# Session 01 — Wake an Agent · Facilitator Guide **Deck:** [deck-session-01.html](deck-session-01.html) · **Handout:** [session-01-exercises.md](session-01-exercises.md) · **Length:** 150 min **Session goal (say it to yourself before walking in):** every student leaves with an agent that has run one real job under supervision and is scheduled to run again tonight, unattended. Nothing else matters. Cut anything before you cut that. **Success metric:** at 2:00 AM tonight, N schedules fire, where N = number of students. --- ## Run of show ### 0:00–0:10 · Welcome + the promise (slides 1–2) - Names, one line each: "the chore you'd most like to never do again." Write these on the board — several become JOB.md briefs in Exercise 3. - Slide 2. Say the promise plainly: *"You leave tonight with a running agent, not notes about one."* Some students won't believe you. Good. ### 0:10–0:20 · The reframe + live demo (slides 3–5) - Slide 4, chatbot vs agent. Don't lecture it — ask *"who here has used ChatGPT?"* then *"who has had it do something while the laptop was closed?"* The silence after the second question is the whole point. - **Slide 5, the live demo.** One instruction to your agent, out loud as you type: calendar conflict → proposed fix → message arrives on your phone. Hold your phone up when it buzzes. - Narrate the *choices* it made, not the magic: "nobody told it to check free slots — the goal implied it." - **If the demo fails:** open this morning's test-run log and narrate that instead. Say: *"this is the log of the same job from 7 this morning — and the fact that there IS a log is lesson one."* This recovery consistently lands as well as the live version. ### 0:20–0:35 · Anatomy: six parts (slides 6–9) - Slides 7–8. Keep pace brisk — the anatomy gets reinforced all course; it doesn't need to be mastered now. - Emphases that pay off later: **the model is the least interesting part** (surprises them, reframes vendor obsession); **the leash is a feature** (slide 9 — read rule zero out loud, have them read it back). ### 0:35–0:50 · EXERCISE 1 · First contact (slide 10) - Solo. Helpers roam. Real purpose: **flush out broken installs now**, not at minute 100. - The teachable beat is the permission prompt. When the first student gets one, freeze the room: *"read that dialog out loud."* Everyone should see one before moving on. - Students done early: have them ask the agent "what tools do you currently have?" — sets up Exercise 2. - ⚠ Any install unfixable in 5 minutes → pair the student, fix at break. ### 0:50–0:55 · Debrief (slide 11) - Goal → plan → tools → report. Have them find each phase in what they just watched. This shape is the course's recurring x-ray. ### 0:55–1:25 · EXERCISE 2 · Give it hands (slide 12) - The emotional peak of the night: **their agent messages their phone.** - Connector steps are in the handout with screenshots `[INSTRUCTOR: capture screenshots from your own setup — consent screens change]`. Budget says 30 min; the slowest third will need all of it. Helpers live in this block. - When phones start buzzing, let it be loud. Do not shush the room. That buzz is why they came. - ☆ SKIPPABLE fallback if running long: students who finish calendar skip Hermes and send the summary to a file instead; Hermes catch-up is 10 minutes of Session 02's arrival window. ### 1:25–1:30 · Break ### 1:30–1:45 · The job file (slides 13–15) - JOB.md's five headings. The line to land: *"writing a good brief for an agent is the same skill as writing a good brief for a person — most people were never taught either."* - Slide 15's bad/good contrast: ask the room what the agent would do with "organize my stuff." Let them generate the horror stories — self-generated caution sticks better than warnings. ### 1:45–2:10 · EXERCISE 3 · The standing instruction (slide 16) - Pairs critique briefs BEFORE running: the partner asks "what will it do if X?" until the brief answers. - Pull chores from the 0:00 board for anyone stuck choosing. - Enforce read-and-report only. If a student's brief says "send" or "delete," have them move it to a DRAFT step — narrate why: rung discipline starts tonight, unnamed. - Watch for the classic failure: briefs with no "done looks like." The agent rambles. Perfect teaching moment — tighten, re-run, contrast. ### 2:10–2:25 · EXERCISE 4 · Set the alarm (slides 17–19) - Schedule the nightly run. Have each student **read their schedule back from the tool** — not from memory. - The ritual matters: *"confirm the schedule, say your never-list to your partner, close the laptop."* Do it as a room, together. The closed laptop is the graduation gesture of night one. ### 2:25–2:30 · Homework + close (slides 20–22) - Homework is sleeping. Forbid watching the 2 AM run — name it: *"watching is a chatbot habit."* - Preview S02 with one number: *"next week your agent reads three hundred emails so you read three."* --- ## Common failure points | Symptom | Cause | Fix | |---|---|---| | "command not found: claude" | Install didn't land in PATH | Loaner laptop now, fix at break | | Connector consent loop | Wrong account picked mid-flow | Sign out of extra accounts, redo — 2 min | | Agent "did nothing" overnight (next week's complaint) | Laptop was closed AND asleep | `[INSTRUCTOR: platform-specific — confirm whether class runtime needs machine awake; if so, teach the caffeinate/power-settings step here]` | | A student races ahead and loosens permissions | Enthusiasm | Praise the energy, restore the leash, promise them Session 04 is entirely about earning that | ## What good looks like at the door Every student: ran a job supervised · schedule confirmed for 2 AM · never-list spoken aloud · laptop closed on the way out.