schedule/cost/risk, and whether the 'clarification' framing is accurate or this is a material scope change."
- **Excellent:** "Act as a contract-literate program PM. Inputs: our scope baseline (sanitised) and vendor CR-007, framed as a 'clarification'. Analyse: (1) exact delta — quote baseline text vs CR text side by side; (2) is 'clarification' accurate, or does this materially transfer scope/risk/cost — argue from the quotes; (3) quantified impact if accepted (schedule, cost exposure, our residual obligations); (4) our options: accept / negotiate (what specifically) / reject (on what grounds), each with relationship and delivery consequences; (5) draft the neutral, non-accusatory response requesting the CCB route. Assume good faith in tone, rigour in analysis."

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. Scope creep is dangerous primarily because…**

   a. It is more or less always fundamentally malicious in intent
   b. It accumulates invisibly, small addition by small addition
   c. It arrives in writing
   d. It only affects budgets

**2. The underrated half of a scope baseline is…**

   a. The main in-scope work item list
   b. The explicit out-of-scope list
   c. All of the various required signatures
   d. The formatting

**3. AI detects scope drift by…**

   a. Intuition
   b. Comparing artifacts: baseline text vs status narratives, minutes, and action logs — flagging work being reported that maps to nothing baselined
   c. Reading minds
   d. Monitoring budgets

**4. A drift candidate flagged by AI becomes "creep" only after…**

   a. The AI confirms it at least twice over
   b. Your own classifying verdict on it
   c. The very same thing appears twice over
   d. The vendor agrees

**5. A CCB-decidable change request contains…**

   a. A long and thoroughly persuasive written essay arguing for it
   b. Quantified impact, options with rejection, a deadline
   c. Maximum urgency
   d. Technical appendices only

**6. The impact assessment AI drafts will be weakest on…**

   a. Formatting
   b. Second-order effects in YOUR program: what this change does to the store calendar, the vendor contract, the parallel workstream
   c. Arithmetic
   d. Grammar

**7. "Gold-plating" differs from scope creep in that it's…**

   a. Always externally imposed on you
   b. Self-inflicted, internal polish
   c. Nearly always beneficial to have
   d. A specific piece of hardware terminology

**8. In a fixed-deadline transition (lease expiry), an approved scope ADDITION must be paired with…**

   a. Plenty of sheer raw enthusiasm
   b. An explicit trade elsewhere
   c. Consistently longer workdays
   d. A whole new baseline meeting

**9. Change requests for the EUC transformation vs the DC transition differ typically in…**

   a. Nothing
   b. Blast pattern: EUC changes ripple across users/adoption/training; DC changes ripple across dependencies/waves/downtime windows
   c. Font requirements
   d. Approval authority only

**10. The assumptions section of a baseline earns its keep when…**

   a. Only when the auditors actually get around to reading it
   b. An assumption breaks and becomes a documented change
   c. Auditors read it
   d. It's long

**11. Running the creep hunt MONTHLY (not once) matters because…**

   a. Good habits tend to impress the PMO
   b. Creep compounds over time
   c. The AI itself needs the practice
   d. The baselines eventually expire

**12. CCB preparation with AI ("summarise these 6 pending CRs with impacts and a recommended agenda order") is legitimate because…**

   a. It isn't — boards really must suffer through it
   b. Preparation is production work
   c. AI votes too
   d. It reduces all the meetings to zero entirely

---

## Phase 4 — Deliver & Systematise (weeks 7–8)

### Module 7 — Execution, Reporting & Cutover

**Guiding question:** How do I run the reporting machine in an hour a week — and walk into cutover weekend with evidence instead of hope?

**Outcome:** Build the status-report generation workflow (inputs → AI draft → your judgment layer), generate dashboard narratives, coordinate multi-team/multi-vendor dependencies, keep decision logs current, and produce cutover plans + readiness assessments with go/no-go criteria that mean something.

**Delivery lens:** Cutover weekend is where transition PMs earn their reputation: the runbook, the rollback triggers, the comms cadence, the readiness evidence. AI drafts the machinery brilliantly — checklists, timeline, comms sequences — and has no idea whether YOUR wave is actually ready. The go/no-go call is the most human moment in the whole discipline.

**Apply-at-work mission — Automate the report, rehearse the cutover:** Build your status workflow: define inputs (plan deltas, RAID changes, milestones), draft with AI, add your judgment layer, ship a real report with it. Then build a cutover readiness assessment for a real/realistic migration wave: criteria, evidence required, go/no-go thresholds, rollback triggers — and dry-run the decision.

**Reflection:** With the reporting grind reduced, what did I do with the recovered hours — and did the readiness framework change how confident (or honestly unconfident) I felt about the go call?

