Skip to main content
/ Development

Cursor AI 2.4: Complete Guide to the Newest Features

Sacha Roussakis-NotterSacha Roussakis-Notter
18 min read
CursorCursor
ClaudeClaude
TypeScript
VS Code
Share

Discover Cursor 2.4's groundbreaking features including Subagents, Agent Skills, Image Generation, and Cursor Blame. Learn how to use each feature with examples, keyboard shortcuts, and best practices for Australian developers.

Cursor Just Became the Fastest-Growing SaaS Ever

Cursor has reached a $29.3 billion valuation and crossed $1 billion ARR — making it the fastest SaaS application to ever reach $100 million ARR (just 12 months). With version 2.4 released on January 22, 2026, Cursor introduces game-changing features that redefine what an AI code editor can do.

Cursor by the Numbers (2026)
Valuation
$29.3BNov 2025
Annual Revenue
$1B+ARR
Daily Active Users
1M++200%
Fortune 500 Adoption
50%+
4 metricsbuun.group

Whether you're a Brisbane startup or an enterprise team, Cursor 2.4 offers features that will transform how you write code. Let's dive into everything new.

What's New in Cursor 2.4

Version 2.4 (January 22, 2026) introduces the most significant update since Cursor 2.0. Here's the complete breakdown.

Changelog
3 releases
Added

Subagents: Independent parallel agents with custom prompts and tools

Added

Agent Skills: Domain-specific workflows via SKILL.md files

Added

Image Generation: Create images from text descriptions

Added

Cursor Blame (Enterprise): AI attribution for code changes

Added

Agent Clarification: Agents ask questions while continuing work

Performance

Browser navigation 10x faster

Performance

Hook commands start 40x faster

Keep a Changelog formatbuun.group

Subagents: Parallel AI Agents

The headline feature of Cursor 2.4 is Subagents — independent agents that run in parallel with their own context, prompts, and tool access.

How Subagents Work

Unlike previous versions where you had a single Agent thread, Subagents let you spawn multiple AI workers that operate independently:

flowchart

Main Agent

Subagent: Research

Subagent: Terminal

Subagent: Parallel Work

Explore codebase

Find patterns

Run tests

Execute commands

Edit files

Create components

Ctrl+scroll to zoom • Drag to pan37%

Default Subagents

Cursor 2.4 ships with three pre-configured subagents:

SubagentPurposeTools Available
ResearchCodebase explorationRead, Search, Grep
TerminalCommand executionBash, Shell
ParallelIndependent work streamsAll tools

Custom Subagent Configuration

You can create custom subagents for specific workflows:

json
1{
2 "subagents": {
3 "test-runner": {
4 "prompt": "Run tests and report failures",
5 "tools": ["Bash"],
6 "model": "claude-sonnet-4.5"
7 },
8 "documenter": {
9 "prompt": "Generate documentation for code changes",
10 "tools": ["Read", "Write"],
11 "model": "gpt-4o"
12 }
13 }
14}

Agent Skills with SKILL.md

Agent Skills let you define domain-specific knowledge and procedures that Cursor can invoke dynamically.

Creating a Skill File

Create a SKILL.md file in your .cursor/ directory:

markdown
1# Deploy to Production
2
3## Context
4This skill handles production deployments for our Cloudflare Workers application.
5
6## Prerequisites
7- Wrangler CLI installed
8- Valid Cloudflare API token
9- All tests passing
10
11## Procedure
121. Run the test suite: 'npm run test'
132. Build the production bundle: 'npm run build'
143. Deploy to Cloudflare: 'wrangler deploy --env production'
154. Verify deployment health: 'curl https://api.example.com/health'
16
17## Rollback
18If deployment fails, run: 'wrangler rollback --env production'

Skills in Action

When you ask Cursor to "deploy to production", it automatically discovers and follows your SKILL.md procedure:

Agent Using Deploy Skill
$npm run test
All 47 tests passed
$npm run build
Build complete in 4.2s
$wrangler deploy --env production
Deployed to production
$curl https://api.example.com/health
{ status: ok }
4 commandsbuun.group

Image Generation

Cursor 2.4 adds inline image generation — create images directly from text descriptions using Google Nano Banana Pro.

How to Generate Images

Simply describe what you need in Agent mode:

Generate a hero image for our landing page showing a futuristic Brisbane skyline with tech elements

The agent creates the image and saves it to your assets/ folder, displaying it inline in the conversation.

Use Cases

  • Hero images for landing pages
  • Diagrams and architecture visuals
  • Icons and UI assets
  • Placeholder images during prototyping
  • Documentation graphics

Cursor Blame (Enterprise)

For enterprise teams, Cursor Blame extends Git blame with AI attribution — tracking exactly how code was created.

