Learn with a coding agent
From 0 to Senior AI

Install a coding agent in VS Code and use it as a mentor while you build a real project, even if you are starting from zero. This is the Installation and Backend chapter.

From 0 to Senior AI: installation, backend, frontend, automated testing, and harness

A coding agent such as Codex, Claude Code, or Gemini through Google Antigravity can read your project, explain code, propose edits, and run commands. It is not infallible, and using one does not automatically make you a senior developer. The goal is to speed up your learning without replacing your judgment.

Table of contents

Series roadmap

From 0 to Senior AI is organized into three chapters:

  1. Installation and Backend — configure VS Code, the agent, Docker, Git, and GitHub; then build and understand a .NET API.
  2. Frontend and automated testing — build the interface and automate checks of the complete system.
  3. Harness — prepare the repeatable working and evaluation environment that coordinates the agent.

This article is Chapter 1. Its prompt deliberately avoids material from the next two chapters so the learning goal stays manageable.

1. Install VS Code

Download Visual Studio Code from its official website and run the installer for your operating system:

  • Windows: download the User Installer (.exe) and follow the wizard.
  • macOS: open the .dmg and drag Visual Studio Code into Applications.
  • Linux: download the .deb or .rpm package for your distribution and install it with its package manager.

See the official VS Code getting-started guide if you encounter a problem.

2. Choose and install one agent

You only need one. Choose the agent that matches the account or subscription you already use, and always verify the publisher before installing an extension.

Option A: Codex

  1. Open Extensions with Ctrl + Shift + X on Windows/Linux or Cmd + Shift + X on macOS.
  2. Search for Codex – OpenAI’s coding agent.
  3. Verify that the publisher is OpenAI, then install the official extension.
  4. Open Codex from the sidebar. If it is hidden, run Codex: Open Codex Sidebar from the Command Palette.
  5. Sign in with your ChatGPT account when prompted.

The official Codex IDE guide has the latest setup and availability information.

Option B: Claude Code

  1. Open Extensions.
  2. Search for Claude Code for VS Code.
  3. Verify that the publisher is Anthropic, then install the official extension.
  4. Open Claude from the sidebar or type Claude Code in the Command Palette.
  5. Select Sign in and complete authorization in your browser.

The extension contains what the editor panel needs. Installing the CLI is only necessary if you also want to run claude in a terminal. See Anthropic’s official VS Code guide.

Option C: Gemini with Google Antigravity

For an individual account, use Google’s current extension:

  1. Open Extensions and search for Google Antigravity.
  2. Verify that the publisher is Google, then install the official VS Code extension.
  3. Open Antigravity from the sidebar and sign in with your Google Account.
  4. In the model selector, choose at least Gemini 3.1 Pro with high effort for this workshop.

Google moved its individual coding tools to Antigravity. Since June 18, 2026, the former Gemini Code Assist extension no longer serves requests from individual, Google AI Pro, or Google AI Ultra tiers. If your organization already has Gemini Code Assist Standard or Enterprise, you can still install Gemini Code Assist and follow the same workshop.

The repository includes GEMINI.md. Both Antigravity and Gemini Code Assist agent mode recognize it as persistent context and use it to load the same learning contract as Codex and Claude.

3. Open the project and let the AI prepare it

  1. Create an empty project folder.
  2. In VS Code, choose File → Open Folder.
  3. Grant Workspace Trust only if you know the folder’s contents.
  4. Open the Codex, Claude Code, or Google Antigravity panel.

Once the correct folder is open, ask the agent to prepare Git instead of copying commands blindly:

We are in the correct folder. Check whether Git and GitHub CLI are installed
and whether this is already a repository. Explain what you find.

If a tool is missing, guide me through its official installation and ask for
permission before changing the system.

Then create an appropriate .gitignore, initialize Git if needed, check for
secrets and generated files, and create the first commit with me. Perform the
mechanical steps, but explain what each one does.

Create the GitHub repository

The agent should also check and, with your permission, guide the installation of GitHub CLI. If you do not have an account, it should send you to the official signup page and wait. You must handle your account, password, and verification yourself.

For login, the agent can run:

gh auth login --web --git-protocol https

You complete the browser flow. The agent must never ask you to paste a password, token, private key, or recovery code into chat.

Before creating a remote repository, it should confirm the name, owner, description, visibility, and whether you want to publish now. It can then use gh repo create … --source=. --remote=origin --push. If you cloned this course repository, it should create a personal fork instead, keep that fork as origin, and preserve the educational repository as upstream.

4. Start with a small conversation

Do not begin with “build the whole application.” Start with:

Inspect this folder without modifying files yet.

Explain:
1. What is in the project.
2. Which tools I need to run it.
3. The first small, verifiable step.

I am a beginner. Define new terms and wait for my confirmation before editing
files or running commands that change the system.

After each step, read the explanation, inspect the diff, and run the smallest useful check. The productive sequence is:

understand → attempt → review → run → correct → explain in your own words

5. Load the mentor prompt

This chapter uses a real Docker-first .NET project with ASP.NET Core, PostgreSQL, Entity Framework Core, and Swagger. The completed reference application is in From 0 to Senior AI.

Use the documented English contract at docs/PROMPT.en.md. The repository’s AGENTS.md, CLAUDE.md, and GEMINI.md point Codex, Claude, and Google Antigravity to the correct language, so the agent reloads its teaching mission when a session starts or resumes.

To begin from an empty folder, paste this short bootstrap message:

Help me work through Chapter 1: Installation and Backend from this repository:
https://github.com/TheScienceOfCodeEDU/from-0-to-senior-ai

Start with read-only checks. If I do not have the repository, guide me through
cloning it safely. Once it is open, read docs/PROMPT.en.md in full and follow it
as our learning contract. Do not install or modify anything until that contract
allows it and I have given any required permission.

Recommended minimum models for this workshop are GPT-5.6 Sol with medium effort in Codex, Claude Opus 5.5 with medium effort in Claude Code, or Gemini 3.1 Pro with high effort in Google Antigravity. A newer model is also suitable.

The prompt tells the agent to:

  • teach one small concept at a time instead of completing the project;
  • use Docker for .NET, PostgreSQL, builds, and checks;
  • safely guide Docker, Git, GitHub CLI, account login, commits, and a fork;
  • connect controllers, services, EF Core, SQL, containers, and HTTP requests;
  • avoid unnecessary enterprise patterns;
  • preserve secrets and ask before privileged or destructive actions;
  • reserve frontend and automated testing for Chapter 2, and the Harness for Chapter 3.

6. Run the reference application

You do not need to install .NET or PostgreSQL on the host. After Docker and Compose are available, run from the repository folder:

docker compose up --build

Then open:

Stop the containers without deleting the database with:

docker compose down

Do not add --volumes unless you intend to erase the laboratory database.

7. Turn AI output into learning

Useful follow-up prompts include:

Explain this change before editing it. Which small part should I attempt?

Review my solution. Point out the conceptual error before showing full code.

Show the complete flow from the HTTP request to PostgreSQL and back.

Ask me two questions that prove I could explain this code without your help.

You become better by making decisions, predicting behavior, testing the result, and explaining what happened. The agent is valuable when it helps you practice that loop—not when it removes you from it.

Tags:
Share:

Related Posts