#### Resources

- [Microsoft — Cloud Adoption Framework: migration waves & cutover](https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/migrate/) — Wave planning and cutover structure from the transition canon — the skeleton your readiness pack fleshes out.
  - Structure wave planning and cutover from the transition canon
  - Flesh a readiness pack out of the documented cutover skeleton
  - Sequence waves so dependencies resolve before the cut
- [Atlassian — incident communication (for cutover comms)](https://www.atlassian.com/incident-management/incident-communication) — Cutover weekend IS a planned incident: same comms cadence discipline applies.
  - Treat the cutover weekend as a planned incident with incident comms cadence
  - Set the status rhythm for a cutover before it starts
  - Apply major-incident comms discipline to a planned change
- [Google — technical writing: reports](https://developers.google.com/tech-writing) — The editing standard your generated status reports and dashboard narratives get held to.
  - Hold generated status reports and dashboard narratives to a real editing standard
  - Cut the padding out of an AI-drafted status report
  - Make a report lead with status and risk, not activity
- [Microsoft — Copilot in Teams/Loop for meetings & status](https://learn.microsoft.com/en-us/copilot/microsoft-365/) — Where your reporting workflow can live natively if your org runs M365 Copilot.
  - Run your reporting workflow natively where your org already runs M365 Copilot
  - Decide when Copilot-in-Teams beats a separate tool for status
  - Wire meeting and status capture into the reporting loop

#### In-world ticket queue

> Wave 1 cutover is in 12 days: 38 servers, one Saturday night, Manchester → colo. Diane wants a readiness view she can defend at the board. Marco-grade honesty required: two criteria are currently amber.

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0061 | P1 | Diane Moreau | Wave 1 cutover readiness assessment — go/no-go in 12 days |
| HZN-0062 | P1 | readiness review | Rollback rehearsal not yet done — Priya needs a window |
| HZN-0063 | task | this week's milestone | Status machine: weekly report eating 4 hours — automate it |

#### Project — Status Workflow + Cutover Readiness (Capstone Milestone 5)

Two builds. (1) The status machine: define your inputs (plan deltas, RAID changes, milestone status, decisions pending), build the AI drafting step, add your judgment layer (meaning, emphasis, honesty), and ship at least one real report through it — plus a dashboard narrative variant for executives. (2) The cutover readiness pack for a real/realistic migration wave: readiness criteria with evidence required per criterion, go/no-go thresholds, rollback triggers and procedure references, comms sequence (T-7 through T+1), and hypercare arrangements. Dry-run the go/no-go decision against honest current evidence. 🎯 Completes Capstone Milestone 5.

**Deliverable:** playbook/w07-delivery-pack.md — the status workflow (with one shipped report), the readiness assessment, and the dry-run decision record.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Status workflow reality | 25% | Inputs defined, draft generated, judgment layer visible (what you changed and why), and a real report actually shipped with it. |
| Narrative quality | 20% | The executive narrative says what changed, what it threatens, what's needed — meaning, not activity. |
| Readiness pack depth | 35% | Criteria are evidence-backed (test results, sign-offs, rollback rehearsed), thresholds numeric where possible, rollback triggers unambiguous. |
| Honest dry-run | 20% | The go/no-go dry-run against real evidence produced a defensible call — including "no-go" if that's what the evidence said. |

#### Scenario drills

**Drill 1.** The rollback rehearsal (HZN-0062) hasn't happened and the window options are: this weekend (vendor B-team) or next weekend (T-5 days before cutover).

**Task:** Build the decision analysis with AI: risk of each option, what a B-team rehearsal actually validates vs doesn't, and the recommendation with reasoning for Diane. Then make the call yourself.

**Drill 2.** It's T-2h on cutover night. Smoke test 4 of 6 passed; test 5 is delayed because the vendor's network engineer is stuck in traffic; the window maths still works — barely.

**Task:** Against YOUR readiness pack's pre-agreed rules: is this go or no-go? Write the decision record either way — what the rules said, what you decided, what you communicated at T-2h.

**Drill 3.** Your first automated status draft says everything is green. Your own judgment says wave 1 is amber (rehearsal debt) and ring 2 is amber (compat backlog).

**Task:** Fix the workflow, not just the report: why did the inputs produce green? Add the inputs/rules that force amber conditions to surface. Ship the corrected honest report.

#### Prompt clinic — The readiness view Diane can defend (HZN-0061)

- **Weak:** "Are we ready for the cutover?"
- **Average:** "Create a go/no-go checklist for our wave 1 datacenter cutover."
- **Good:** "Build a cutover readiness assessment for wave 1 (38 servers, Saturday night, Manchester → colo): readiness criteria grouped by category (technical, people, comms, rollback), evidence required per criterion, and go/no-go thresholds."
- **Excellent:** "Act as a cutover manager. Build the wave 1 readiness pack (38 servers, Saturday 22:00–06:00, Manchester → colo): (1) criteria table: criterion | evidence REQUIRED (artifact, not assertion) | current status slot | threshold (numeric where possible); include the unfashionable ones — rollback rehearsed with timing, backup restore TESTED, vendor staffing confirmed by name, store-freeze conflict check, hypercare rota signed; (2) go/no-go decision rules: what combination of ambers forces no-go, who holds the call, when it's made (T-48h and T-2h); (3) rollback triggers during the window: objective conditions + decision deadline ('if smoke tests incomplete by 04:30 → roll back'); (4) comms sequence T-7 → T+1 per audience. Format so a director can read the STATUS column and defend the decision at board level."

