Choosing an Agent
PathMX is designed to work with AI coding agents. This guide explains what that means, why it matters, and which tools to consider.
What Is a Coding Agent?
If you have used ChatGPT, Claude, or Gemini in a browser, you have already worked with an AI -- but not with a coding agent.
A coding agent is an AI that can read, edit, and create files directly in your project. You talk to it in natural language, but instead of giving you text to copy-paste, it works inside your actual repository: reading your files, following links between documents, running commands, and making changes on your behalf.
Think of it this way:
- ChatGPT is a conversation partner. You describe your project, paste in snippets, and copy answers back into your files.
- A coding agent is a collaborator sitting in your editor or terminal. It sees your entire project, understands the file structure, and can generate or edit content directly.
This distinction matters for PathMX because the whole methodology depends on agents navigating the repository, following links between files, reading conventions from AGENTS.md, and generating content that fits the existing structure. That requires direct file access that chat apps do not provide.
Chat Apps vs. Coding Agents
Both are useful. They serve different purposes.
Chat apps (ChatGPT, Claude web, Gemini, etc.)
- Great for brainstorming, explaining concepts, and drafting text
- No file access -- you copy-paste everything manually
- No awareness of your repository structure or conventions
- Cannot follow instructions in an
AGENTS.mdfile automatically
Coding agents (Cursor, Claude Code, GitHub Copilot, etc.)
- Live inside your editor or terminal
- Can see your entire project and read/write files directly
- Follow repository-level instructions (
AGENTS.md) automatically - Maintain context across many files in one session
You can start by brainstorming in a chat app and then hand off to a coding agent for the actual work. Both are useful, but PathMX needs an agent that can work with files.
Quick Recommendation
If you want to skip the details:
- Best overall: Cursor or Claude Code (~$20/mo)
- Free option: GitHub Copilot (free for educators)
- If you only have ChatGPT: Use it for planning and drafting, but pair it with a free agent for file editing
What Makes an Agent Good for PathMX?
Not every AI tool is equally suited for curriculum authoring. Here is what matters:
- File navigation: can it browse and read files in a repository? This is essential for following links and type hints.
- Multi-file editing: can it create and edit several markdown files in one session?
- AGENTS.md support: does it automatically read repository-level instructions?
- Markdown fluency: does it work well with prose and structured text, not just programming code?
- Repository context: does it understand the overall project structure (file tree, search, embeddings)?
Comparison
| Agent | Cost | Type | PathMX Fit |
|---|---|---|---|
| Cursor | Free tier / $20/mo Pro | IDE (VS Code fork) | Best overall -- reads AGENTS.md, deep file context, agent mode |
| Claude Code | $20/mo (Claude Pro) | Terminal CLI | Most capable model, great for batch generation and multi-file work |
| GitHub Copilot | Free for educators | VS Code extension | Easy setup, solid free option |
| Windsurf | Free tier / $15/mo Pro | IDE (VS Code fork) | Budget paid alternative to Cursor |
| OpenAI Codex | $20/mo (ChatGPT Plus) | Web + CLI + IDE | Good if you already pay for ChatGPT |
Other options worth knowing about: Aider (open-source, terminal, Git-focused), Cline (VS Code extension with plan/act workflow), and OpenCode (open-source, bring-your-own-model).
Getting Started With Your First Agent
If you have never used a coding agent before, here is the fastest path:
- Install Cursor from cursor.com. It works like VS Code, so any extensions and settings carry over.
- Open your PathMX repo (e.g., the cs1-example) as a folder in Cursor.
- Open the agent panel (Cmd+I on Mac, Ctrl+I on Windows) and try a simple prompt:
Read README.md and AGENTS.md, then summarize this repository.
- The agent should describe the repo structure, the file types, and the conventions. That tells you it is working.
- Now try something real:
Generate a themed variant of lab-01.spec.md for students interested in music.
- Review the output, iterate, and you are off and running.
If Cursor is not an option, the same workflow applies to Claude Code (in your terminal) or VS Code with GitHub Copilot. The prompts are the same -- the only difference is where you type them.
Free Options for Educators
Cost should not be a barrier to trying PathMX.
- Cursor offers a free year for students through their student program. Some educators qualify as well.
- GitHub Copilot is free for verified educators through the GitHub Education program.
- Claude Code and Cursor Pro both cost ~$20/mo if free tiers do not apply. A single month is enough to evaluate PathMX for your course.
Prices and features change frequently. This guide reflects what is available as of early 2026.
Next Steps
- CS1 Quickstart -- clone the starter repo and adapt it with your agent
- Ontology Guide -- understand type-hinted filenames
- CLI Guide -- the PathMX build and preview toolchain