Lesson 32 of 36 · Track 4: Automate and code with AI

Claude Code vs Cursor vs Codex: which AI coding tool should you use?

Beginner12 min5 steps1 videoFree · no login

The short answer

All three are coding agents that can read your project, edit files and run commands. The real differences are where you work (terminal, editor or app), which models you get, and which subscription you already have. Claude Code suits Claude subscribers, Cursor suits people who want an AI editor, and Codex suits ChatGPT subscribers. Non-coders who just want an app can start with a browser app builder.

1
First, the three shapes of AI coding tools
AI coding tools come in three shapes. Editors put AI inside the app where you write code: Cursor, or VS Code with GitHub Copilot. Agents take a task and work through it on their own, in a terminal, editor panel, desktop app or the cloud: Claude Code, Codex, and Cursor’s Agent. App builders like Lovable and Bolt run in the browser, so you describe an app and get a working one without opening any code.
The lines are blurring. Claude Code now has editor extensions and a desktop app, Codex lives in ChatGPT and in editors, and Cursor has cloud agents. So pick by how you like to work, not by labels.
1. App builder: describe, get an app
2. AI editor: you code, AI helps
3. Coding agent: you assign, AI codes, you review
2
The big three, side by side
Claude Code (Anthropic) started in the terminal and now also runs in VS Code, JetBrains, a desktop app and the web. It uses Claude models. It needs a paid Claude plan: Pro is $20 a month, Max $100, Ultra $200, or pay per use through the API (as of Nov 2026). Instructions go in CLAUDE.md.
Cursor is a full code editor built on VS Code, with Tab autocomplete, an Agent, Plan Mode and cloud agents. Plans as of Nov 2026: Free, Plus $8 (Grok models only), Premium $30 (adds Claude, GPT and other third-party models at API list price), Super $100 and Ultra $200. Instructions go in .cursor/rules or AGENTS.md.
Codex (OpenAI) is a coding agent inside ChatGPT, plus a desktop app, an editor extension and an open-source CLI. It uses OpenAI models and your ChatGPT login. ChatGPT Free lists limited Codex access, and Plus ($20 a month) gives more. Instructions go in AGENTS.md.
- Pick Claude Code if you already pay for Claude, or like working in the terminal with any editor.
- Pick Cursor if you want to see and edit code in one AI-first editor.
- Pick Codex if you already pay for ChatGPT, or want cloud tasks running while you do other things.
3
GitHub Copilot and the app builders
GitHub Copilot works inside VS Code and other editors you may already use. As of Nov 2026, Copilot Free gives 2,000 code completions a month, Pro is $10 a month, Pro+ $39 and Max $100, and paid plans include the Copilot coding agent. Verified students can get the Copilot Student plan through GitHub Education, which makes it a strong first pick for students.
Lovable and Bolt are browser app builders for non-coders: you chat, and they build and host a web app. They are great for a quick prototype or a portfolio demo. We could not verify their current prices, so check lovable.dev/pricing and bolt.new/pricing. When your app grows, you will want to understand the code, and that is where the tools above help.
4
A simple decision guide
You don’t need the "best" tool. You need one you will use every day. Start with what you already pay for, try the free tiers, and give one tool a full week before switching.
- Never coded, want a quick app: try an app builder, or Claude Code’s desktop app or the Codex app, which need no terminal.
- Learning to code: Cursor or VS Code with Copilot, so you see every line. Students should check the free Copilot Student plan.
- Already pay for ChatGPT Plus: start with Codex. Already pay for Claude Pro: start with Claude Code.
- Smallest budget: Copilot Free, Cursor Free and ChatGPT Free (limited Codex) cost nothing to try.
- Working in a team: use the tool your team uses, and share one instruction file (AGENTS.md works in Codex, Cursor and Claude Code).
Key idea
The habits matter more than the tool: small clear tasks, a test to check, and a careful review of every change.
5
What we could not verify, and what changes fast
We took every price and feature here from each company’s official pages in Nov 2026. We did not run our own head-to-head tests, so we don’t claim one tool writes better code. Model rankings change every few months, and any single test result is a snapshot.
Prices and limits change often too. Cursor rebuilt its whole plan line-up in 2026, and "usage" means different things on each service (weekly pools, five-hour windows, credits). Before paying, open the pricing page on the day you buy.
Key idea
Try two tools on the same small task and keep the one you understand better.

From the people who built it

Straight from the source

The original idea, then the same idea taken apart, then how you use it today.