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. The status report workflow's division of labour is…**

   a. The AI writes it, and you just send it
   b. AI drafts; you add meaning
   c. You write it all, the AI just formats
   d. The PMO writes the whole thing

**2. A dashboard narrative exists because…**

   a. The dashboards are simply ugly
   b. Numbers don't self-explain
   c. Executives can't read the charts
   d. Templates require text

**3. Cutover weekend most resembles…**

   a. Just a regular, normal working sprint
   b. A carefully planned incident
   c. A nice relaxing holiday period
   d. A demo

**4. A go/no-go criterion is real (not theatre) when…**

   a. Everyone agrees verbally
   b. It names required EVIDENCE: "UAT sign-off received", "rollback rehearsed successfully on <date>", "backup verified"
   c. It's on a slide
   d. The vendor asserts it

**5. Rollback triggers must be defined BEFORE cutover because…**

   a. The auditors always ask about it
   b. Pre-agreed triggers beat panic
   c. Actual rollbacks are really quite rare
   d. The vendors all strictly require it

**6. The go/no-go call is "the most human moment" because…**

   a. It's emotional
   b. It weighs evidence, risk appetite, and accountability
   c. The AI happens to be entirely offline over the weekends
   d. It requires seniority

**7. Multi-vendor coordination during execution most needs…**

   a. A good many more regular meetings
   b. Explicit handoffs and ownership
   c. A much bigger RAID log to maintain
   d. Shared pizza

**8. A decision log earns its existence when…**

   a. It is finally complete only at the project closeout
   b. "Why did we do X?" months later has a real answer
   c. It's long
   d. Auditors visit

**9. Hypercare differs from BAU support because…**

   a. It's cheaper
   b. Heightened everything: staffing, monitoring, escalation speed, and rollback readiness for a defined post-cutover window
   c. It's optional
   d. Only vendors do it

**10. For the EUC transformation, the execution-phase signal to watch hardest is…**

   a. The overall server CPU utilisation levels throughout
   b. Ring health — the metrics predicting the next ring
   c. Badge swipes
   d. License counts

**11. For the DC transition, the execution-phase signal is…**

   a. Overall meeting attendance rates throughout
   b. Wave burn-down against the deadline
   c. General office printer status reports
   d. Team sentiment and general morale only

**12. Automating the status report should NOT automate…**

   a. Data collection
   b. The honesty: if the truth is amber, no workflow should smooth it green — the judgment layer exists to protect candour
   c. Formatting
   d. Distribution

### Module 8 — Your AI-Powered Project System (Capstone)

**Guiding question:** What does my personal AI operating system for projects look like — and what proves it works?

**Outcome:** Assemble the Personal AI Project Manager Playbook: prompt library, artifact templates, workflows, governance rules — integrated with your real tool stack (M365/Copilot, Project, Jira, Teams, ServiceNow), measured against your Week 1 map, with a 90-day adoption plan.

**Delivery lens:** The difference between "I use AI sometimes" and "I run an AI-augmented practice" is a system: which artifact, which prompt, which review step, which tool — decided once, then executed every project. Your capstone applies it end-to-end to one program; your playbook makes it repeatable on every one after.

**Apply-at-work mission — Ship the Playbook, run the system:** Complete the Playbook in the tracker, apply the full system end-to-end to your capstone program (charter → risks → comms → change → reporting → cutover), compare your Week 1 task-time map against now, and present the result to a real audience — your PMO, your manager, or a recorded walkthrough.

**Reflection:** Final entry: reread Week 1. Which task did I most underestimate AI on, which did I most overestimate — and what does "being a great infrastructure PM" mean now that document manufacturing is cheap?

#### Resources

- [Microsoft — M365 Copilot adoption resources](https://adoption.microsoft.com/en-us/copilot/) — Integrating your practice into the tenant most enterprises actually run.
  - Integrate your practice into the tenant most enterprises actually run
  - Plan adoption of your workflow inside M365, not beside it
  - Set what "adopted" means for your personal system
- [PMI — AI in project management (thought leadership)](https://www.pmi.org/learning/thought-leadership/ai-impact) — Re-read from Week 1 with practitioner eyes: what do you now agree and disagree with?
  - Re-read Week 1's framing with practitioner eyes and revise your view
  - Name what you now agree and disagree with about AI in PM
  - Turn eight weeks of practice into a defended position
- [NIST — AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) — The governance vocabulary for the professional-responsibility section of your playbook.
  - Bring governance vocabulary to the professional-responsibility section of your playbook
  - Frame your AI use in terms an audit would accept
  - Cite the core risk functions in your own standards
- [The Batch — DeepLearning.AI newsletter](https://www.deeplearning.ai/the-batch/) — Your bounded staying-current channel after the course.
  - Set one curated weekly read as your staying-current channel
  - Filter the noise down to one reliable signal
  - Keep your playbook current after the course ends

#### In-world ticket queue

> Wave 1 went in (with one rollback trigger nearly pulled at 03:40 — your pre-agreed criteria held the line). Diane, at steering: "Whatever you've been doing differently — write it down before you forget it." That's the capstone.

| Ref | Priority | From | Request |
| --- | --- | --- | --- |
| HZN-0071 | capstone | this week's milestone | Assemble the PM Playbook + before/after numbers |
| HZN-0072 | capstone | Diane Moreau | Present the system at PMO forum — Diane volunteered you |
| HZN-0073 | task | you, to yourself | Ring 2 + wave 2 ahead: 90-day plan for running both on the new system |

#### Project — Personal AI Project Manager Playbook + Capstone (Milestone 6)

Assemble the complete Playbook: prompt library (40+ by now), artifact templates (charter, register, comms pack, CR, readiness pack), your workflows (status machine, RAID refresh, creep hunt, minutes→actions), tool integrations for your real stack, and your governance rules (sanitisation, review gates, what AI never touches). Then the capstone: apply the system END-TO-END to one real or realistic infrastructure program — both workstream types if possible — producing the full artifact chain. Compare your Week 1 task-time map against now. Write the 90-day adoption plan. Present to your PMO, manager, or as a recorded walkthrough. 🎯 Completes the programme.

