Capstone, build a research brief

This capstone ties the track together. You will produce a one-page brief on a topic you care about, using Claude as a thinking partner, and you will check it as carefully as you would check a colleague's work.

Pick something real: a decision you face, a local issue, a tool you are evaluating, a topic for a talk. Avoid anything with private or sensitive data. Use public material so you can practice freely.

The workflow

Step 1. Frame it (lesson 2). Write a prompt with context, task, constraints, and format. Example:

I am preparing a 10-minute talk for a local
meetup on how small shops can use AI assistants.
Audience: shop owners, not technical. Give me a
one-page outline with three main points and one
concrete example for each. Plain language.

Step 2. Ask questions first. Add: "Before you start, ask me up to five clarifying questions." Answer them honestly, including "I do not know."

Step 3. Bring in material (lesson 3). Attach two or three public sources, such as articles or reports you have read. Ask Claude to base the outline on them and to quote the lines it relies on.

Step 4. Choose the model (lesson 4). Draft on a faster model if you have a choice. For the final pass, try a more capable one and compare. Keep the better version.

Step 5. Check privacy (lesson 5). Before you attach anything, ask yourself whether it is public and whether you are allowed to share it. Redact if not.

Step 6. Verify. This is the step people skip. Ask Claude: "List every factual claim in this brief that I should verify, and mark which ones came from my sources and which came from your general knowledge." Then verify the top three yourself using a real source. Fix or remove anything you cannot confirm.

Step 7. Edit in your voice. Rewrite at least two sentences by hand so the piece sounds like you. You are the author. Claude is the assistant.

What good looks like

A good capstone has these marks:

• A clear opening sentence stating the point.
• Every factual claim traceable to a source or flagged as unverified.
• No invented quotes, numbers, or names.
• A short note on how you used Claude and what you changed.

Reflect

Write four sentences at the bottom of your brief:

1. Which prompt change helped the most?
2. What did Claude get wrong or overstate?
3. Which model choice did you make, and why?
4. What would you do differently next time?

Share it

Bring the brief to a club meetup or post it in the community channel. Reading other people's prompts is one of the fastest ways to improve your own. If you want to keep going, Track B moves from chatting with Claude to working with it inside a real project folder.

Try it

The whole lesson is the exercise. Set a timer for 45 minutes and complete steps one through seven. If you only have 20 minutes, do steps one, two, six, and seven, and skip the rest. Finishing a smaller version beats planning a perfect one.