Image source: Generated by Unbounded AI

The development of generative AI in 2024 is crazy, with large model wars spreading to various tracks, and vertical application crazes waxing and waning. Especially in the field of AI programming, algorithm progress has been rapid, many new unicorns have been born, investors have poured in intensively, and billion-level financing has been announced from the beginning of the year to the end of the year.

But if nominated for the annual AI programming dark horse, Cursor, whose valuation has increased by 550% in 4 months, is definitely the most popular.

Just in the past six months, it has set off a hurricane of change in the programming field by making it easier to write code, converting a large number of developers into its die-hard fans. Engineers and researchers from international technology giants and top scientific research laboratories to "top" generative AI companies such as OpenAI, Midjourney, and Perplexity are all paying to use this tool.

How popular is Cursor? Basically, everyone on the Internet has praised it, and everyone who has used it says it smells really good.

Cloudflare Vice President Ricky Robinett posted a video showing his 8-year-old daughter building a chatbot using Cursor AI in just 45 minutes. The "Kitten Fill Light" app developed in just one hour using Cursor quickly reached the top 20 on Apple's iOS app list. Andrej Karpathy, co-founder of OpenAI and former Tesla AI director, also cheered for this rising star of AI programming, praising that Cursor’s experience has surpassed GitHub Copilot.

The best thing is that its monthly price of US$20 is much higher than GitHub Copilot, attracting a large number of subscribers who are willing to pay for it, which shows how much it hits the pain points and joys of programmers.

▲Cursor official website (Source: Cursor)

So how useful is Cursor?

To put it simply, it changes the way of learning programming, making typing code line by line a thing of the past, and turning programming into something that can be accomplished by communicating in natural language. You come up with ideas, and AI is responsible for helping you implement them. Backed by large third-party models such as Claude 3.5 Sonnet and GPT-4o, Cursor AI can automatically generate code, check for errors and fix them, and it can also instantly answer questions when you encounter programming problems.

Not only is it friendly to beginners, it is also a time-saving magic weapon for professional software developers to significantly speed up their workflow.

Its investors have gathered a group of celebrities, including Google AI leader Jeff Dean, OpenAI co-founder John Schulman, Noam Brown, the father of poker AI, and former GitHub CEO Nat Friedman, Stripe co-founder Patrick Collison, Dropbox co-founder Arash Ferdowsi and more. The latest valuation reached US$2.6 billion (approximately RMB 19 billion).

The start-up company behind this product that has accumulated a huge reputation has only 12 engineers and researchers as of August 2024, and it has been established for less than two years.

How did they do it?

Let us review the birth of an annual hot-selling AI native product.

01. "Graduated" from OpenAI, the valuation increased by 550% in 4 months, ‍The rise of an AI code editor

p>

In 2022, MIT students Michael Truell, Sualeh Asif, Arvid Lunnemark, and Aman Sanger founded a company called Anysphere.

▲ A group photo of the four co-founders of Anysphere (Source: Forbes)

The four young people have an ambitious ambition: "Create a magical tool that one day can Write all the software in the world."

Their ultimate goal is to take the trouble out of building and changing software.

In its second year of entrepreneurship, Anysphere graduated from the OpenAI accelerator program and completed an $11 million seed round led by the OpenAI Startup Fund.

By the second half of 2024, the capital transaction process will suddenly accelerate. Anysphere first announced the completion of US$400 million in Series A financing in August, and in November announced the acquisition of Supermaven, the company behind the AI ​​code model Babble. Then in December, it was revealed by foreign media that it had completed US$100 million in new financing. The company’s valuation increased from August to of US$400 million jumped to US$2.6 billion (approximately RMB 19 billion), a full increase of 550%.

One of the possible reasons for the surge in valuation is linked to its rapidly growing sales.

