Working with AI · IS218Build a game you can trust.

Your game, one small feature ​

You get a Python app with a browser interface. The game rules are unfinished. Work with AI to make one round playable: choose a move, see the computer's choice, and find out who won.

What you will learn ​

• Turn an idea into a short spec with examples everyone can check.
• Guide AI through a focused change, and ask it to test and review before you play.
• Use tests, code analysis, and formatting checks as repeatable feedback.
• Inspect the result and leave a clear trail from the request to the code and checks.

Learn through the conversation. The agent gives one small step, then waits for you. Predict a round, explain one test together, and describe something you saw while playing. Short answers are fine; ask for a hint or example when you're unsure.

The checks are prepared: pytest checks behavior, Pylint analyzes Python, and Black checks formatting. Valid and invalid cases describe what you try; unit, integration, and real-browser checks describe how much of the system you exercise. See the examples and limits.

The route ​

Start: your GitHub copy → SSH clone → local AI project. Follow the starter README, then open the cloned folder in ChatGPT desktop's Codex mode. Your first conversation explains the project and guides setup one checkpoint at a time, until you can run the unfinished game. Another coding agent can work with the same files and terminal.

Stay in that project to agree on one feature, save the spec, build, test, and review. Keep the change with terminal Git. Use a browser to play; VS Code is optional.

Done looks like: a playable round, checks for all nine matchups, your own browser review, and a merged PR linked to the issue and SPEC.md. No API key or deployment is needed.

Then make it yours: choose one follow-up feature with less guidance—first-to-two scoring, an explanation of why a move won, or your own visual style. Use the same process again.

Start with the starter README →