Huge monolithic React page blocks reuse
The page becomes composable and testable; two new features are added the same afternoon by reusing the extracted components.
Track 13 · Advanced
13.2.3
Cursor
VS Code reimagined with an AI agent that writes and runs your code
1Overview
A fork of VS Code where AI isn't bolted on as an extension — it's built into the editor itself, from inline suggestions to a full autonomous agent mode.
Cursor is a code editor built on VS Code — familiar if you've used VS Code, but AI is woven into every layer. It autocompletes as you type, answers questions about your own files in a chat sidebar, and in Agent mode reads your project, writes changes across multiple files, runs them, and iterates until it works. → University students get a full year of Cursor Pro free via cursor.com/students.
1.1Before you start
AI coding assistants— Reading AI coding assistants introduces collaborative code-base tools, so you'll grasp Cursor's role as an example.A repo you can push to— First, work in GitHub—store repositories, handle pull requests, Issues, and Actions, where AI agents already convert Issues into completed PRs.
1.2After this chapter you can
→Use Agent mode to read a project, write multi-file changes, and run them
→Ask the chat sidebar plain-English questions about your own files
→Pick the right frontier model per task — Claude, GPT, or Gemini
1.3Best for
Describing what you want in plain English instead of writing syntax from memory — "load this CSV, plot a heatmap, flag rows with >20% missing values" becomes a working script.
1.4Watch out
The free Hobby plan hits its request limits fast; heavy use needs the $20/mo Pro plan. Sends code to cloud AI providers — check institutional data policies for sensitive research data.
1.5Free vs paid
Hobby (free) hits its usage limits fast under real work; each paid tier mainly buys more headroom, not new features. Your code also goes to outside AI providers on any tier, since it's cloud-based.
2Lessons 7
2.1 Install Cursor and open its chat sidebar
Open Cursor and display the built‑in chat pane for coding assistance
- Visit cursor.com/download in a web browser and click the appropriate Download button for your operating system
- Run the installer you saved and follow the on‑screen prompts to finish installation
- Launch Cursor from your applications menu
- Open a folder with File → Open Folder… or create a new file
- Click the chat icon in the activity bar to open the chat pane
- You'll see The editor UI appears with a chat panel on the right side of the window
- Takeaway The AI‑enabled editor is ready for instant code assistance
- Check Which sequence of actions lets you install Cursor, open a project folder and then reveal the chat pane for AI help?
2.2 Explore your project with Cursor’s AI Agent
The Agent window that can answer questions about the files in a opened folder.
You will be able to ask the Agent to locate definitions, trace call paths, and list key modules in your codebase.
- Open a local repository folder in Cursor.
- Press Cmd+K to open the chat sidebar if it is not already visible.
- In the chat input type “@agent show me the entry point of this project”.
- Read the Agent’s response and verify that the file path it mentions exists in the Explorer panel.
- You'll see The Agent returns a file name with line numbers, and clicking the link opens that file at the indicated location.
- Takeaway AI agents can quickly surface structural information from large codebases without manual searching.
2.3 Activate a free year of Cursor Pro
Activate a free year of Cursor Pro using a verified student account
- In Cursor, click Sign Out to log out of any existing account
- Open a browser and go to cursor.com/students
- Enter your .edu email address in the field and press Submit
- Open the verification email you receive and click its Verify link
- Log back into Cursor with the newly created student credentials
- You'll see The dashboard shows a Pro badge and indicates that free usage credits are available for the year
- Takeaway Verifying student status unlocks premium AI features at no cost
- Check What steps do you follow to verify your .edu email and obtain the Pro badge inside Cursor?
2.4 Plan a new feature using Cursor’s Plan mode
Plan mode, where the Agent asks clarifying questions and produces a step‑by‑step implementation plan.
You will generate a concrete development plan for adding a “search” endpoint to an existing API.
- In the chat sidebar type “@agent I want to add a search endpoint that returns JSON results”.
- When the Agent asks follow‑up questions, answer them to clarify language (e.g., TypeScript) and location (e.g., src/api).
- Accept the generated plan when the Agent presents it.
- Copy the plan into a new Markdown file in the project for reference.
- You'll see The Agent outputs an ordered list of tasks with file names, function signatures, and test suggestions.
- Takeaway Plan mode turns high‑level ideas into actionable, scoped work items that can be handed to the agent for execution.
2.5 Refactor a React page into typed sub‑components
Transform a monolithic React file into separate typed components and a context provider while keeping the UI unchanged
- Open the project folder in Explorer
- Show the Agent sidebar by clicking the Agent icon
- Enter the refactoring request in the Prompt box and submit
- Examine the generated plan and click Approve
- Allow the Agent to create sub‑components, add typings, and extract a context provider
- Open Terminal, run
npm start, and confirm the UI looks unchanged
- You'll see The original file shrinks, new .tsx component files appear in the explorer, and the app renders exactly as before
- Takeaway AI agents can automate large‑scale refactoring into modular, typed code with minimal effort
- Check Which steps involve interacting with the Agent sidebar to request, approve, and let the AI generate new typed sub‑components for your React project?
