Project brief

Everything converges here. Your capstone is not an essay about AI — it's a working, documented AI workflow for a real recurring task in your actual job, plus a short briefing that could teach a colleague to run it. It's designed to outlive the course: a workflow you keep running every week, and one your team could adopt.

The three deliverables

1. The workflow one-pager (Module 3 format): stations marked [AI]/[ME], the actual prompts, checkpoint placement (Module 5), risk notes (Module 4), and measured before/after time.
2. The evidence pack: real inputs and outputs from at least three live runs, plus one verification log showing your (c)-list and what you checked.
3. The 5-minute briefing (write-up or slides): what the task was, how the workflow works, what it saves, where its limits are — pitched so a colleague with zero AI experience could adopt it.

Choosing your task — the four filters

1. 1Recurring: happens at least weekly. One-offs can't demonstrate a workflow.
2. 2Meaningful: currently costs 30+ minutes per occurrence. Trivial tasks make trivial capstones.
3. 3Green-zone core: the heavy lifting is drafting, compression, translation, or structuring (Module 3's patterns) — not judgment or red-zone data.
4. 4Yours: you own it end-to-end and can run it for real, this week, three times.

The rubric (how your capstone will be graded)

• Prompts (25%) — five-part briefs, template used, examples where style matters.
• Workflow design (25%) — sensible AI/human split; checkpoints upstream where errors are naked.
• Responsibility (25%) — data classes respected, redaction where needed, no judgment delegated.
• Evidence & honesty (25%) — real runs, real timings, limits stated plainly. An honest 'saves 40 minutes, fails on weeks with unusual inputs' beats an inflated claim every time.