What Cursor Blame Tracks

Attribution TypeDescription
Tab CompletionsCode accepted from autocomplete
Agent RunsMulti-line changes from Agent mode
Human EditsManual developer modifications
Chat SuggestionsCode copied from Chat responses

Why It Matters

  • Audit trails for compliance and security reviews
  • Productivity metrics showing AI vs human contribution
  • Code quality analysis comparing AI-generated vs human code
  • Training insights for improving AI suggestions

Agent Clarification Questions

Agents can now ask clarifying questions while continuing work in parallel. No more blocking waits.

Before vs After

Agent Clarification Flowtext
44
1-Agent: I need to know the database type before proceeding.
2-[Waits for response]
3-[No work happens]
4+Agent: I'll start implementing. Quick question: PostgreSQL or MySQL?
5+[Continues building API structure]
6+[Creates models]
7
8 User: PostgreSQL
9
10-Agent: [Now starts working]
11+Agent: [Incorporates answer, adjusts connection code]
7 → 7 linesbuun.group

This parallel approach significantly speeds up development workflows.

Core Features Deep Dive

Cursor's power comes from understanding when to use each mode. Here's the complete decision guide.

Feature Progression

flowchart

Tab Autocomplete

Chat

Composer

Agent Mode

Single line/block

Q and A with context

Multi-file edits

Full autonomy

Ctrl+scroll to zoom • Drag to pan45%

When to Use Each Mode

Choosing the Right Mode
Use for single-line completions, repetitive edits, and small utility functions. Accept with Tab, reject with Esc.
5 optionsbuun.group

Keyboard Shortcuts Cheat Sheet

Master these shortcuts to 10x your Cursor productivity.

Cursor Essential Shortcuts

Mac shortcuts shown. Replace Cmd with Ctrl on Windows/Linux.

Accept autocompletecommon
Tab
Reject suggestioncommon
Esc
Partial accept (word)tip
Cmd + Right Arrow
Inline editcommon
Cmd + K
Open Chatcommon
Cmd + L
Open Composercommon
Cmd + I
Full-screen Composeradvanced
Cmd + Shift + I
Add selection to Chattip
Cmd + Shift + L
Reference filecommon
@filename
Reference foldercommon
@src/components
Search codebasecommon
@codebase query
External docstip
@Docs React
Web searchtip
@Web latest API
Git historyadvanced
@Git recent commits
Command palettecommon
Cmd + Shift + P
Settingscommon
Cmd + ,
Toggle sidebarcommon
Cmd + B
Toggle terminalcommon
Cmd + backtick
Switch layoutstip
Cmd + Option + Tab
3 sections • 19 itemsbuun.group

AI Model Selection Guide

Cursor supports multiple AI models. Here's when to use each.

flowchart

Yes

No

Quick autocomplete

General coding

Complex architecture

Long context needed

Stuck in loops

Select Model

Privacy Required?

DeepSeek or Local Models

Task Type?

cursor-small

Claude Sonnet 4.5

Claude Opus 4.5

Gemini 2.5 Pro

o3 Reasoning

Fast and cheap

Best daily driver

Critical logic

1M token context

Deep reasoning

Ctrl+scroll to zoom • Drag to pan37%

Model Comparison

ModelBest ForContextSpeed
cursor-smallAutocomplete, boilerplateSmallVery Fast
Claude Sonnet 4.5Daily coding, most tasks200KFast
Claude Opus 4.5Architecture, critical logic200KMedium
GPT-4oGeneral tasks, multimodal128KFast
Gemini 2.5 ProLong files, large context1MMedium
DeepSeekPrivacy-sensitive codeVariesMedium
o3Breaking out of loopsVariesSlow

Cursor vs GitHub Copilot vs Windsurf

How does Cursor 2.4 compare to the competition?

FeatureCursorGitHub CopilotWindsurf
BaseVS Code forkExtensionVS Code fork
Multi-file editingExcellentLimitedStrong
Agent modeFull autonomyPreviewCascade
Parallel agentsUp to 8NoNo
Codebase indexingSemantic RAGBasicSemantic
Model flexibilityClaude, GPT, Gemini, localGPT-4o, ClaudeProprietary
Background agentsCloud, 99.9% uptimeNoNo
Pricing (Pro)$20/mo$10/mo$15/mo
EnterpriseSSO, audit, blameExtensiveGrowing

Pricing and Plans

pricing

Cursor Pricing

Hobby

Free/forever
  • +Limited Tab completions
  • +50 slow AI requests
  • +Basic features

Pro

$20/per month
  • +Unlimited Tab completions
  • +$20 credit pool for premium models
  • +500 fast agent requests
  • +Background Agents
  • +Max context windows

Pro+

$60/per month
  • +Everything in Pro
  • +3x usage credits
  • +Priority support

Ultra

$200/per month
  • +Everything in Pro+
  • +20x usage credits
  • +Early access to features
  • +Elite model access
  • *Business/Teams: $40/user/month with SSO and admin controls
  • *Enterprise: Custom pricing with pooled usage and audit logs
  • *Prices in USD. Annual billing available with discount.
Last updated: January 2026buun.group

Best Practices for Australian Developers

Brisbane and the South East Queensland region are emerging as a global tech hub. Here's how local developers are leveraging Cursor.