2.6 Let the Agent implement the planned feature
Agent execution, where the AI writes code across multiple files based on a plan.
You will have a new endpoint file created, route registration updated, and a passing test added automatically by the Agent.
- Open the Markdown plan you saved in Lesson 2.
- Select all text of the plan and type “@agent execute this plan”.
- Confirm any preview the Agent shows before it writes files.
- When the Agent finishes, run the built‑in test command (Ctrl+Shift+
thennpm test`). - Observe that the new tests pass and the endpoint returns expected JSON.
- You'll see New source files appear in the Explorer, the test runner reports all green, and a console log shows the endpoint response.
- Takeaway Cursor’s agent can turn a written plan into real code changes and verify them with automated tests.
2.7 Add a custom skill to extend Cursor’s capabilities
A Skill, defined in the Docs under “Customize → Skills”, that lets you invoke reusable commands from the chat.
You will install a sample skill and use it to generate boilerplate for a React component with TypeScript types.
- Open the Docs sidebar (Cmd+K then type “skills”).
- Find the example skill named “component‑boilerplate” and click “Add”.
- Return to the chat sidebar and type “/component‑boilerplate Button.tsx”.
- Accept the generated file and open it to see a ready‑to‑use component skeleton.
- You'll see A new Button.tsx file appears with imports, props interface, and a functional component stub.
- Takeaway Skills let you package common patterns as one‑click actions, speeding up repetitive development tasks.
3You’ll know it worked 18 checkable outcomes in this chapter
- ✓The original page loads with the same UI, but its code is split into typed sub-components and a context provider
- ✓Discrepancies beyond $0.01 trigger an email to ar-team@company.com
- ✓Cursor shows a custom model entry named "GLM 5.2" and uses it for chat completions
- ✓The loop stops after the score reaches four or five, or after a set number of attempts
- ✓A suggestion appears inline while typing and is inserted after pressing Tab
- ✓Agent responses include references to the exact file names you mentioned
- ✓Billing page lists "Pro" as the active subscription and shows $20 remaining API credit
- ✓Account dashboard shows "Ultra" as active and $400 of remaining usage
18 outcomes in all — one per recipe below.
4FAQ, Tips & How-to 41
one problem, one solution, one action
Internal tools & ops1
Dashboards & analytics1
Slow dashboard caused by ORM N + 1 queries
Dashboard load time drops from 4.2 s to 0.6 s; the migration is production-ready with rollback included.
Research & data tools1
Manual spreadsheet formulas for billing reconciliation
A reconciliation step that previously required a 200-line manual spreadsheet formula is replaced by a reproducible, version-controlled script; discrepancies are caught the same day they arise.
Tracking team performance on Cursor helps prioritize development focus
Cursor API key swap for GLM 5.2
Configure Cursor to use GLM 5.2 by swapping the OpenAI key for a Z AI key and overriding the endpoint
Using a single coding environment ensures all models run under the same hardware and file-system constraints
Want to use MiniMax M3 in your code editor
You can use MiniMax M3 with Cursor by pointing to its API endpoint and registering a custom model name
Want UI clicks to run ImageKit transforms
Configure Cursor to call ImageKit's API without writing code
Need to speak your project specs
Use voice input to generate a detailed prompt for Claude
Need an architecture document
Ask Claude to produce a structured architecture document
Want longer chats without cutting off text
Increase the context size to keep longer conversations
Need a new GitHub repo that enforces commit rules automatically
Let Cursor manage GitHub integration and enforce commit policies
AI pipeline keeps failing
Use error messages as prompts for Claude to fix the pipeline
Code gets low review scores
See how Cursor can repeatedly pull a review score, fix code, and push until the target is reached
My code is scattered and I need the AI to see it
Opening a folder lets the AI see your codebase and generate files in the right place
You retain control by inspecting the suggested modifications before they become part of your project
You can type faster by letting Cursor suggest whole lines or functions as you write code
Press Tab to accept an autocomplete suggestion
Pressing Tab confirms the autocomplete suggestion, inserting it instantly
Use the chat to understand code snippets and request concise modifications rather than large rewrites
Need to edit many files at once
Activating Agent mode lets the AI read your project and write changes across many files automatically
Need a script to clean a gene CSV and plot a clustered heatmap
Providing a clear English description to the agent yields a ready-to-run script spread over the necessary files
Knowing your plan's request quota helps you allocate Agent runs to high-value features
Not sure which AI model fits your task
Enabling Auto lets Cursor automatically pick the most suitable model for each task
Choosing a more capable (heavier) model improves answer quality on difficult tasks
Different models have different API costs, so picking a model directly influences how fast you consume included usage
The agent can answer using actual project files instead of guessing
Project knowledge disappears after each chat
Shared-context cloud agents let multiple users work with the same project knowledge
Need a free AI assistant with limited requests
You can start using Cursor without paying by selecting the Hobby tier
- **Hobby** — free tier to start.
