AdapttoAI Academy — Internal Draft
AI Cowork Bootcamp — Session Map
Draft v1, from a brainstorm pass on 2026-08-28. 4 live lessons — theory
and demo only, delivered by us — with async pre-work carrying the actual participant
effort. One "big, hairy, scary" project per participant threads through the whole cohort.
1The Agentic Mindset
Theory
- Cowork vs. chat — the agentic loop: plan → execute → report
- "Delegating the work, not the judgment"
- What makes a task Cowork-shaped: multi-step, file-producing, tool-using
- Shaping beat — is your big project Cowork-shaped? Five-ingredient test applied live to the room's declarations
Demo
Camera on Cowork. A well-shaped, impressive task delegated cold: watch it plan, ask clarifying questions, execute, and produce a real deliverable. Shows the ceiling — what delegation looks like when it works.
Pre-work
- Setup checklist, paid-plan access confirmed (hard rule)
- One-page warm-up reading: what Cowork is
- Declare your big, hairy, scary project — the major time drain or long-deferred ambition
Big-project thread → declare
2Delegating Well
Theory
- The delegation process: name deliverable / inputs / nuance
- Answer clarifying questions specifically; steer when it drifts; review with discernment
- Context intro — "the more it knows, the better the next one"
- Picking the right worker — Haiku / Sonnet / Opus + effort level, as a decision habit, not a spec sheet
Demo
Camera on us. A deliberately messy task: first pass comes back mediocre — watch how we phrase the ask, catch the drift, steer, and judge the result. Re-run with richer context to show the delta. Shows the road — what the operator does.
Pre-work
- First real delegation — carve the first Cowork-delegable slice off your big project and hand it over
- Bring the result, good or bad (the misses feed L2's steering beat)
Big-project thread → first slice
3Skills & Reach
Theory
- Recap-bridge (~5 min): standing context — "you know this move from the bootcamp; here's what it feeds now"
- Skills — the centerpiece. What they are, the work→skill loop, when a repeated gesture earns one
- Extending reach across time: scheduled tasks
- Extending reach across systems: Claude in Chrome — the no-connector workaround
- Safety, taught adjacent to Chrome: permission modes, the five hazards, prompt injection on live web
Demo
Capture a skill live from a repeated gesture · a scheduled task firing · Chrome acting inside a logged-in system · one hazard scenario walked end-to-end ("Cowork prepares; you ship").
Pre-work
- Set up your big project as a Cowork Project (instructions + context)
- Re-delegate a slice with that context; compare against the pre-L2 run
- Spot and write down your repeated gesture — the skill candidate
Big-project thread → institutionalize
4From Task to System
Theory
- The ladder you climbed: task → context → skill as one compounding loop
- Evals — "an eval is just a try-out," the confidence gate before trusting a skill unattended
- From my skill to our plugin — team sharing, individual skills becoming org expertise
- The maturity roadmap — four proactive levels (inform → suggest → draft → self-improve) as your post-course trajectory
- Bookend with L1: judgment stays yours
Demo
Open by showcasing 2-3 participant projects ourselves, curated from pre-L4 submissions (week-0 vs. now). Then run the eval loop live: request in → judge the output → tell it what to fix → re-run. No live participant demos.
Pre-work
- Build the skill from your identified gesture; run one try-out (eval) on it
- Submit your progress snapshot: big project at week 0 vs. now — feeds the L4 showcase
Big-project thread → show progress
Locked design decisions
Track shapeHorizontal content, task-flavored by each participant's own project — not department modules
Entry barAssumes AI Literacy Bootcamp graduates — no compressed on-ramp
Vendor scopeClaude-first, category-framed — ChatGPT Work stays a one-line comparison
Paid-plan ruleStated in prework (Pro/Max/Team/Enterprise required)
CapstoneShip a skill extracted from the big project — no live participant demos
Cowork 101 course mapping
Source: knowledge/sources/anthropic-cowork-101-course/ — 14 official lessons compressed into our 4.
| Our lesson | 101 lessons folded in |
|---|
| L1 | L1 What is Cowork · L2 Setup · L3 What it can do for you (own-task ID starts here in the source too) |
| L2 | L4 Hand Claude your first task · L5 Get better results faster |
| L3 | L6 Global instructions/projects (recap only) · L7 Skills · L8 Plugins · L11 Safety (full) · L9 Claude in Chrome retained (reach-extender framing) — L10 M365 dropped |
| L4 | L12 Validating skills (evals) · L13 Share with your team · L14 Wrap up |
Open items before this locks
- Verify Claude in Chrome plan/rollout availability against the wiki before promising the L3 demo.
- Check the two standing wiki contradictions before any lesson asserts them as fact: Projects availability (desktop-only vs. all-surfaces) and Compliance API coverage.
- Who pays for participant seats (client org / participants / bundled) — pricing/sales question, does not block curriculum design.
instructor-notes.md needs explicit guidance on reshaping a declared big project that isn't Cowork-shaped, and recovery moves when a task doesn't fit a lesson's exercise shape.- Per-lesson production files not yet built — this session map is the design layer.