Andrej Karpathy

Founding member of OpenAI and former director of AI at Tesla, where he saw neural networks replace hand-written code in Autopilot.

Software Is Changing (Again), keynote at Y Combinator’s AI Startup School (June 2025)

1 · Understand: Karpathy calls LLMs a new kind of computer that you program in English, which he names Software 3.0. His advice is to build "partial autonomy" tools, where people choose how much to hand to the AI, rather than flashy fully automatic agents. In his words, "less Iron Man robots and more Iron Man suits."
2 · Break it down: Software 1.0 is code people write. Software 2.0 is the learned weights of a neural network. Software 3.0 is prompts in plain language that program an LLM.
The autonomy slider: in Cursor, for example, you can go from Tab completion, to editing a selected chunk, to letting the agent change the whole project. You decide how far to slide.
The AI generates and the human verifies, so the faster you can check its work (clear diffs, screenshots, tests), the faster you go.
Keep the AI on a leash: he works in small chunks, because a huge AI change is slow to check and can hide bugs.
3 · Use it: Judge each tool by two questions: how easily can I slide between small help and big tasks, and how quickly can I check what it did? Whichever tool you choose, keep tasks small enough that you can review every change.

5 terms

Words to know

Coding agent: An AI that takes a task, reads and edits files, and runs commands in a loop until the task is done.
AI code editor: A code editor with AI built in, like Cursor, or VS Code with Copilot.
App builder: A browser tool like Lovable or Bolt where you describe an app in chat and it builds and hosts it.
AGENTS.md: A shared Markdown instruction file that Codex, Cursor and Claude Code can all read.
Usage limit: How much AI work a plan includes before you wait for a reset or pay more. Each company measures it differently.

1 video · plays here

Watch

Hand-picked from the people who explain this best. The video loads only when you press play.

Claude Code vs Codex vs Cursor (an honest comparison)
Theo - t3․gg · 38 min
One experienced developer’s strong opinions after using all three daily. Recorded in May 2026, before Cursor’s plan changes, so use it for workflow ideas, not prices.

Hands-on · 10 minutes or less

Try it

Run the same task in two tools

1Pick two tools you can use for free or already pay for, for example Cursor Free and Codex in ChatGPT, or Copilot Free and Claude Code.
2Make two empty folders, one per tool.
3Give both the same prompt: "Build an expense splitter in one index.html file. I enter names and amounts paid, and it shows who owes whom. Explain how you tested it."
4Test both in your browser with three people: A paid 900, B paid 300, C paid 0. Each should owe 400, so C pays A 400 and B pays A 100.
5Note which one was easier to follow and review. That is your tool for now.

GitHub Copilot plans (includes the free tier)

3 questions · instant feedback

Check yourself

Pick an answer to see if you are right, and why.

Question 1 of 3

You already pay for ChatGPT Plus and want a coding agent. Which is the most natural first try?

Question 2 of 3

A student wants to learn coding and see every line the AI writes, on a small budget. What fits best?

Question 3 of 3

Which statement is fair?

Lesson 32 of 36

Done? Keep going

Mark this lesson complete to keep your place. Progress is saved on this device, no login.

Up next · Lesson 33 · 12 minPrompting vs RAG vs fine-tuningStart with prompting, always. Add RAG (retrieval-augmented generation) when the model is missing knowledge, such as your private documents or facts that change often. Use fine-tuning only when the model is missing behaviour, like a strict format or style, that good prompts and examples cannot fix. Let a small test set decide, not hype.← Previous: OpenAI Codex for beginners

Questions people ask

Which AI coding tool is best for beginners?

The one that fits what you already have. If you pay for ChatGPT, try Codex. If you pay for Claude, try Claude Code. If you want to learn to code, an editor like Cursor or VS Code with Copilot keeps the code in front of you. If you never want to see code, start with an app builder.

Is Cursor better than Claude Code?

Neither is better for everyone. Cursor is a full AI code editor; Claude Code is an agent you use from the terminal, your editor, a desktop app or the web. You can even use both, because Claude Code has an extension that installs in Cursor. Try both on the same small task.

Is GitHub Copilot free for students?

GitHub’s plans page says verified students can access the GitHub Copilot Student plan through GitHub Education. Anyone can also use Copilot Free, which includes 2,000 code completions a month as of Nov 2026.

Claude Code vs Cursor vs Codex: which AI coding tool should you use?

First, the three shapes of AI coding tools

The big three, side by side

GitHub Copilot and the app builders

A simple decision guide

What we could not verify, and what changes fast