- **Pro** — $20/mo (Pro+ $60, Ultra $200).
- **Teams** — $40/user/mo; **Enterprise** — custom.
Cursor Pro plan
Upgrading to Pro gives you $20 of monthly API usage and higher limits for Auto and Composer
Pro+ offers more generous monthly credits for heavier users at a higher price
Cursor Ultra tier
Ultra delivers the highest monthly credit for power users or teams needing large API budgets
Want a single bill for all users
Teams lets multiple users share billing, a marketplace, and shared-context agents
Want a full year of premium features for free
Eligible students receive a full year of Pro features at no cost
What is Cursor and what is it for?
Cursor is a code editor built on VS Code with AI woven into every layer, made by Anysphere. If you have ever used VS Code the interface is immediately familiar, but as you type it autocompletes whole lines and functions, a chat sidebar answers questions about your own files in plain English, and Agent mode reads your project, writes changes across multiple files, runs them, and iterates until they work. It is for writing and building software without needing syntax from memory: you describe what you want, review the proposed change, and accept it. It understands pandas, NumPy, matplotlib, and Jupyter notebooks natively, so it suits data and analysis work as well as app building.
Can I keep my VS Code extensions and settings in Cursor?
Yes. Cursor is a fork of VS Code, so the layout, shortcuts, and extension ecosystem are the same, and the apps you already know carry over. That includes the wider VS Code extension marketplace and native Jupyter notebook support, so you can keep using your familiar tools alongside Cursor's AI features. On paid plans you can also extend the agent with MCPs, skills, and hooks: MCPs connect the agent to external tools and data sources, while skills and hooks shape how it works. The practical upshot is that you do not have to abandon your existing editor setup to adopt Cursor; you layer the AI on top of it.
What is Tab completion and how do I use it?
Tab is Cursor's AI autocomplete. As you type it suggests code ahead of your cursor as grayed-out text, based on your recent edits, the surrounding code, and any linter errors. Press Tab to accept a suggestion, or keep typing to ignore it. It is more capable than ordinary autocomplete: it can change multiple lines at once, add missing import statements, and suggest coordinated edits across related code. After you accept a suggestion, pressing Tab again can predict your next edit location and jump you there. You can snooze or configure it from the Tab status indicator in the bottom-right corner, under Cursor Settings then Tab.
What is Agent mode and when should I use it?
Agent mode is Cursor's autonomous mode in the chat sidebar. Instead of giving you a single answer, the agent reads your project, plans the steps, writes changes across multiple files, can run terminal commands with your approval, and iterates on errors until the task works. You use it for whole features rather than tiny edits, for example loading a CSV, cleaning it, and producing a plotted heatmap from one description. Always review the diff it proposes before accepting. Keep the request specific and small-scoped, naming the file, columns, and exact rules, because vague instructions let the agent drift while precise ones keep it on track.
Which AI models can Cursor use, and does the choice matter?
Cursor is not tied to a single model. The model picker in the chat and Agent bar gives you access to the major frontier models from Anthropic, OpenAI, and Google, switchable per task, and an Auto setting that lets Cursor pick for you. The choice matters for both quality and cost: reach for a heavier model for tricky multi-file reasoning and a faster one for routine edits. Because different models have different underlying API costs, your model choice is also a cost lever, governing how quickly your plan's included usage is consumed. Frontier-model access is part of the paid plans rather than the free Hobby tier.
How do I fix an error in my code with Cursor?
Run your code, and when it breaks, copy the error message or traceback and paste it straight into the chat. Cursor diagnoses the cause from your actual code rather than a generic guess and explains it in plain English. In Agent mode it can go further: it can run the code itself, see the failure, and iterate, installing a missing package, fixing a path, or correcting logic and re-running until the script succeeds. Review each fix before accepting it. To spend fewer Agent requests, batch the error and the goal into one message, for example pasting the traceback and adding make the script run end to end.
Want a coding environment ready to go
Get Cursor installed on your machine and ready to use
Student with a .edu email wants a free year of the AI coding tool
Students can receive a free one-year subscription to Cursor Pro by verifying their university email address
The same set on /recipes, filtered by tool and role.
5Videos 3
How to Install and Use Cursor AI Editor (2026)
The quick setup video. Watch this to get Cursor running before the longer tutorials.
