# From Prompt to Production: a cohesive talk

This edition has **70 slides: 50 narrative slides and 20 embedded activity slides**. The original 50-activity deck remains intact in the parent slides folder. This version is a presentation with a story, selected demonstrations, and invitations to practice.

## The story

A faculty member begins with a teaching need and a blank page. The opening explains why current systems can help produce a usable resource, what a language model does, and why its output still needs inspection. The middle follows that resource through a brief, agent roles, learning design, evidence, visuals, assessment, and a small interactive build. The closing returns to the original teaching need and asks what changed in the quality of the learner's experience.

The recurring campus transport example anchors the workflow. Ecology and association-versus-causation examples show how the process transfers to other disciplines. They are explicitly separate teaching examples, not facts about the same campus project.

## Suggested 90-minute presentation

| Time | Slides | Purpose | Participation |
|---|---|---|---|
| 0–13 min | 1–14 | Why this moment; how LLMs and agents work | Invite a prediction about a possible next word, then distinguish prediction from verification. |
| 13–31 min | 15–28 | Start with learning; preview tools; specify a brief; divide work and revise | Use a prepared brief for the tool previews. Give participants two minutes to improve their own brief on slide 22. |
| 31–41 min | 29–35 | Ideas, Bloom's taxonomy, and UDL | Compare two tasks and discuss whether alternate formats preserve the same outcome. |
| 41–49 min | 36–39 | Evidence trails and tentative gaps | Show one real source and one claim; have the audience decide whether the source supports it. |
| 49–58 min | 40–45 | Images, revision, and slide stories | Compare visual versions and ask which specific revision helps learning. |
| 58–68 min | 46–51 | Rubrics, calibration, and meaningful cases | Have participants briefly rate a fictional response before revealing the agent's rationale. |
| 68–77 min | 52–58 | Build a working learning resource | Open the local example module, test an incorrect answer, and demonstrate one targeted revision. |
| 77–90 min | 59–70 | Synthesis, judgment, reusable workflows, and next steps | Revisit the original artifact; ask what evidence would make participants willing to teach with it. |

The opening and wrap-up are approximately 13 minutes each. The 20 activity slides are opportunities for demonstrations, brief audience decisions, or later practice. Completing every full prompt live requires a substantially longer workshop. Activity durations in the prompt book describe standalone exercises, not an obligation to pause for all of them in this presentation.

## A 60-minute version

Keep a 12-minute opening and a 12-minute closing. Use 36 minutes in the middle: 10 for the brief/plan/revision workflow, 6 for Bloom and UDL, 5 for evidence, 5 for visual iteration, 5 for rubric calibration, and 5 for the working module. Select one of the two tool previews for a live demonstration. Show the other as a comparison. Use the activity prompts as the take-home resource.

Do not rush through the mechanism slides as a list of terminology. Explain the relationship between learned patterns, current context, generated output, and external tools. If time is tight, narrate several mechanism images as one continuous explanation without a separate interaction on each.

## A longer workshop

Use the same opening and closing but extend the middle. Teams select one artifact and carry the same brief through several activities. Keep researcher, writer, and critic outputs distinct, then compare the revised result with the first draft. Do not attempt twenty unrelated builds simply because twenty activities are available.

## Demonstration preparation

1. Open a copy of faculty-starter-kit/. Its course brief, fictional campus scenario, synthetic data, sample responses, and working example-module.html are ready for use.
2. Test your current Cursor or Antigravity installation and account. The workspace illustrations are conceptual, not interface screenshots. The first tool previews use the supplied brief; writing a new brief follows on slide 22.
3. Prepare one first draft, a critique, and a revision for the campus task. A useful visible improvement is replacing “choose a proposal” with “use two observations, explain a tradeoff, and name a limitation.” Preserve the full outcome in the course brief.
4. For research, choose actual papers you have opened. The starter scenario is not a research publication. A gap activity needs a verified source set and a real account of search coverage; missing search records must remain missing, not be generated as fiction.
5. For image iteration, generate a first draft from the reviewed image brief before presenting. Keep a factual correction, a clarity revision, and a style revision separate. The explanatory slides illustrate the process; they do not substitute for inspection of the actual image.
6. Use the supplied fictional responses to calibrate a rubric. Recalculate any numerical claim against sample-data.csv before assessing the reasoning.
7. Open example-module.html locally as a reliable fallback. It works without external libraries or accounts. Ask the audience to predict the feedback for an incorrect answer, then show it.
8. Share the real-text gallery or prompt book. Describe important visual relationships aloud when presenting the image slides.

## Three questions to carry through the talk

- **Does it function?** The file opens, controls work, and the learner can complete the path.
- **Is it supported?** Claims and calculations match the supplied evidence, and limitations remain visible.
- **Does it teach the intended thing?** The student task produces evidence of the desired learning.

These are different questions. A working page can contain unsupported content; an accurate handout can assess the wrong outcome; an attractive task can exclude learners. The closing should bring these questions together through the original example.

## Transitions worth saying aloud

After the mechanism opening: “Now that we know what the system is doing, we can give it a better job.”

Before the learning-design section: “We have a way to produce things. What would make those things worth teaching with?”

Before research: “The model can help organise the inquiry. The evidence still has to come from somewhere we can inspect.”

Before the interactive build: “So far, students could read and discuss the resource. Now they can do something with it and see a response.”

Before the closing: “We have changed the material several times. Which changes actually improved the learning, and who made those decisions?”

At the close: “Choose one outcome and one artifact. Build a version small enough to review, then use your judgment to make it worth using.”