The strong demand for AI programming has been transformed into a strong willingness to pay. According to TechCrunch, Anysphere’s annualized recurring revenue (ARR) increased more than 10 times from April to October, reaching $48 million. According to the forecast of the research institution Sacra at that time, Anysphere's ARR is expected to exceed US$65 million in November, a year-on-year increase of 64 times, and its monthly revenue has reached US$4 million (approximately RMB 29.2 million).

▲Anysphere ARR forecast (Source: Sacra)

According to The Information, Anthropic’s software development and code generation business has experienced a 10-fold annualized revenue increase in 3 months. . Microsoft previously disclosed GitHub Copi in its financial report released in July.Lot's annualized recurring revenue (ARR) reached $300 million, accounting for 40% of GitHub's overall revenue growth. In comparison, Cursor’s growth rate is astonishing.

AI programming tools generally adopt a subscription model, and Cursor is no exception.

However, its pricing is not user-friendly. There is a two-week free trial period. After the trial period, users can choose to purchase the professional version for US$20 per month (approximately RMB 146), or purchase the professional version per month. Purchase a business subscription designed for large teams and organizations for $40 (approximately RMB 292).

Although Cursor charges $10 per month (approximately RMB 73) higher than GitHub Copilot, it still attracts the use of more than 30,000 customers around the world, including OpenAI, Midjourney, Perplexity, Replicate, Shopify, Instacart and other well-known companies.

Feeling the threat of competition, Microsoft quickly launched a free version of GitHub Copilot. Free users can enjoy 2,000 code completions per month. Microsoft CEO Satya Nadella also personally tweeted in support of Copilot.

In 2024, when Cursor was booming, the combination of AI and programming became a hot spot for venture capital. In addition to GitHub Copilot, the mainstream choice for developers, many other startups on the market such as Augment, Codeium, Poolside, Magic AI, Bolt.new, etc. are also emerging in the fierce competition and have received large amounts of financing.

How does Cursor stand out?

02. Challenge Microsoft and use AI to change the way software is built

The core functions of AI programming can be divided into two major categories: automatic code generation and code completion/suggestion.

Cursor is a branch of VS Code. As an independent local IDE, Cursor has greater control over the editor's user interface, allowing for stronger AI integration.

Unlike GitHub Copilot, which focuses on providing code suggestions and completions based on context, Cursor is designed to provide intelligent code tips based on the entire code base. Therefore, Cursor can predict developer needs in real time. Just press the Tab key, and the AI ​​editor can predict and automatically complete the code, and recommend the next possible code edit based on the context.

▲Analysis of the project code base (Source: Cursor)

Cursor’s vision is to create a code that is more helpful and fun than ever in the next few years. Editor.

VSCode used to be the preferred IDE for most developers. But the new AI era calls for a new tool that interacts more naturally and integrates seamlessly into the workflow. GitHub Copilot serves as an enhancement tool for existing IDEs, and Cursor builds a more specialized IDE that deeply integrates AI. Standing on the shoulders of cutting-edge third-party large models, the Cursor team focuses on polishing the product on optimizing the development of user experience and model interaction methods, and how to make the model give better answers.

In 2024, more and more developers will begin to turn to Cursor. Compared with traditional IDEs, this AI programming tool focuses more on the seamless integration of AI and development processes. Its core advantage is that it not only has code completion and conversational programming functions, but also introduces the "Composer" function, which allows users to put forward requirements for the entire project through natural language and perform cross-file editing. After acquiring Supermaven, Cursor has further enhanced its technical strength in processing complex code and improving context awareness.

1. Agent mode and Yolo mode: improving automated task execution

On November 24, Cursor released version 0.43, introducing the "Agent mode". This mode supports interacting with the Agent via Composer, selecting context and executing terminal commands. Agent mode not only supports semantic search, but also helps developers automatically complete tasks such as code completion, error repair, and multi-file operations. Through semantic search, Agent mode can intelligently recommend solutions to further improve work efficiency.

