Using AI code completion as a daily habit

Write the comment first; let the AI write the code

After this, you'll be able to use Cursor (a free code editor with AI built in) or VS Code with GitHub Copilot to produce a working block of code from a plain-English description, without needing to know any syntax first.

Before you start

Complete Chat Fluency first; this lesson builds on writing specific, context-rich requests that AI can act on.

The idea

You've avoided writing code because you don't know the syntax. That's not a gap you need to close first. Cursor, GitHub Copilot, and similar editors predict the next line of code the same way they predict text: based on the surrounding code in your file. This means you can describe what you want in plain English as a comment, and the tool translates that description into code. No syntax memorization required.

Here is the before and after: without AI completion, you would search online for the syntax, click through several links, copy a snippet, and edit it to match your variable names; the whole process takes 10 to 15 minutes. With AI completion in Cursor, you type a comment like '# Read the CSV at filepath and return a list of rows as dictionaries' and press Tab. The editor predicts the code using your file as context. Always review what it generates before accepting.

Now try it. Open Cursor or VS Code with Copilot, create a new file, and write a comment describing one small thing you want the code to do. Tab accepts the suggestion; Escape rejects it. The exercise below uses Claude's chat as a second path when you want a complete block without opening an editor. Both paths require the same skill: describing what you want precisely enough that the AI can write it.

Try it (10 min)

Watch out for

Tab accepts whatever the AI predicted, including mistakes. Accepting without reading lands broken logic in your file with no indication anything went wrong.
A vague comment like '# do the thing' produces a vague suggestion. The AI predicts from what you wrote, not what you meant.
A comment that names the task but not the input and output ('# process the files') leaves the function signature ambiguous. The AI picks its own assumptions, and they may not match yours.
Generated code often uses placeholder file paths and variable names that don't exist in your project. Running it without reviewing them first produces errors unrelated to the logic.

Paste this into Claude

I want to practice writing the precise comment that makes AI code completion work. My task is [describe what you want to automate: what it does, what goes in, what comes out; be specific]. Write me three things: first, the one-line comment for this function in the format '# [verb] [what it does] from [input] and return [output]'; second, the function signature on the next line; third, the complete function body. Show all three so I understand what a precise comment generates. Then tell me: was my description specific enough to write the comment without guessing, or did you have to fill in details I left out?

What good looks like

The comment names the input and the output explicitly, not just the task
Reading the comment alone tells you exactly what the function does, takes, and returns
The function name in the signature matches the action the comment describes. The signature is the line that names the function and lists what it takes in.
You could hand this comment to someone else and they would write the same function without asking follow-up questions

When this breaks

Breaks when the comment is ambiguous because the AI picks the most common interpretation, which may not match what you meant
Breaks when input and output types are unstated because the AI guesses. The function signature then reflects its guess instead of your data.
Breaks when the verb is fuzzy ('handle', 'process', 'manage'). The generated function lands near what you wanted, not on it.

You can now

Open Cursor or VS Code with Copilot, paste your comment as a # line in a new file, and press Tab. If the generated code matches what Claude wrote in the chat exercise, your comment is precise enough. If they diverge, your comment was ambiguous. Tighten the input and output names and try again.

Key takeaways

Writing a precise comment is the real skill. AI completion translates your description into working code. The better your description, the better the code.

AI completion predicts code based on the comments and variable names already in your file
Writing a plain-English comment and pressing Tab is the core workflow; no syntax memorization needed
Tab accepts, Escape rejects; you always review the suggestion before it lands in your file
You are learning to describe code precisely, not to memorize syntax

Was this helpful?

← Back to lessons

I want to practice writing the precise comment that makes AI code completion work. My task is [describe what you want to automate: what it does, what goes in, what comes out; be specific]. Write me three things: first, the one-line comment for this function in the format '# [verb] [what it does] from [input] and return [output]'; second, the function signature on the next line; third, the complete function body. Show all three so I understand what a precise comment generates. Then tell me: was my description specific enough to write the comment without guessing, or did you have to fill in details I left out?