To build on the excellent points made by John6666, I would suggest a structured approach for those just starting their journey into AI-assisted development.
A first step is securing a professional subscription to either Google Gemini, OpenAI, or Claude Code all of which are around the $20 monthly mark. After subscribing, install the respective CLI clients for your system to bridge the gap between the AI and your local environment.
Set up a GitHub account, if not already done. Create a fine-grained API key within the account settings to provide to the agent. This enables the AI to handle GitHub pushes and commits directly, which can make the workflow easier and more automated.
For a first build, start with Python and Pygame. Have the CLI agent guide through a simple project, such as a Pong remake or a bullet hell scroller. The goal is to fully flesh out that simple project. In coding projects, remember the 90/90 rule: when you think you are 90% done, you still have 90% to go.
Quick Start: Installing AI Agent Tools (Windows) <-assumption #
The most current way to use these models on Windows is through their official desktop applications or CLI tools. Modern tools like Codex and Claude Code now favor browser-based OAuth, which means no more searching for API keys.
1. OpenAI Codex (Desktop App & CLI) #
OpenAI has shifted toward the Codex Desktop App as the primary command center for Windows. It includes a built-in terminal and better visual diffing than a standard CLI. [1, 2, 3]
Install App: Download directly from the Microsoft Store. #
Install CLI (Optional): If the terminal is preferred, runnpm install -g @openai/codex in PowerShell. #
Authenticate: Open the app or runcodex login in the terminal. The browser will open to sign in via the ChatGPT account.
2. Claude (Anthropic) CLI #
Claude Code is a high-speed agent that lives in the terminal.
3. Google Gemini CLI #
Google provides a dedicated CLI for interacting with the Gemini 1.5 Pro and Flash models.
Pro Tip: If any of these npm commands fail, ensure the latest LTS version of Node.js is installed from nodejs.org.
To make your life easier, you’ll want to **give your agent a **Fine-Grained Personal Access Token (PAT)****. This is more secure than old-school tokens because you can limit the agent’s access to only the specific projects it needs to work on.
Step-by-Step: Creating a GitHub Fine-Grained Token #
Go to Settings: Log in to GitHub, click your profile picture in the top right, and selectSettings . 2. Developer Settings: On the left sidebar, scroll all the way to the bottom and clickDeveloper settings . 3. Personal Access Tokens: ClickPersonal access tokens >Fine-grained tokens . 4. Generate New Token: ClickGenerate new token . You may be asked to confirm your password. 5. Configure the Basics: #
Token name: Give it a name likeAI-Agent-CLI . #
Expiration: Set this to 90 days (or whatever you’re comfortable with). #
Description: “Token for AI CLI agent to manage pushes and commits.” 6. Repository Access: #
Select Only select repositories . #
Pick the specific repository you are using for your game (e.g., your Pong orBullet Hell repo). This ensures the AI can’t touch your other projects. 7. Permissions: Click thePermissions dropdown and set the following underRepository permissions : #
Contents: Set toRead and write (This allows the agent to commit and push code). #
Metadata: Set toRead-only (Usually set by default). #
Workflows: Set toRead and write (Optional—only if you want the AI to manage GitHub Actions). 8. Generate: Scroll to the bottom and clickGenerate token .
Important: Copy that token immediately and save it in a secure place (like a password manager). GitHub will never show it to you again.