Queensland AI Hub Resources

The Queensland AI Hub operates nine regional chapters across the state, providing resources for AI tool adoption:

  • Brisbane Chapter: Community engagement, responsible AI
  • Gold Coast Chapter: AI workforce solutions
  • Sunshine Coast Chapter: AI literacy programs

Australian AI Adoption Stats

AI Adoption in Australia
SMEs Adopting AI
40%Q1 2025
Daily AI Users
26%of workers
Time Saved
4+ hrsweekly
AI Skills Premium
56%salary boost
4 metricsbuun.group

Local Meetups and Events

  • Brisbane AI Developers Meetup: 5,000+ members
  • GDG Brisbane: DevFest and monthly events
  • River City Labs: Innovation hub events

Getting Started with Cursor 2.4

Quick Start Guide
1 / 6

1Download Cursor

Get the latest version from cursor.com. Available for macOS, Windows, and Linux.

code
# Visit cursor.com/download
Step 1 of 6
6 stepsbuun.group

Project Rules Configuration

The modern way to configure Cursor AI behavior is with .mdc files in .cursor/rules/.

Example Rule File

markdown
1---
2description: TypeScript coding standards for this project
3globs: ["**/*.ts", "**/*.tsx"]
4alwaysApply: false
5---
6
7# TypeScript Standards
8
9## Code Style
10- Use functional components with hooks
11- Prefer const over let
12- Use explicit return types for functions
13- Use interface over type for object shapes
14
15## Naming Conventions
16- Components: PascalCase
17- Functions: camelCase
18- Constants: UPPER_SNAKE_CASE
19- Files: kebab-case
20
21## Error Handling
22- Always handle Promise rejections
23- Use try-catch for async operations
24- Provide meaningful error messages

Rules Priority Order

  1. Local (manual): Included with @ruleName
  2. Auto Attached: Matched by glob patterns
  3. Agent Requested: AI chooses if needed
  4. Always: Included in all contexts

Security Best Practices

Essential Security Settings

SettingRecommendationWhy
Privacy ModeEnable for proprietary codeZero data retention
.cursorignoreAdd .env, credentialsPrevent leaking secrets
Code ReviewHuman review all AI codeCatch vulnerabilities
Terminal ApprovalKeep enabledPrevent unintended commands

Privacy Mode Details

  • Over 50% of Cursor users have Privacy Mode enabled
  • Zero retention agreements with OpenAI, Anthropic, Google
  • SOC 2 Type II certified
  • GDPR and CCPA compliant

Frequently Asked Questions

Common Questions
8 items

For professional developers writing code daily, the productivity gains typically justify the cost. Users report 1.5-2x productivity increases. The free tier lets you try before committing.

Yes, Cursor is a fork of VS Code and supports all VS Code extensions. You can import your entire setup with one click.

Enable Privacy Mode to ensure zero data retention by AI providers. Cursor has zero-retention agreements with OpenAI, Anthropic, and Google.

Chat is conversational Q&A with your codebase. Agent mode is fully autonomous — it can edit files, run commands, create new files, and iterate until the task is complete.

Download Cursor, import your VS Code settings, and start using it. Your Copilot subscription is separate. Many developers use both during transition.

Yes, Cursor supports all languages through its LLM foundation. It works with Python, JavaScript, TypeScript, Go, Rust, Java, C++, and every other major language.

Subagents are independent AI agents that run in parallel with their own context and tools. They handle specialized tasks like research, testing, or documentation while the main agent coordinates.

Cursor requires internet for AI features. You can use it as a code editor offline, but AI assistance needs connectivity.

0 of 8 expandedbuun.group

Conclusion

Cursor 2.4 represents a massive leap in AI-assisted development. With Subagents for parallel work, Agent Skills for domain-specific workflows, Image Generation for visual assets, and Cursor Blame for enterprise auditability, it's clear why Cursor has become the fastest-growing SaaS application in history.

For Australian developers — especially those in Brisbane's thriving tech ecosystem — Cursor offers a competitive edge. The combination of powerful AI assistance, privacy-first options, and enterprise-grade features makes it an essential tool for modern development.

We're seeing an increase of over 25% in PR volume and over 100% in average PR size. Together, that means we're shipping about 50% more code." — Cursor Customer

The future of coding is agentic, and Cursor 2.4 is leading the way.

Ready to supercharge your development workflow?

Topics

Cursor AI 2.4Cursor AI features 2026Cursor SubagentsCursor Agent SkillsAI code editorCursor vs CopilotCursor changelogAI coding assistant

Share this post

Share

Comments

Sign in to join the conversation

Login

No comments yet. Be the first to share your thoughts!

Found an issue with this article?

/ Let's Talk

Want to work with us?

Whether you need help with architecture, development, or technical consulting, our team is here to help bring your vision to life.