AI-Assisted Teaching Materials Sprint

Tuesday of Week 3 is one continuous all-day event. The day opens with the AI-Assisted Prototyping and Teaching Methods primer (~40 min, 25 slides), then runs the activity below for the rest of the day. This page covers the activity itself: the day flow, what to bring, the four critical-review axes, and the share-out structure. The primer page covers the tool overview and the workflow framing.

Each participant uses cs-class-scaffolding (a Socratic syllabus-builder tool) plus a local coding agent (Claude Code or Codex) to produce, for a real course they would teach: a finalized syllabus, a scaffolded repository, one drafted lecture, and one drafted lab or assignment. The day closes with a 5-minute share-out per participant. The point is not a polished course; the point is to inhabit the prototype-and-critique loop — draft fast with AI, read critically, iterate — and walk out with a workflow you can use the next time you prep a class.

Day flow

Block	Duration	What happens
Topic primer	~40 min	Tool walkthrough, day flow, four critical-review axes, share-out structure.
Setup checkpoint	15 min	Clone the tool, npm install, confirm your Claude CLI login (or paste the institute-issued API key into .env as a fallback), run npm run dev, open http://localhost:5173. Help desk during this block.
Morning sprint: Syllabus + scaffold	90 min	Run the five-phase Socratic dialogue end-to-end against your own course. Finalize the syllabus. Generate the scaffold ZIP. Unzip it. Open the repo in your coding agent.
Working lunch	60 min	Optional: review the AI integrity stance the dialogue drafted for you. Begin reading the scaffold structure.
Afternoon build: Lecture + lab/assignment	150 min	Hand the scaffold to your coding agent. Use the lecture-builder skill to draft one lecture. Use the matching <type>-builder to draft one lab or assignment. Iterate: read, push back, rerun.
Share-out prep	20 min	Prepare a 5-minute show-and-tell: syllabus headline, lecture excerpt, lab/assignment statement.
Share-outs + structured discussion	60 min	5 minutes per participant + 2 minutes peer feedback per slot. Closing 10-minute group discussion on the four critical-review axes.
Closing debrief	15 min (protected)	What worked, what failed, what you would adopt in your own course.

It is OK if the lab/assignment is the weakest of the three artifacts. The closing share-outs are protected even at the cost of cutting the afternoon build short. The loop is what we are after, not a polished course.

Before you arrive

• A laptop with Node.js 18+ and npm 9+ installed.
• A coding agent installed locally — Claude Code (recommended), Codex, OpenCode, or OpenClaw.
• A working Claude CLI login (claude login from earlier institute days). The web app and the afternoon coding agent both use it. If claude isn’t ready on your laptop, the institute will provide an Anthropic API key as a fallback.
• Git installed and a place to clone repositories.
• A code editor you are comfortable with (VS Code, anything).
• A course you would actually teach. If you do not have one in mind, the activity will load a bundled example (CSCI 1013 Intro to AI Programming or CSCI 6531 Computer Security) so you can do the day without blocking on course choice — but real courses produce better artifacts.

What you’re building

By end-of-day, each participant should have produced and shared:

• A finalized syllabus.md in their scaffolded repo, covering the 29-item institute checklist baked into the tool.
• A course-meta.json in their scaffolded repo, structured for downstream agent expansion.
• One lecture in lectures/ (or the location the scaffold chose), generated by handing the scaffold to a coding agent and invoking the lecture-builder skill.
• One lab or assignment in labs/, homework/, project/, or the corresponding scaffold directory.
• A short reflection note on the four critical-review axes as applied to today’s artifacts.

The four critical-review axes

Every AI-generated artifact you produce today gets reviewed against these four axes before you trust it. These are also the axes the share-out peer feedback uses.

• Correctness. Does the syllabus’s week-by-week schedule actually match the topics? Does the lecture’s example code run? Does the lab have a working solution that passes its own tests?
• Feasibility. Can a real student finish this lecture in one class period, or this lab in one assignment cycle? Is the workload honest?
• Privacy realism. Does the policy reflect what your institution actually allows? Does any handled data respect IRB/DSS expectations?
• Security realism. If the course is AI- or security-flavored, does the threat model in the lecture match the lab? Does the lab fail safely for a student who goes off-rails?

These four axes are not a checklist to grind through. They are the questions to ask of every AI-generated artifact, every time, before you trust it.

The tool

The activity depends on cs-class-scaffolding (also called “The Class Factory”), a full-stack web app that drives a five-phase Socratic dialogue and emits a course-repo scaffold. The tool is hosted in the institute organizer’s repository; participants clone it locally for the day.

