# FAST-02 Lab Guide: Generative AI for Everyday Work

Provider: 7formations  |  Course: FAST-02  |  Instructor: Wadood Chaudhary  |  Four labs (about 50 minutes each), one per session, plus the final applied task
Audience: trainees (steps, worksheets) and instructors (setup, answer notes, timing). Instructor-only notes are marked **Instructor**.

All companies, people, IDs, prices, and messages in this guide are fictional. Labs run in the employer's approved tool (Microsoft Copilot in Word, Outlook, Teams, and Copilot Chat; ChatGPT Business or Enterprise; Gemini for Google Workspace with Docs, Gmail, and NotebookLM; or Claude for Work), in the 7formations sandbox, or on paper. Nothing in the labs connects to an employer's systems or uses real customer, patient, or employee data.

---

## 0. Lab environment and safety rules (read before Lab 1)

**Tools**

- Use only the employer-approved tool, signed in with a work account. If none is approved, use the 7formations sandbox account (fictional data only, spend-capped, disabled at the end of the cohort) or the printed outputs in the course folder.
- The instructor demonstrates in the employer's tool only. Where steps differ, short tool notes are given.
- **Currency warning (October 2026).** Saved-helper features are changing: OpenAI is retiring custom GPTs on December 11, 2026 (February 11, 2027 for some workspaces) and migrating them to plugins, where instructions become a "skill"; Google is moving Gems on personal accounts to "skills" from November 2026; Microsoft's documentation now uses "Microsoft Copilot" for what was "Microsoft 365 Copilot," and builds agents with Agent Builder. The instructor confirms the current names and menus in the employer's tenant the week before Lab 4.

**Data**

- Fictional data only. Massachusetts G.L. c. 93H and 201 CMR 17.00 treat a resident's name combined with an SSN, a driver's license or state ID number, or a financial account or card number as personal information. Never paste such combinations.
- No PHI, no confidential business data (pricing, bids, unreleased financials, HR matters, source code, passwords).
- Helpers (agents, GPTs or plugins, Gems, projects) keep their instructions and reference files. Never put restricted data in a helper's knowledge.
- Shared-printer test: if you would not leave it on the office printer, do not paste it.

**Fictional organizations in this course**

- **Cape Harbor Inn**: 40-room inn, Chatham, MA (Lab 1).
- **North Shore Builders**: residential contractor, Beverly, MA (Lab 2).
- **Bayside Precision Parts**: CNC machine shop, Worcester County, MA (Lab 3 and Lab 4).

**Completion**

- Each worksheet is scored Not yet / Meets / Strong. A "Not yet" worksheet can be resubmitted once within the course window.

---

## Lab 1 (Session 1): Grounded customer notice from a policy sheet

**Scenario.** Cape Harbor Inn (fictional, Chatham) is resurfacing its pool in May and changing its early check-in fee. The owner wants a notice emailed to guests arriving in May. Last year an AI-written notice promised "complimentary breakfast" that the inn doesn't offer; the owner now insists every claim be traceable to the policy sheet.

**Goal.** Produce W1: a grounded notice, drafted in rounds, with a Gaps line and every claim traced.

**Data.** `lab1/cape-harbor-policy-sheet.txt` (fictional):

```text
CAPE HARBOR INN - POLICY SHEET (rev. April 2026)
Check-in 3:00 PM; check-out 11:00 AM.
Early check-in from 1:00 PM: $30 (was $25), subject to availability. Effective May 1.
Ferry shuttle to Chatham dock: 8:15 and 10:15 AM daily.
Pets: two dogs up to 40 lb, garden rooms only, $35/night. No cats.
Cancellation: full refund up to 7 days before arrival; within 7 days, first night charged.
Breakfast: Lantern Room 7:00-10:00 AM; gluten-free on request.
Pool: closed for resurfacing May 1-21. Reopens May 22 (weather permitting).
Front desk: 7:00 AM-11:00 PM. After hours: on-call manager via lobby phone.
```

**Steps**

1. **Thin prompt first (5 min).** Run: "Write a notice to May guests about the pool and check-in." Save the output. Highlight every claim not in the sheet.
2. **Round 1, outline (10 min).** Paste the sheet between `---` markers. Ask for a heading-only outline of a 120-word notice for guests arriving May 1 to 21. Confirm the job: pool closure, early check-in fee change, nothing else unless needed.
3. **Round 2, fill (10 min).** Ask the tool to fill each heading using only the sheet, writing "not in source" for missing facts and adding a "Gaps:" line.
4. **Round 3, polish (5 min).** Ask for a warmer tone and 120 words maximum, with "Do not add, remove, or change any facts."
5. **Trace (10 min).** Number each sentence. In W1, write the sheet line each claim comes from. Mark any untraceable claim and remove it.
6. **Compare (5 min).** Compare the thin and grounded outputs. Count invented claims in each.

