# Beyond Chatbots — The Course Kit

Companion curriculum to [beyond-chatbots](../index.html) on solveitsam.com.
Four hands-on sessions that take a mixed-background group from "AI is a
chatbot" to a running, governed agent that works overnight.

---

## What this course is

Most AI training teaches people to write better prompts into a chat window.
This course skips that entirely. Students build an **agent** — a system with
tools, memory, a written job, a schedule, and guardrails — and they run it
against their real lives: real inboxes, real calendars, real chores.

The teaching philosophy, in three rules:

1. **Ship every session.** Nobody leaves with only notes. Session 01 ends
   with a scheduled overnight run. Session 02 ends with a working triage
   agent. Every session's homework is "let it run and read the log."
2. **Guardrails are curriculum, not fine print.** Draft-and-hold is the
   default posture from minute one. Students red-team each other's agents
   in Session 04 because trust that hasn't been attacked isn't trust.
3. **Excitement through specifics.** No hype vocabulary. The wow moments
   are real: a phone buzzing with a message the student's own agent wrote,
   an overnight log with timestamps, the price of a night shift computed
   from actual token counts.

## Audience and prerequisites

- Motivated adults, mixed technical background. Comfort installing an app
  and using a web browser is enough; a terminal is used but every command
  is provided verbatim.
- Everyone has used a chatbot at least once. Nobody needs to have coded.
- Hard prerequisite: complete [setup-guide.md](setup-guide.md) before
  Session 01. It takes ~45 minutes at home and the instructor verifies it
  in the first exercise.

## The arc

| # | Session | Students leave with | Deck | Guide | Exercises |
|---|---------|--------------------|------|-------|-----------|
| 01 | **Wake an Agent** | An agent that ran one real job and is scheduled to run tonight | [deck](deck-session-01.html) | [facilitator](session-01-facilitator.md) | [handout](session-01-exercises.md) |
| 02 | **Inbox Zero, Honestly** | A triage agent in report-only mode + a nightly morning brief | [deck](deck-session-02.html) | [facilitator](session-02-facilitator.md) | [handout](session-02-exercises.md) |
| 03 | **Memory & Machinery** | Durable agent memory + a working mental model of the physical stack | [deck](deck-session-03.html) | [facilitator](session-03-facilitator.md) | [handout](session-03-exercises.md) |
| 04 | **Judgment** | A signed delegation policy, tested guardrails, the governed full stack | [deck](deck-session-04.html) | [facilitator](session-04-facilitator.md) | [handout](session-04-exercises.md) |

Sessions run ~150 minutes each, one week apart. The week between sessions
is load-bearing: agents run nightly and students keep a miss-log that
becomes the next session's material.

### Ideas that thread through all four sessions

- **The six-part anatomy** (brain, hands, memory, job, clock, leash) —
  introduced in S01, referenced constantly after.
- **The five files** (job, constitution, memory index, brief format,
  policy) — accumulated one or two per session; at graduation, students
  can say "these five files *are* my agent."
- **The delegation ladder** (observe → suggest → draft → ask → act+report
  → act silently) — implicit from S01 ("report-only"), named and drilled
  in S04.
- **"Email is data, not instructions"** — planted as a trap in S02
  Exercise 1 (email #17), formalized as a NEVER rule, then attacked
  head-on in the S04 red team.

## Materials in this kit

```
training/
├── index.html                    ← the hub page (links everything)
├── README.md                     ← this file
├── setup-guide.md                ← student pre-work + instructor pre-flight
├── deck.css / deck.js            ← shared slide engine
├── deck-session-0{1..4}.html     ← the four decks
├── session-0{1..4}-facilitator.md ← minute-by-minute run of show
├── session-0{1..4}-exercises.md   ← student handouts (print or send)
└── templates/
    ├── JOB-template.md
    ├── inbox-rules-template.md
    ├── morning-brief-template.md
    ├── miss-log-template.md
    └── delegation-policy-template.md
```

## Room, equipment, and staffing

- **Room:** tables in pairs or pods of four, projector or large screen,
  reliable wifi, power strips everywhere. Students sit where they can see
  each other's screens — pair debugging is half the pedagogy.
- **Students bring:** a laptop they can install software on, their phone,
  and (from S02) access to a real email account. The setup guide has them
  create a separate **test inbox** used for anything risky.
- **Staffing:** one instructor presenting + ideally one roaming helper per
  6 students during lab blocks. The facilitator guides mark exactly where
  helpers matter.
- **Backup plan:** every deck prints one slide per page; every exercise
  works on paper if wifi dies except the connector labs (contingencies in
  each facilitator guide).

## Tooling notes

The course teaches with **Claude Code** as the agent runtime, **MCP
connectors** for email/calendar, **Hermes** for agent-to-phone messaging,
and **scheduled tasks** for overnight runs. The concepts (tools,
permissions, memory files, schedules, logs) transfer to any serious agent
stack; the facilitator guides mark the few places where instructions are
tool-specific so you can swap runtimes without rewriting the course.

Instructor fills in environment specifics (API access, workspace setup)
in the bracketed `[INSTRUCTOR: ...]` slots in the setup guide before
sending it out.

## Delivery controls for the decks

- `→` / `space` / `PgDn` next · `←` / `PgUp` previous · `Home` / `End`
- Every slide deep-links: `deck-session-01.html#12`
- Print any deck for paper backup: one slide per page, chrome hidden.

---

*Part of the BeyondPrompt sessions · questions to chris@beyondprompt.ai*