Coding with AI: build something small and verify what it does

Choose a project you can understand

A unit converter, a reading list, or a small expense summary is a better first experiment than a system that handles payments or private customer records. Describe a few examples of correct behavior and at least one invalid input. The clearer the specification, the easier it is to tell whether the result works.

Coding assistants can offer suggestions, discuss a codebase, or prepare changes for review. GitHub's documentation describes several of these capabilities, but the level of access depends on how the assistant is being used. GitHub: What is GitHub Copilot?

Ask for a bounded first version

Build a small browser-based unit converter for inches and millimeters. State the conversion factor, allow decimal inputs, reject invalid text clearly, and show the units beside the result. Keep the code readable and explain how to run it. Do not add accounts, analytics, or external services.

Run the smallest version before asking for extra features. Check a known value and the reverse conversion. Test zero, a decimal, and an invalid entry. If the result is wrong, report the exact input, expected output, actual output, and any error message.

Review the changes

Ask which files changed and why. Read the part that handles calculations and data. An assistant may introduce a dependency to solve a problem that needs only a few lines of code. A generated project may also include placeholders or simulated data that look real. Remove or label them before sharing.

Never paste API keys or passwords into public code. Use the platform's normal secret-handling mechanism for a project that needs credentials. If the assistant can execute commands, keep its access within the project and review consequential operations. A coding workflow can affect more than what appears in the browser.

Know when a prototype needs more work

A personal utility and a public service have different requirements. A public app may need security review, accessibility, backup, monitoring, and a way to handle user mistakes. Passing a few examples does not establish that the software is safe for sensitive or high-volume use.

Keep a short README explaining the purpose, setup, limitations, and checks performed. Save a known working version before extending it. If you cannot explain a critical calculation or permission, pause to understand it. AI can shorten the path to a prototype, but ownership of the result still includes knowing what it does and how to recover when it fails.