**Tool notes.** Copilot in Word: use "Draft with Copilot" and reference the sheet file; or Copilot Chat with the text pasted. Gemini in Docs: "Help me write" with the sheet pasted, or the side panel referencing the file. ChatGPT or Claude: paste the sheet in the chat or add it to a project.

**Deliverable: W1 Grounded draft with Gaps line**

| Sentence # | Claim | Source line in sheet | Kept / removed |
| --- | --- | --- | --- |
| 1 | | | |

Gaps line (copy from output, then add who will answer each gap):

**Completion criteria.** Notice drafted in three rounds; every claim traced to a sheet line; Gaps line present with an owner for each gap; thin-prompt comparison recorded with a count of invented claims.

---

## Lab 2 (Session 2): Two-audience summary with fact-check log

**Scenario.** North Shore Builders (fictional, Beverly) issued a 30-page 2026 site safety manual. The site supervisor needs a decision-oriented brief; new laborers need day-one rules. Both must be accurate.

**Goal.** Produce W2: two summaries of the same manual, each with a five-fact check and a coverage check.

**Data.** `lab2/nsb-safety-manual-2026.pdf` (fictional, 30 pages, 9 sections). Key excerpts the instructor key relies on:

```text
§2 PPE: Hard hats required in all active work zones, including the material yard.
   Safety glasses with side shields for cutting, grinding, nailing.
§4 Fall protection: required at 6 feet or more above a lower level. Guardrails preferred;
   personal fall arrest systems inspected before each use.
§5 Ladders: extend 3 feet above the landing; 4:1 setup angle; no top-step standing.
§7 Incidents: report to the site supervisor within 1 hour; written report within 24 hours.
§8 Heat: water, rest, shade; 10-minute break each hour above 90°F heat index.
§9 Changes from 2025: heat rule added (§8); ladder inspections now weekly (was monthly).
```

**Steps**

1. **Chunk (10 min).** Summarize section by section (9 short summaries). Ask after each: "List anything in this section you did not include."
2. **Supervisor brief (10 min).** Combine into a decision-oriented brief (max 150 words): what changed from 2025, what to enforce this week, open questions.
3. **New-hire summary (10 min).** From the same section summaries, write 6 to 8 plain bullets (max 100 words), keeping every rule and number exact.
4. **Five-fact check (10 min).** For each summary, pick five facts (numbers, rules, deadlines). Find each in the manual; record match, wrong, or missing. Fix and recheck.
5. **Coverage check (5 min).** Compare the combined summaries against the nine section headings. Which sections did each summary omit, and was that right for the reader?

**Deliverable: W2 Two-audience summary and five-fact check**

| Summary | Fact checked | Manual section | Match / wrong / missing | Fix made |
| --- | --- | --- | --- | --- |
| Supervisor | 1 | | | |
| New hire | 1 | | | |

Coverage: sections omitted from each summary, with a one-line reason.

**Completion criteria.** Two summaries for named readers; five facts checked for each with results recorded; every wrong fact fixed and rechecked; coverage note complete.

---

## Lab 3 (Session 3): Three-quote comparison table with audit marks

**Scenario.** Bayside Precision Parts (fictional, Worcester County) buys acetal resin rod stock. Three suppliers sent quotes in different formats: a typed email, a PDF table, and a scanned form. The purchasing lead wants a comparison table she can trust.

**Goal.** Produce W3: an extraction table with zero invented values, audited cell by cell, with comparison math done in a spreadsheet.

