Build one small behavior you can test. A working demo, clear limits, and a readable change are a better first milestone than a large unfinished app.
Pick a project that teaches you how to check
Claude Code can work with project files and help make software changes. For a first exercise, build a small follow-up queue using fictional enquiries. Each item has a name, next action, and due date; a button marks it complete. This is deliberately small enough to inspect without understanding an entire application stack.
The learning goal is to turn an intention into observable behavior. “An enquiry can be completed and removed from the active list” is testable. “An intelligent sales platform” is too broad. Write the expected behavior before generating code, and keep that short brief beside the finished page.
Get into a clearly bounded project
Follow Anthropic's current quickstart for your operating system and account access. This tutorial uses its terminal interface; other interfaces have their own setup. After installation, run claude --version to confirm the command works. Create a practice folder, open a terminal there, and start claude. Complete the offered sign-in flow.
A project folder is simply the place containing this exercise's files. Keep it separate from important documents and production work. If you already have a repository, use a new branch and understand its existing instructions. Ask Claude to tell you its working directory so you can verify it is looking at the intended material.
Describe the smallest useful version
Ask for an explanation of the proposed files and how you will open the result. Specify a single local web page with sample entries, a date sort, and a completed state. No account system, outgoing messages, or live customer information is needed for this exercise. Ask it to label the data as fictional in the interface.
Decide persistence explicitly. For the first version, it is fine for changes to disappear on refresh if the page clearly says so. Browser storage can become a separate second lesson. Without this decision, a demo may look like a usable record system while silently losing work.
Make the review part of the task
Anthropic's guidance emphasizes concrete verification and planning when the approach is uncertain. Ask for checks tied to the behavior you requested. The queue should sort dates correctly, keep two enquiries with the same name distinct, handle an empty list, and let a keyboard user activate completion.
Review permission requests in relation to the task. If the assistant proposes installing a large framework for a tiny page, ask what requires it. If it reports success, request the evidence: what it opened, which checks ran, and what it could not inspect. An explanation of expected behavior is different from observing the running page.
Try to break your finished demo
Open the page yourself. Complete the first item, then the last one. Try a long enquiry name. Refresh and confirm the behavior matches the persistence notice. Test it on a narrow screen and use the keyboard without a mouse. Report a problem with the action you took, the result you expected, and the result you saw.
Once it works, ask for a short README explaining how to run it, what it does, and its limitations. If using version control, review the changed files before committing. A sensible next iteration adds one capability, such as saving sample entries locally, together with a check for that capability. Resist adding ten features before understanding the first.
Prompts to put it into practice
Replace the brackets with your context. Use information you have permission to share.
Build the practice queue
In this practice folder, build one local web page for a fictional follow-up queue. Show six sample enquiries with name, next action, and due date. Let me sort by due date and mark individual enquiries complete. Label the examples fictional and explain that changes reset on refresh. Use the simplest suitable implementation. First explain the files and how I will open it. Then build it and check sorting, duplicate names, empty state, and keyboard operation. Report what you verified and any remaining limits. Do not connect external services or deploy it.
Try it on your work
Complete the demo, deliberately test three awkward cases, and make one small revision based on what you observe. Keep the original brief and the final README together.
Check the result
- The page runs and the visible sample data is clearly fictional.
- Sorting and completion work for individual records.
- The refresh behavior matches the notice shown to the user.
- You can describe the changed files and the checks performed.
Sources & further reading
Product documentation checked October 16, 2026. Features, labels, and access can change. Examples and practice prompts are original AI Teachers teaching material.