**Deliverable:** playbook/ — the assembled Playbook, the end-to-end capstone artifact chain, before/after numbers, 90-day plan, and the presentation.

**Assessment rubric**

| Criterion | Weight | What good looks like |
| --- | ---: | --- |
| Playbook completeness | 25% | All components present and organised: a colleague PM could adopt any workflow from it in minutes. |
| End-to-end capstone | 30% | The full chain (charter → risks → comms → change → reporting → readiness) produced with the system, quality-checked, coherent as one program. |
| Measured delta | 25% | Week 1 map re-scored honestly; the delta stated per task with method — hours/week recovered, quality observations included. |
| Adoption plan + presentation | 20% | 90-day plan with named next steps and KPIs; presentation tells the change story in outcomes, not features. |

#### Scenario drills

**Drill 1.** The PMO forum presentation (HZN-0072): 15 minutes, eight PMs, at least two convinced "AI status reports" means "PM who stopped reading their own program".

**Task:** Build the presentation with AI: lead with delivery outcomes and the judgment-layer design (what the human still owns, visibly), include the wave-1 03:40 story, end with the starter kit you'd hand them.

**Drill 2.** Your before/after map shows status reporting 4h→1h, RAID upkeep now actually happening, charters days→hours — but escalation writing unchanged and stakeholder time UP 3 hours.

**Task:** Write the honest analysis: why judgment-bound tasks didn't compress, why stakeholder time INCREASING might be the best outcome on the whole map, and what that says about where recovered hours should go.

**Drill 3.** Ring 2 and wave 2 start next month. Sam wants your templates; Priya wants the readiness pack "but adapted properly, not copy-pasted".

**Task:** Design the handover: what transfers as-is, what needs their-workstream adaptation, and the 30-minute onboarding each of them gets. Put it in the 90-day plan with dates.

#### Knowledge check (12 questions)

**Self-test prompts. Answers and explanations are not published here — take the quiz at https://ragentic.netlify.app/#/courses/ai-it-pm to check yourself.**

**1. A personal AI "operating system" for projects means…**

   a. A custom GPT
   b. Decided-once, executed-always per stage
   c. Simply using AI whenever you remember to
   d. A dedicated automation platform of some kind

**2. Integrating with your real stack (M365/Copilot, Project, Jira, Teams) matters because…**

   a. The vendors tend to reward loyalty
   b. Practice survives where work lives
   c. It happens to be entirely contractual
   d. Standalone tools are simply banned here

**3. The professional-responsibility line for an AI-augmented PM is…**

   a. Disclose nothing
   b. You own every artifact and decision, whatever drafted it
   c. Add clear AI disclaimers absolutely everywhere you possibly can
   d. Blame the tool

**4. The before/after comparison uses the Week 1 map because…**

   a. A certain symmetry is quite pleasing
   b. Same tasks, honestly re-scored
   c. The PMO already filed it away safely
   d. It is simply what the rubric says

**5. An honest capstone delta sounds like…**

   a. "AI transformed everything"
   b. "Status: 4h→1h weekly. RAID upkeep: now 20 min weekly, was skipped. Charters: days→hours. Escalations: unchanged — judgment-bound."
   c. "10x productivity"
   d. "Results vary"

**6. The 90-day plan's primary enemy is…**

   a. Budget
   b. Regression: deadline pressure returns, old habits return, the playbook gathers dust — unless next steps are named and calendared
   c. Tool churn
   d. Colleague envy

**7. Sharing your playbook with the PMO is strategic because…**

   a. Hoarding is difficult
   b. Team-level adoption multiplies the value considerably
   c. It reduces your own personal workload right down to zero
   d. It's required

**8. Applying the system to BOTH workstream types in the capstone proves…**

   a. Stamina
   b. Generality: the system flexes across both workstream types
   c. Broad general-purpose tool compatibility right across the board
   d. Nothing extra

**9. Your staying-current loop should be…**

   a. Reading every single AI newsletter going
   b. Bounded and sustainable, not exhaustive
   c. Nothing at all really — done is simply done
   d. Annual training

**10. The KPIs to attach to your adoption plan are…**

   a. The total number of prompts you have written
   b. Ones your PMO already tracks and owns
   c. Overall raw token consumption levels throughout
   d. Model benchmarks

**11. Presenting the capstone, you lead with…**

   a. The full prompt library tour
   b. The delivery outcomes
   c. Detailed model comparisons
   d. A live demo of absolutely everything

**12. The identity shift this course argues for is…**

   a. PM → prompt engineer
   b. PM → PM whose document manufacturing is delegated and whose week concentrates on judgment, influence, and the go/no-go moments
   c. PM → developer
   d. No shift

## Toolkits

### PM Data Sanitisation Checklist

**Unlocks in module 2.**

The pre-flight check for project artifacts: commercial + personal confidentiality before any AI contact.

```markdown
# PM Sanitisation Checklist — before ANY project artifact reaches an AI tool

## Strip or tokenise (commercial)
- [ ] Client/company names → "the client" / "Company A"
- [ ] Vendor names + day rates + contract values → "Vendor A", "<rate>"
- [ ] Budget figures, contingency amounts (unless the analysis needs them — then approved tools only)
- [ ] Contract terms, penalty clauses, negotiation positions
- [ ] Unannounced org changes, restructures, strategy

## Strip or tokenise (personal/political)
- [ ] Names of individuals → role labels ("EUC lead", "vendor PM")
- [ ] Anything implying performance ("X is struggling", "Y always slips")
- [ ] Stakeholder analysis attached to real names — power/interest maps use ROLES
- [ ] HR-adjacent content: absences, disputes, staffing changes

## Keep (the working substance)
- Structures, dates, durations, dependencies, quantities (servers, endpoints, stores)
- Sanitised risk descriptions, generic constraints ("retail trading freeze mid-Nov–Jan")

## Combination check
- [ ] Could role + site + situation identify one person or one vendor anyway?
- [ ] Would I be comfortable if this exact prompt appeared in a procurement dispute?

**Rule: the plan's structure travels; the program's secrets don't.**
```