**Data.** Three fictional quotes. Bank details have already been removed by the instructor (and that removal is part of the discussion).

| Quote | Format | Key content (abridged) |
| --- | --- | --- |
| A: Granite State Polymers | Email | $3.15/lb, min 200 lb, ships in 10 business days, FOB origin (freight extra), Net 30 |
| B: Pioneer Plastics Supply | PDF table | $42.00 per 10-lb bundle, min 50 bundles, Net 45, freight included over $1,500; no lead time stated |
| C: Bay Colony Materials | Scanned form | $3.05/lb, min 500 lb, 3 weeks, freight $0.12/lb, Net 30, "certs available on request" |

**Steps**

1. **Define columns (5 min).** Supplier | Unit price (USD/lb) | Minimum order (lb) | Lead time (business days) | Freight terms | Payment terms | Certifications stated | Source (quote and line).
2. **Extract (10 min).** One quote at a time, labeled A, B, C. Require blanks for missing values. No totals, no recommendation.
3. **Partner audit (15 min).** Swap tables. Mark every cell correct, wrong, or invented, with the source line next to it.
4. **Fix and rerun (5 min).** If any cell is invented, tighten the prompt ("leave blank if not stated") and rerun only that quote.
5. **Spreadsheet math (10 min).** Paste the audited table into the provided sheet. Use formulas to compute landed cost for 1,200 lb from each supplier (unit price × qty + freight). Note which supplier is cheapest and which has unknowns.

**Deliverable: W3 Extraction table audit**

| Cell | Extracted value | Source line | Audit mark | Fix |
| --- | --- | --- | --- | --- |
| B, lead time | | | | |

Plus the spreadsheet with formulas visible and a two-sentence note to the purchasing lead.

**Completion criteria.** Every cell audited; zero invented values in the final table; units consistent; landed cost computed by formula, not by the AI; unknowns stated in the note.

---

## Lab 4 (Session 4): Configure a helper and complete the before-and-after log

**Scenario.** You do the same writing task every week or month. Today you build a saved helper for it in the approved tool, have a partner run it cold, and measure whether it actually saves time once checking is counted.

**Goal.** Produce W4 Helper specification sheet and W5 Before-and-after time log.

**Data.** Your own recurring task (sanitized), or the Bayside quote-comparison task from Lab 3.

**Steps**

1. **Pick the task (5 min).** Recurring, low-to-medium risk, with a source you can supply safely.
2. **Write the spec (15 min).** Complete W4: job, reader, reference files, format, missing-fact rule, check step, refusal rule for restricted data.
3. **Build it (10 min).** In the employer's tool:
   - Microsoft Copilot: create an agent with Agent Builder; add instructions and approved SharePoint or OneDrive files as knowledge.
   - ChatGPT Business/Enterprise: create a plugin with a skill (or, until retirement, edit an existing GPT); add reference files.
   - Gemini: create a Gem (or skill, where migrated) with instructions and knowledge files; or a NotebookLM notebook for source-grounded Q&A.
   - Claude for Work: create a project with custom instructions and project knowledge.
   - Paper fallback: write the spec as a reusable prompt card.
4. **Partner cold test (10 min).** Your partner runs the helper on a new example without asking you anything and checks the output against the source. Record every question they wanted to ask; each one is a missing instruction.
5. **Log (10 min).** Complete W5 with honest times, including checking and fixing.

**Deliverable: W4 Helper specification sheet**

| Field | Your entry |
| --- | --- |
| Helper name | |
| Job (one sentence) and reader | |
| Reference files (approved, no restricted data) | |
| Output format and length | |
| Missing-fact rule | |
| Check step shown to the user | |
| Refusal rule (restricted data) | |
| Partner test result and changes made | |

**Deliverable: W5 Before-and-after time log**

| Task | Minutes by hand | Minutes with AI (run) | Minutes checking and fixing | Errors caught | Errors missed (found later) | Quality note | Keep / adjust / stop |
| --- | --- | --- | --- | --- | --- | --- | --- |

**Completion criteria.** Spec complete; helper built (or prompt card in paper mode); partner ran it cold and at least one instruction was improved; time log includes checking time and an honest decision.

---

## Appendix B: Rubric (Not yet / Meets / Strong)

| Criterion | Not yet | Meets | Strong |
| --- | --- | --- | --- |
| Grounding | Untraceable claims remain | Every claim traceable to the supplied source; gaps labeled | Gaps assigned to owners and resolved |
| Verification | Fewer than five facts checked | Five facts checked with result recorded | Errors traced to prompt fixes |
| Extraction | Any invented value | Zero invented values; math in spreadsheet | Unknowns surfaced as decisions |
| Reusability | Helper needs the creator to explain it | Helper instructions run correctly for a partner | Spec portable across tools |
| Value | Time log missing checking time | Time log complete with honest quality note | Keep/adjust/stop decision backed by numbers |