How to Use an AI Coding Assistant: A Beginner’s First Session, Step by Step

Your first session with an AI coding assistant sets habits that will either make you faster or quietly make you dependent. This walkthrough takes you from nothing installed to a small working program you actually understand — using GitHub Copilot’s free tier inside the free VS Code editor, the lowest-cost setup GitHub offers. Every step has a purpose, and the last three steps are the ones beginners skip: running the code, questioning the code, and writing down what you learned. If you have not read the pillar guide to AI for coding beginners yet, start there for the concepts; this article is the hands-on companion.

Step 1: Install the editor and the assistant

Download VS Code from Microsoft (free, Windows/Mac/Linux) and create a free GitHub account if you do not have one. Inside VS Code, open the Extensions view and install the GitHub Copilot extension, then sign in with your GitHub account when prompted. GitHub’s plan documentation describes Copilot Free as a plan that “includes limited access to a selection of Copilot features allowing you to try AI-powered coding assistance at no cost” — no card, no trial clock. Students can do even better: verified students get Copilot’s features free under the Student plan, and GitHub says verified teachers and maintainers of popular open-source projects get Pro at no cost. For this walkthrough, the free tier is enough; you are learning a workflow, not buying capacity.

Step 2: Create a deliberately tiny project

Make a new folder called first-session and open it in VS Code. Create one file — quiz.py if you are learning Python, or quiz.js for JavaScript. The project is a five-question quiz that asks questions, checks answers and reports a score. Tiny is a strategy, not a lack of ambition: on a small project you can hold the whole program in your head, which means you can actually review what the assistant suggests. Reviewing is the skill this session is really teaching. On a huge project you would be rubber-stamping — the failure mode the checking checklist exists to prevent.

Step 3: Get your first suggestion — and interrogate it

At the top of your empty file, type a comment describing the first piece: # Ask the user their name and greet them. Pause. The assistant will offer a greyed-out completion. Before you press Tab, read it as if a stranger handed it to you: What does it print? What does it store? Is the variable name sensible? Then accept it, run the file, and watch it work. Now the important move: open the Copilot chat panel and ask, “Explain the code in quiz.py line by line, as if teaching a beginner.” Compare its explanation with what you thought the code did. Where they differ, one of you is wrong — find out which. GitHub describes this assistive mode plainly in its Copilot documentation: the tool “responds while you work, suggesting code, answering questions, or explaining code. You review and apply each suggestion.” Review first, apply second. That order is the entire discipline.

Step 4: Build the quiz in slices, not in one gulp

Resist the urge to type “write me a quiz program” and accept 80 lines at once. Instead, work in slices you can test individually: (1) a list of questions and answers; (2) a loop that asks each question; (3) a check that compares the reply, ignoring capitalisation; (4) a score counter and a final message. For each slice, write the comment or the first line yourself, let the assistant complete, read the completion, run the file, and only then move on. When a suggestion is longer than about ten lines, that is your signal to slow down: ask the chat to explain it before you keep it, or reject it and ask for a smaller piece. You are practising the professional workflow in miniature — small changes, constant testing — and the assistant’s speed makes it tempting to skip exactly the steps that make the speed safe.

Step 5: Break it on purpose and debug with the chat

Now deliberately introduce an error: change a variable name in one place only, or delete a colon. Run the file, read the error message, and before asking for help, write down your own guess about what is wrong. Then paste the error into the chat and ask: “Explain this error in plain English and point me to the line — don’t rewrite the file.” Compare its diagnosis with your guess. Error-driven learning is where assistants shine for beginners: the feedback loop that used to require a teacher, a forum post and a day’s wait now takes a minute. Stack Overflow’s 2025 survey found 51% of professional developers use AI tools daily; debugging help and explanations are a large part of why — but the guess-first habit is what turns a crutch into training, because it keeps your own diagnostic muscle working. The same guess-first habit powers the general routine in our guide to checking AI answers.

Step 6: Review the whole file against the checklist

Finish by reading the complete program top to bottom and running through the beginner’s review checklist: Can I explain every line? Did I test an edge case (an empty answer, a wrong answer, all answers right)? Is there anything here I would be embarrassed to call mine because I don’t understand it? Anything that fails the test gets a chat question (“explain lines 20–25”) or gets rewritten by you, smaller. Finally, save a note in the project — three lines: what you built, one thing the AI got wrong, one thing you learned. That note is your evidence that this was a learning session and not a copying session.

What this means

A first session feels like a setup task, but it is really a negotiation about who does the thinking. The tools are designed to make accepting effortless — one keypress, and code appears — and every part of this walkthrough exists to slow that moment down just enough for your judgement to stay in the loop: tiny project, small slices, explain-before-accept, guess-before-asking, review-before-finishing. Beginners who build those frictions early get the documented benefits of these tools — GitHub’s research found users reporting less frustration and better flow — without the dependency that makes other learners helpless the first time the assistant is unavailable or wrong. There is also a longer game: the habits of slicing work, testing constantly and reviewing output are precisely the habits professional teams use with agentic tools that edit whole files. Your toy quiz is not a toy exercise. It is the professional workflow at beginner scale, and repeating it weekly is how the assistant ends up training you instead of replacing you. This matters beyond your desk, too: as assistants become the default first screen a new programmer sees, the quality of a beginner’s first session increasingly decides whether the next generation of developers reads code or merely approves it.

There is a wider point here for anyone starting out in the US tech market: the first-session habits scale directly into workplace expectations. Teams adopting agentic tools report the same pattern this walkthrough teaches — small tasks, reviewed diffs, tests run early — because it is the only way the speed survives contact with real code. A beginner who practises on a quiz file is quietly rehearsing the stand-up meeting where they will one day explain what the AI changed and why it is safe. Nobody asks how fast you can press Tab; they ask whether the change can be trusted, and that answer is built in sessions exactly like this one.

Bottom line

Slow the accept key down. Small slices, explanations before acceptance, a guess before every answer — the habits of a good first session turn a code generator into a teacher instead of a crutch.

Read Next

• AI for Coding Beginners: How AI Coding Assistants Work and How to Start Learning With Them
• How to Check AI-Written Code: A Beginner’s Review Checklist

Sources

• About GitHub Copilot — GitHub Docs (accessed 7 November 2026)
• Plans for GitHub Copilot — GitHub Docs (accessed 7 November 2026)
• AI section, 2025 Developer Survey — Stack Overflow (accessed 7 November 2026)