### PM Prompt Pack (50+)

**Unlocks in module 2.**

Ready-to-use prompts across planning, risk, comms, change, and delivery — the pack your 25 templates grow into.

```markdown
# PM Prompt Pack — 50+ ready-to-use prompts

Replace <placeholders>. Sanitise first (commercial + personal). You own every artifact shipped.

## Planning & initiation (9)
1. "Act as a senior infrastructure PM. Draft a one-page charter for <initiative>: measurable objectives, scope boundary (in AND out), success criteria with numbers, governance anchored to <real bodies>. Context: <sanitised>."
2. "Create a two-level WBS for <project>. Type: <transformation|transition>. Constraints: <deadline/calendar>. Include the endings (decommissioning, contract exit, hypercare). Flag critical-path candidates."
3. "For a DC exit with lease expiry <date>, plan BACKWARDS from the deadline: what must be true by when? Milestone spine with float analysis."
4. "For an endpoint transformation (<N> devices, personas: <list>), propose ring structure with compat gates and per-ring success criteria."
5. "From these two (sanitised) workstream plans, find collision candidates: date conflicts, shared-resource conflicts, undefined handoffs. <PASTE BOTH>"
6. "Draft a RACI for these 10 activities. Exactly one A per row. Roles: <list>. <ACTIVITIES>"
7. "What would you need to know to make this plan realistic? Rank the questions by how much the answer changes the plan. <PLAN>"
8. "Interrogate this estimate: what must be true for <estimate>? Which assumptions are testable now, and what would ring/wave 1 need to measure? <CONTEXT>"
9. "Convert this messy inherited plan into: current-state summary, obvious gaps, staleness flags, and the 10 questions to ask the previous team. Do not fill gaps silently — list them. <PASTE>"

## Risk (9)
10. "Risk identification sweep for <project type>, category by category: technical, vendor, resource, business, compliance, cross-workstream. For each risk: description, mechanism, early indicator."
11. "Score these risks probability × impact (1–5 each) with reasoning I can challenge; rank. I will re-score from program knowledge. <LIST>"
12. "For the top 10, draft responses (avoid/mitigate/transfer/accept): concrete action, owner role, trigger, leading indicator. No 'monitor closely'."
13. "What-if scenario: <e.g. wave 1 cutover fails at 02:00>. Build the response plan: decision points, comms, rollback, first 24 hours."
14. "From these (sanitised) minutes, extract risk candidates that were mentioned but never logged — quote the line each came from. <PASTE>"
15. "Steelman the case AGAINST my top risk being real. What am I over-weighting?"
16. "Given ring/wave 1 actuals (<data>) vs plan (<data>), extrapolate scenarios for remaining work. Argue both directions on representativeness. State assumptions."
17. "Which three risks on this register could actually KILL the program (not just hurt it)? Kill mechanism for each. <REGISTER>"
18. "Refresh this RAID from this week's status + minutes: propose adds, updates, closes — as a delta list for my review, never auto-applied. <RAID + SOURCES>"

## Stakeholders & communication (10)
19. "Build a power × interest map from these role descriptions (roles only, no names): quadrant, core concern, engagement strategy each. <ROLES>"
20. "One verified truth, three renderings: exec brief (1 page, decisions-focused), business update (plain language, impact-on-them), technical detail (dependencies, dates). <FACTS>"
21. "Draft an escalation that gets a decision: impact, options with trade-offs, specific ask by <date>, recommended option. Non-alarmist, decidable in one read. <SITUATION>"
22. "Decision-request memo: context (5 lines), options table (incl. do-nothing), trade-offs, recommendation, deadline, decider. <DETAILS>"
23. "From this (consented) transcript/minutes: decisions | actions with owner + date | open questions | scope-watch quotes. Mark ambiguity ⚠, never guess owners. <PASTE>"
24. "Rewrite for a store-manager audience: tills-and-trading language, what changes when, what never changes, who to call. 150 words. <PASTE>"
25. "Rehearse a difficult conversation: I need to tell <role> that <situation>. Anticipate their responses, stress-test my framing, propose non-inflammatory phrasing. Do NOT script it word-for-word."
26. "Critique this draft as a sceptical <CFO|steering member|store manager> with 4 minutes: what's unclear, what's missing, what would you challenge? <DRAFT>"
27. "Draft the freeze-confirmation: change freeze <dates>, what continues (and why it can't affect <their world>), escalation contact, honest emergency-change clause. Forwardable verbatim. <CONTEXT>"
28. "Summarise this thread for someone joining cold: positions, agreements, open items, recommended next step. <PASTE>"

## Scope & change (8)
29. "Sharpen this scope baseline: tighten in-scope wording, generate the explicit OUT-of-scope list from these known adjacent asks, make assumptions testable. <BASELINE + CONTEXT>"
30. "Drift hunt: compare this baseline against these status reports/minutes. Flag work being reported that maps to nothing baselined — with quotes. <BASELINE + ARTIFACTS>"
31. "Draft a CCB-ready change request: description, reason, quantified impact (schedule/cost/risk/resource), options incl. reject, recommendation, decision deadline. <DETAILS>"
32. "Audit this 'basically free' addition: support, training, security, imaging, lifecycle costs at <scale>. What does free actually cost? <PROPOSAL>"
33. "Baseline vs CR text, side-by-side quotes: what exactly changes? Is 'clarification' accurate or is this material scope transfer? <BOTH>"
34. "This change is requested in a fixed-deadline program. Draft the trade options: descope what, add what resource, or accept what risk. <CONTEXT>"
35. "Prepare the CCB agenda from these pending CRs: summary each, impacts, interdependencies between CRs, recommended order. <CRS>"
36. "Which of these approved changes have quietly expanded since approval? Compare approval text vs current descriptions. <BOTH>"

## Delivery, reporting & cutover (10)
37. "Status draft from these inputs (plan deltas, RAID changes, milestones, pending decisions): what changed, what it threatens, what's needed. Meaning over activity. I will edit for emphasis and honesty. <INPUTS>"
38. "Dashboard narrative: these RAG statuses + metrics → 5 sentences an exec acts on. Amber items get a 'to protect X, do Y by Z' clause. <DATA>"
39. "Single-source multi-audience status: master report with sections tagged [EXEC] [BUSINESS] [WORKSTREAM]. One truth, no version drift. <INPUTS>"
40. "Cutover readiness pack for <wave>: criteria table (criterion | evidence REQUIRED | status | threshold), amber-combination no-go rules, rollback triggers with decision deadlines, comms sequence T-7→T+1."
41. "Draft the cutover runbook skeleton: timeline with owners, checkpoints, comms cadence, rollback procedure reference, hypercare handoff. <WAVE DETAILS>"
42. "Go/no-go dry run: against these rules and this evidence, what does the framework say? Where is judgment still required beyond the rules? <PACK + EVIDENCE>"
43. "Decision log entry from these minutes: decision, date, options considered, decider, rationale. <MINUTES>"
44. "Multi-vendor handoff table from these plans: deliverable | from | to | date | acceptance criteria. Flag handoffs with no named receiver. <PLANS>"
45. "Hypercare plan for <wave/ring>: window, staffing, monitoring focus, escalation speed, exit criteria."
46. "Post-wave retrospective structure: what the metrics say, what surprised us, what changes for the next wave. <DATA>"

## System & governance (6)
47. "Review my prompt library for gaps against this task map: which recurring tasks have no template? <MAP + LIBRARY>"
48. "What sensitive content remains in this text I'm about to paste? Commercial and personal. <PASTE>"
49. "List every claim in this AI-drafted artifact I should verify before shipping, ranked by consequence. <ARTIFACT>"
50. "Design the review gate for this workflow: what the human checks every time vs samples vs trusts. <WORKFLOW>"
51. "My 90-day adoption plan draft: critique for regression risk — where will old habits reassert, and what countermeasure per point? <PLAN>"
52. "Turn this course's before/after data into a PMO-facing one-pager: what changed, what it saves, what the practice requires (gates, sanitisation, review). <DATA>"
```