Cursor: coding agents tutorial (2026)
From a widely respected developer (ex-Vercel). Watch to learn the agent workflow, not just the buttons.
Cursor 2.0 - Full Tutorial for Beginners
A thorough, no-fluff intro from a top coding educator. Start here if you've never opened Cursor.
6FAQ 6
What is Cursor and what is it for?
Cursor is a code editor built on VS Code with AI woven into every layer, made by Anysphere. If you have ever used VS Code the interface is immediately familiar, but as you type it autocompletes whole lines and functions, a chat sidebar answers questions about your own files in plain English, and Agent mode reads your project, writes changes across multiple files, runs them, and iterates until they work. It is for writing and building software without needing syntax from memory: you describe what you want, review the proposed change, and accept it. It understands pandas, NumPy, matplotlib, and Jupyter notebooks natively, so it suits data and analysis work as well as app building.
Can I keep my VS Code extensions and settings in Cursor?
Yes. Cursor is a fork of VS Code, so the layout, shortcuts, and extension ecosystem are the same, and the apps you already know carry over. That includes the wider VS Code extension marketplace and native Jupyter notebook support, so you can keep using your familiar tools alongside Cursor's AI features. On paid plans you can also extend the agent with MCPs, skills, and hooks: MCPs connect the agent to external tools and data sources, while skills and hooks shape how it works. The practical upshot is that you do not have to abandon your existing editor setup to adopt Cursor; you layer the AI on top of it.
What is Tab completion and how do I use it?
Tab is Cursor's AI autocomplete. As you type it suggests code ahead of your cursor as grayed-out text, based on your recent edits, the surrounding code, and any linter errors. Press Tab to accept a suggestion, or keep typing to ignore it. It is more capable than ordinary autocomplete: it can change multiple lines at once, add missing import statements, and suggest coordinated edits across related code. After you accept a suggestion, pressing Tab again can predict your next edit location and jump you there. You can snooze or configure it from the Tab status indicator in the bottom-right corner, under Cursor Settings then Tab.
What is Agent mode and when should I use it?
Agent mode is Cursor's autonomous mode in the chat sidebar. Instead of giving you a single answer, the agent reads your project, plans the steps, writes changes across multiple files, can run terminal commands with your approval, and iterates on errors until the task works. You use it for whole features rather than tiny edits, for example loading a CSV, cleaning it, and producing a plotted heatmap from one description. Always review the diff it proposes before accepting. Keep the request specific and small-scoped, naming the file, columns, and exact rules, because vague instructions let the agent drift while precise ones keep it on track.
Which AI models can Cursor use, and does the choice matter?
Cursor is not tied to a single model. The model picker in the chat and Agent bar gives you access to the major frontier models from Anthropic, OpenAI, and Google, switchable per task, and an Auto setting that lets Cursor pick for you. The choice matters for both quality and cost: reach for a heavier model for tricky multi-file reasoning and a faster one for routine edits. Because different models have different underlying API costs, your model choice is also a cost lever, governing how quickly your plan's included usage is consumed. Frontier-model access is part of the paid plans rather than the free Hobby tier.
How do I fix an error in my code with Cursor?
Run your code, and when it breaks, copy the error message or traceback and paste it straight into the chat. Cursor diagnoses the cause from your actual code rather than a generic guess and explains it in plain English. In Agent mode it can go further: it can run the code itself, see the failure, and iterate, installing a missing package, fixing a path, or correcting logic and re-running until the script succeeds. Review each fix before accepting it. To spend fewer Agent requests, batch the error and the goal into one message, for example pasting the traceback and adding make the script run end to end.
7Glossary 8 terms
Show the 8 terms
Cursor
sns.clustermap- A function from the seaborn visualisation library that draws a heatmap with rows and columns automatically reordered so that similar values cluster together.
utils.py- A Python file named by convention to hold small helper functions shared across multiple scripts in the same project, keeping the main files shorter and easier to read.
Agent mode- A mode where Cursor autonomously reads your project and writes changes across multiple files in one session, rather than answering a single question.
Tab- Cursor's line-by-line code suggestion that appears as you type; press Tab to accept it or keep typing to ignore it.
model picker- The dropdown that lets you switch which model handles a request (Claude, GPT, Gemini, or Auto) on a per-task basis.
diff- The view Cursor shows of what the AI changed in a file; you review it and choose to accept or reject before it is saved.
ModuleNotFoundError- A Python error meaning a required package is not installed; in Agent mode Cursor can detect this and install the missing package automatically.
Auto- The model-picker option that lets Cursor choose the best model for each request automatically, useful for routine edits.
8See also
💬 Discuss this chapter
Ask, share, or report — over on the Heidelberg AI community forum.