⬇ Save as PDF / Print

Stack Basics · Module 3 of 5

Module 3 — Coding agents: your build crew

Everything in this course — and everything in the paid kit — assumes you're not typing every line yourself. A coding agent is an AI that reads your repo, runs commands, edits files, and iterates until the thing works. The difference between people shipping in 2026 and people watching tutorials is mostly that the shippers picked one of these and got comfortable.

Here's the honest tour. Prices are ballpark as of this writing — check current pages before you commit.

The main players

Claude Code (Anthropic — terminal). Lives in your terminal, works directly on your files, runs your commands, fixes its own errors. This is what I use to run my whole fleet — servers, deploys, debugging, even the marketing. Comes with Anthropic's subscription plans (Pro ~$20/mo gets you real usage; heavier plans exist), or pay-per-token via API. Strengths: long multi-step tasks, real ops work (ssh, systemd, logs), honesty about what failed.

Cursor (~$20/mo). A full code editor (VS Code fork) with the agent built in. If you want to SEE the code while it changes and click around like a normal IDE, start here. Very popular first agent for a reason.

Codex CLI (OpenAI — terminal). OpenAI's answer to Claude Code, bundled with ChatGPT plans. Same terminal-agent shape. If you already pay for ChatGPT, you may already have it.

Aider (free, open source — terminal). Bring-your-own-API-key. Pairs beautifully with cost hacks from Module 4 (point it at cheap or free models). More manual than the big two, but the price of the tool itself is zero.

Windsurf, Zed, and friends. The field moves monthly. The good news: the skills transfer. Learn to work with one agent and you can switch tools in an afternoon.

Which one, for you

• Never coded, want visual guardrails → Cursor.
• Comfortable in a terminal, want maximum leverage → Claude Code.
• Already pay for ChatGPT → try Codex CLI before buying anything.
• Zero budget for tools → Aider + cheap API models.

Any of them can build everything this course describes. Pick one and stop shopping.

The five habits that make agents actually work

Owning an agent isn't the skill — working with one is. These five habits are most of it:

1. Small asks, verified. "Set up the systemd unit and show me it survives a kill" beats "build my whole server." Small steps, checked each time, compound fast; giant asks produce giant messes.
2. Git before everything. git init on day one, commit after every working step. The agent can then experiment freely because YOU can always roll back. This is your seatbelt — non-negotiable.
3. Make it verify. End asks with "…and prove it works" — run the test, curl the endpoint, show the output. Agents that must demonstrate success lie to you far less.
4. Give it context files. A short note in the repo (commonly CLAUDE.md or AGENTS.md) with what the project is, how to run it, what not to touch. Ten minutes writing this saves hours of the agent guessing.
5. You own the decisions. The agent executes; the choices — what to build, what's on or off, what's worth money — stay yours. That's Module 5's whole subject, and honestly, the entire reason this course exists.

The mental shift

You're not "learning to code" in the old sense. You're learning to be a very small company's technical director, with a tireless employee who types 1000× faster than you and needs clear instructions and code review. Every module in this course is written to be handed to that employee — literally paste it in and say "help me do this."

Next module: what all of this costs to run — model pricing, the OpenRouter free-model hack and its tradeoffs, and when a local LLM on your own hardware is genuinely enough.

---

Disclosure: some links in this course are affiliate links (including Amazon). They help fund the free course — your price never changes, and every product here is one we'd recommend anyway.