### Project Charter Template (one page)

**Unlocks in module 3.**

The sponsor-signable structure every AI charter draft gets poured into.

```markdown
# CHARTER: <Initiative name>

**Type:** ☐ Transformation (changes how things work) ☐ Transition (moves where they run)
**Sponsor:** **PM:** **Date:** **Baseline version:**

## Why (2–3 sentences)
Business driver + what happens if we don't.

## Objectives (measurable)
1. <e.g. 8,000 endpoints on Win11/Intune by <date>, ≥95% first-pass success>
2. <e.g. Manchester DC vacated by <lease date>, zero unplanned P1s attributable to migration>

## Scope boundary
**In scope:** ______
**Explicitly OUT of scope:** ______ ← the drift-prevention list; be specific
**Assumptions (testable):** ______

## Success criteria (numbers + baselines)
| Criterion | Baseline | Target | Measured how |
|-----------|----------|--------|--------------|

## Governance (real bodies only)
Steering: <who, cadence> · Change authority: <CCB, thresholds> · Escalation path: ______

## Top 5 risks (headline only — register holds the rest)
## Milestone spine (5–8 dates)
## Budget envelope + contingency basis

*AI-drafted structure; every fact, date, and name verified by the PM. Sponsor signature = baseline.*
```

### Risk Register + Response Playbook

**Unlocks in module 4.**

The living-register structure: scoring, responses with triggers, and the weekly refresh ritual.

```markdown
# Risk Register — <program>

## Register
| ID | Risk (mechanism, not vibe) | Cat | P (1-5) | I (1-5) | Score | Response | Owner | Trigger | Leading indicator | Status |
|----|---------------------------|-----|---------|---------|-------|----------|-------|---------|-------------------|--------|
| R1 | | | | | | | | | | |

Categories: technical / vendor / resource / business / compliance / cross-workstream
Responses: avoid / mitigate / transfer / accept — with a concrete ACTION. "Monitor closely" is banned.

## Response playbook (top 10 risks)
### R<id>: <name>
- Mechanism: how it actually fires
- Leading indicator: what we'd see first (metric + threshold)
- Response action: what happens, by whom, funded how
- Trigger: the objective condition that activates the response
- What-if rehearsed: ☐ (link to scenario plan)

## Weekly refresh ritual (15 min, AI-assisted)
1. Feed week's status + minutes to the refresh prompt (#18) → delta list
2. Verdict each proposed add/update/close — YOUR call, never auto-applied
3. Re-check top-5 leading indicators against reality
4. Anything mentioned-but-unlogged from minutes sweep (#14)? Log or consciously discard.

## Scenario plans on file
| Scenario | Rehearsed | Response plan link |
|----------|-----------|--------------------|
```