• Five phases: Discovery → Structure → Assessment → Policies → Finalize.
• Scaffold output: syllabus.md, course-meta.json, README.md with init prompt, plus builder skills under .claude/skills/ and .codex/skills/ (initialize-repo, lecture-builder, website-builder, one <type>-builder per assessment type).
• Runtime: Claude Code CLI (recommended for the day; most participants already have claude login from earlier institute days). The tool also supports Codex, OpenCode, and OpenClaw CLI dispatchers, and Anthropic API direct as a fallback.

The institute mirror URL is projected during the setup checkpoint. Six commands get a laptop to a running install:

git clone <institute mirror URL>
cd cs-class-scaffolding
npm install
cd server && npm install && cd ..
cp .env.example .env                # leave blank if using Claude CLI; paste ANTHROPIC_API_KEY only as a fallback
npm run dev                         # opens at http://localhost:5173

Share-out structure

Each participant gets 5 minutes + 2 minutes of peer feedback for a total of ~7 minutes per slot.

The 5-minute share-out covers, in order:

1. The course. One sentence: title, level, term length, audience.
2. The syllabus. Headline shape: how many weeks, what the assessment types are, what your AI integrity stance ended up being.
3. The lecture. Show one slide or paragraph. Name one thing you accepted and one thing you rewrote.
4. The lab/assignment. Show the task statement. Name one strength and one remaining flaw.
5. The honest take. One sentence: would you actually use this draft in your course, or is it a starting point?

The 2-minute peer feedback uses the four critical-review axes as a checklist. Peers do not rate the artifact; they name the axis on which they have the most useful question for the presenter.

Peer discussion prompts

These prompts run during the closing 10-minute group discussion after all share-outs:

• Where in your existing course preparation does AI-assisted drafting clearly save real time, and where does it create new work?
• What review discipline did you discover today that you would impose on yourself going forward when using AI assistance?
• The dialogue’s “policies” phase asked you to commit to an AI integrity stance for your own course. Did that conversation surface anything you had not yet thought through?
• The agent draft of your lab/assignment — does it pass the security realism axis? If it is an AI-flavored or security-flavored course, would the lab fail safely for an off-rails student?
• What is the smallest piece of this workflow you could realistically take back and use next semester?

Classroom-safety norms (read aloud at the start of the morning sprint)

• Drafts are starting points, not commitments. Anything the AI produces is a draft until you have read it and decided to keep it.
• Do not paste real student records into the dialogue. No FERPA-protected content. No real student names. Anonymized course descriptions are fine.
• Treat the AI integrity-stance draft as a serious document. It will end up in your real syllabus next semester if you keep this work. Write the policy you actually believe in.
• The share-outs are not a critique theater. Frame peer feedback as axis-anchored questions, not as judgments on the artifact.

Common failure modes (read before the morning sprint)

• AI generates a plausible-sounding but wrong week-by-week schedule. Push back; the dialogue prompt is designed to support pushback. If it does not, edit the markdown directly.
• AI drafts a lab whose solution does not run. Read the code. If you do not test it, the lab is not finished.
• AI writes a policy that violates your institution’s rules. Especially the AI-integrity policy. Check it against your department’s actual stance before adopting.
• You spend all of “morning sprint” tuning the syllabus. Hard timer: 90 minutes. If the syllabus is not finalized at 75 minutes, accept what you have and move to scaffold generation.

Using AI assistance during the day

Use it. Coding-agent assistance (Claude Code, Codex, OpenCode, OpenClaw) is the entire point of the afternoon build. The agent is supposed to write the first draft of your lecture and your lab.

Understanding is still required. AI helps you move faster, but you still need to read code, reason about systems, and decide whether each artifact actually meets the four critical-review axes. If the agent generates a lab whose solution doesn’t run, the lab is not finished — regardless of how plausible the task statement reads.

AI assistance in the design of this activity

This activity — its primer slides, its activity stub, the cs-class-scaffolding tool, and these website pages — was drafted with substantial assistance from Claude (Anthropic) and reviewed by the institute team across many iterations. The pedagogical design (the four critical-review axes, the share-out structure, the day flow) was authored collaboratively; the AI helped articulate, draft, and stress-test it.

This is itself a piece of the day’s content: faculty will use AI assistance throughout the day on their own course materials, and the activity that asks them to do so was itself built with AI assistance.

Acknowledgements

The cs-class-scaffolding tool was developed by the institute organizer. The pattern of pairing a tool walkthrough with an all-day build sprint and a structured share-out follows Topic 10 / Activity 10 (Build-it / Break-it / Fix-it). The four critical-review axes draw on Topic 08 (Responsible Use) and Topic 09 (Privacy in AI/ML).