### Stakeholder Engagement Plan

**Unlocks in module 5.**

Power × interest mapping with per-stakeholder concerns and comms strategy — roles, not names, when AI-processed.

```markdown
# Stakeholder Engagement Plan — <program>

## Map (ROLES only if this touches an AI tool)
| Role | Power | Interest | Quadrant | Core concern (their words) | Strategy | Owner | Cadence |
|------|-------|----------|----------|---------------------------|----------|-------|---------|
| Program sponsor | H | H | Manage closely | | | | |
| CFO | H | L→M | Keep satisfied | "Why this contingency?" | | | |
| Head of <business area> | H | H | Manage closely | "Will my <operations> break?" | | | |
| Workstream leads | M | H | Keep informed++ | | | | |
| Vendor PM | M | H | Manage actively | Optimises own SOW | | | |

## Per-audience communication design
| Audience | What they need | What to omit | Format | Frequency |
|----------|----------------|--------------|--------|-----------|

## Escalation routes
Business concern → ______ · Delivery slip → ______ · Vendor dispute → ______

## The comms pack (linked artifacts)
- [ ] Exec briefing (1 page) - [ ] Business-facing update - [ ] Escalation template
- [ ] Decision-request memo - [ ] Freeze/impact confirmations - [ ] Forwardable store-manager brief

## Meeting → action workflow
Consent/recording policy checked: ☐ Minutes prompt (#23) → decisions/actions/scope-watch → YOU verify owners → distribute within 24h.
```

### Change Request + Impact Assessment

**Unlocks in module 6.**

The CCB-decidable structure: quantified impacts, real options, and the fixed-deadline trade rule.

```markdown
# CHANGE REQUEST CR-<id>: <title>

**Raised by:** **Date:** **Baseline affected:** **Decision needed by:**

## Change description
What exactly changes, quoted against baseline text where possible.

## Reason
Why now, what happens if rejected.

## Impact assessment (quantified)
| Dimension | Impact | Basis |
|-----------|--------|-------|
| Schedule | | |
| Cost | | |
| Risk | new/changed register entries | |
| Resources | | |
| Other workstream | ripple effects | |

## Fixed-deadline trade (if program end date is immovable)
This change is absorbed by: ☐ descoping ______ ☐ adding resource ______ ☐ accepting risk ______

## Options
| Option | Consequence | Cost |
|--------|-------------|------|
| Approve as requested | | |
| Approve modified: ______ | | |
| Defer to <phase> | | |
| Reject | | |

**Recommendation + rationale:**

## Scope-creep hygiene
- [ ] This CR exists because governance caught it (good) not because it leaked in and got papered over
- [ ] Related "small asks" bundled and assessed at true scale (× stores/endpoints/servers)

**CCB decision:** ☐ approved ☐ modified ☐ deferred ☐ rejected · **Date:** · **Decider:**
```

### Cutover Readiness + Go/No-Go Pack

**Unlocks in module 7.**

Evidence-based readiness, pre-agreed no-go rules, and rollback triggers with decision deadlines.

```markdown
# Cutover Readiness Pack — <wave/ring>

**Window:** <date, start–end> · **Scope:** <N servers/endpoints> · **Decision points:** T-48h and T-2h
**Go/no-go holder:** <one named role>

## Readiness criteria (evidence, not assertion)
| # | Criterion | Evidence REQUIRED | Status | Threshold |
|---|-----------|-------------------|--------|-----------|
| 1 | Migration runbook complete | dry-run record | | 100% steps timed |
| 2 | Rollback rehearsed | rehearsal report + timing | | within window budget |
| 3 | Backups verified | TESTED restore evidence | | restore < ___ h |
| 4 | Vendor staffing confirmed | named roster | | all roles named |
| 5 | Business freeze conflict check | calendar sign-off | | zero conflicts |
| 6 | Comms sequence loaded | T-7→T+1 drafts approved | | all audiences |
| 7 | Hypercare rota signed | rota + escalation tree | | window covered |
| 8 | Smoke test suite ready | test list + owners | | 100% owned |

## No-go rules (pre-agreed, arithmetic not argument)
- Any RED at T-48h → no-go unless steering explicitly overrides in writing
- ≥2 AMBER in criteria 1–4 at T-2h → no-go
- <your rules> ______

## Rollback triggers (during the window)
| Trigger condition | Decision deadline | Action |
|-------------------|-------------------|--------|
| Smoke tests incomplete | by <time> | roll back per <runbook ref> |
| <critical system> unrecoverable | immediately | roll back + invoke <plan> |

## Comms sequence
T-7 all audiences · T-1 confirmation · T-0 start/major checkpoints · T+1 outcome + hypercare contacts

## Decision record
T-48h call: ______ by ______ · T-2h call: ______ by ______ · Rationale: ______
```

### Status Report Workflow Spec

**Unlocks in module 7.**

The reporting machine: inputs → AI draft → judgment layer → ship. One hour, not four.

```markdown
# Status Workflow — <program>

## Inputs (collected before drafting — garbage in, fiction out)
- [ ] Plan deltas since last report (milestones moved, % complete changes)
- [ ] RAID deltas (new/changed/closed, top indicators status)
- [ ] Decisions pending (with deadlines and owners)
- [ ] Workstream leads' 5-line updates (their words, on file)
- [ ] Metrics: <ring health / wave burn-down vs deadline / budget vs baseline>

## Draft step
Prompt #37 (or #39 for multi-audience). Sanitisation checklist on all inputs.

## Judgment layer (the actual job — minimum 15 minutes)
- [ ] MEANING: does the report say what changed, what it threatens, what's needed?
- [ ] EMPHASIS: is the most important thing FIRST, not buried at item 7?
- [ ] HONESTY: ambers stay amber. No workflow smooths truth. If my gut says amber and the draft says green, the INPUTS are wrong — fix them.
- [ ] ASKS: every problem paired with a specific request (decision, resource, time)

## Ship + archive
Distribute per stakeholder plan · archive input set with the report (audit trail).

## Candour audit (monthly)
Compare 4 weeks of reports against reality: did anything surprise stakeholders that the reports should have flagged? If yes → input or judgment-layer fix, in writing.
```

### 90-Day Adoption Plan (PM edition)

**Unlocks in module 8.**

The anti-regression plan: cadences, team scaling, and PMO-ready KPIs.

```markdown
# 90-Day Adoption Plan — <name>, <date>

## Evidence base (Week 8)
Task-time deltas: status __→__ · RAID upkeep __→__ · charters __→__ · minutes __→__
Unchanged (and why): ______ · Stakeholder time change: ______

## Days 1–30 — cement
- [ ] Status machine + RAID refresh + creep hunt running on live program(s)
- [ ] 2 templates/week added or field-note-updated
- [ ] Candour audit #1 completed

## Days 31–60 — extend
- [ ] Readiness pack adapted for next wave/ring by its OWNING lead (not copy-pasted)
- [ ] Onboard 2 colleagues: 30-min starter kit (3 prompts, sanitisation rule, judgment layer)
- [ ] Decision log + minutes workflow adopted by one governance body

## Days 61–90 — lead
- [ ] PMO presentation: outcomes, gates, what the practice requires
- [ ] Propose one PMO-level standard (sanitisation checklist or CR quality bar)
- [ ] Candour audit #2 + before/after refresh

## KPIs (from the PMO's existing scoreboard)
| KPI | Baseline | 90-day target |
|-----|----------|---------------|
| On-time milestone % | | |
| CR cycle time | | |
| Forecast accuracy (planned vs actual) | | |
| Stakeholder surprise incidents | | (count them honestly) |

## Guardrails I keep forever
1. I own every artifact and decision, whoever drafted it.
2. Sanitise (commercial + personal) before every paste.
3. The judgment layer never gets skipped — meaning, emphasis, honesty, asks.
4. Go/no-go calls are mine, informed but never delegated.

## Review dates (calendar NOW)
30: ______ · 60: ______ · 90: ______
```

### Apply-at-Work Mission Log

**Unlocks in module 1.**

The running record of every weekly mission: what you did, what happened, what you learned.

```markdown
# Apply-at-Work Mission Log

| Week | Mission | What I actually did | Outcome / time saved | What I'd do differently |
|------|---------|---------------------|----------------------|-------------------------|
| 1 | Tool up and map your load | | | |
| 2 | Build your first 25 templates | | | |
| 3 | Charter and decompose a real initiative | | | |
| 4 | Build the register that would have saved you | | | |
| 5 | Ship the comms pack | | | |
| 6 | Baseline, then hunt your own creep | | | |
| 7 | Automate the report, rehearse the cutover | | | |
| 8 | Ship the Playbook, run the system | | | |
```

## Capstone

### Focus area · EUC Transformation Playbook

Endpoint modernisation programs: ring planning, compat risk, adoption comms, persona-based delivery.

### Focus area · Datacenter Transition Playbook

DC exits and migrations: deadline-anchored planning, wave management, dependency discovery, cutover craft.

### Focus area · Cloud Migration Program Playbook

Hybrid/cloud programs: CAF-aligned planning, cost governance, multi-vendor coordination.

### Focus area · Network Transformation Playbook

Network refresh/SD-WAN programs: site-wave planning, business-calendar constraints, cutover-heavy delivery.

### Focus area · Multi-Workstream Program Playbook

The Program Horizon pattern: transformation + transition under one governance — cross-workstream craft.

### Your own · Your own program's reality

The best playbook mirrors YOUR delivery world — build around whatever you actually run.

### Milestones

- **Module 2 — PM prompt library started (25 templates)**
- **Module 3 — Charter + WBS for a real initiative**
- **Module 4 — Risk register + response playbook**
- **Module 5 — Stakeholder plan + comms pack**
- **Module 7 — Status workflow + cutover readiness pack**
- **Module 8 — PM Playbook + 90-day plan shipped**

### Portfolio checklist

- PM prompt library: 40+ tested templates with field notes
- Charter + WBS + dependency artifacts (with local-knowledge annotations)
- Risk register + response playbook + one rehearsed scenario
- Stakeholder map + four-piece comms pack
- Scope baseline + creep-hunt output + CCB-ready change request
- Status workflow spec with one shipped report
- Cutover readiness + go/no-go pack with dry-run record
- Before/after task-time map with method
- 90-day adoption plan, presented (PMO, manager, or recorded)