On December 17, Cursor released the latest version 0.44, and introduced the Yolo mode based on the Agent mode. The goal of Yolo mode is to allow agents to execute commands more intelligently and support parallel processing of multiple tasks. For example, developers can instruct the agent to make code modifications in multiple files simultaneously, or to perform complex terminal operations without human intervention. With the help of GPT-4 model, Yolo mode enables Agent to understand commands more accurately and execute them automatically.

2. Intelligent code completion and automatic generation

One of the biggest highlights of Cursor is its deep learning capabilities based on advanced AI models such as GPT-4 and Claude 3.5 Sonnet, which can enable real-time Anticipate developer needs and generate corresponding code. This allows Cursor to not only automatically complete, fix errors and reconstruct code, but also generate new code snippets, greatly improving development efficiency.

Compared with traditional VSCode or GitHub Copilot, Cursor is better at contextual understanding. It can intelligently predict developers' programming intentions, not only relying on code fragment matching, but also focusing on the logic of the overall project, truly realizing code intelligence from a global perspective.

Cursor’s Tab function is a powerful and intuitive AI-assisted programming tool, which can be called an enhanced version of auto-completion. This feature not only automatically completes code based on the cursor position, but also supports multi-line completion and code rewriting. Cursor prediction is another advanced feature of Cursor. It can predict the developer's next operation based on context, greatly improving development efficiency.

▲Tab key automatically completes the code (Source: Cursor)

3. Context awareness and full project support

Cursor breaks through the traditional IDE that is limited to the current File limitations enable analysis of the entire project. It understands project structure, coding style, and provides code suggestions based on a global perspective. @Codebase is a powerful feature of Cursor. It analyzes the entire project through vectorization technology and provides developers with comprehensive contextual support. It is especially suitable for large projects or scenarios that require cross-file analysis.

Through natural language query, developers can easily issue instructions, such as "modify all functions in userService" or "add unit tests for this module", and Cursor will automatically generate appropriate code snippets. Compared with GitHub Copilot, Cursor has advantages in global project understanding and cross-file coordination.

▲Edit with natural language (Source: Cursor)

4. Composer: Intelligent Coding Assistant

"Composer" is Cursor's intelligent coding assistant. Helping developers easily write and modify code. Users simply interact with Composer through natural language instructions, such as "Add login functionality" or "Optimize this function," and Composer will automatically generate the code and optimize it. This greatly simplifies the programming process, allowing developers to focus on higher-level design without the need to frequently switch contexts.

5. Cursor Chat: Interactive code search

Cursor Chat provides an interactive code search function, allowing developers to interact with the code base through conversations. Users can enter simple natural language questions and Cursor Chat will provide relevant code snippets or explanations based on the context of the project. Cursor Chat is unique in that it combines contextual references (@) and the “Apply” function for code modifications, which enables developers to directly apply code suggestions provided by AI to projects.

Although similar conversational programming tools already exist, Cursor Chat simplifies the code modification process through the "Apply" function, providing a more efficient solution than traditional code editors.

6. Privacy and Security

For developers, the privacy and security of the code are crucial. Cursor provides strong privacy guarantees in this regard, ensuring code through privacy modeIt is not stored on the server and data protection is enhanced with SOC 2 certification. All code processing occurs locally to ensure that developers’ data security and privacy are not violated.

In short, Cursor promotes the evolution of IDEs through its deep integration of AI functions, intelligent code completion and generation, semantic search, and intelligent task execution in Agent and Yolo modes. Especially in terms of full project support, context awareness and privacy protection, Cursor provides a more efficient and intelligent solution than traditional IDEs.

03.Copilot, Cursor and Bolt.new: Three development paths for AI programming tools

Globally, GitHub Copilot, Cursor, AI coding tools such as Augment and Codeium have been widely used in the market. Domestic manufacturers such as Alibaba Cloud, ByteDance, Huawei, etc. have also launched their own AI programming products, such as Alibaba's Tongyi Lingma and Baidu's Wenxin Kuaima. wait.

Twitter user auchenberg used data provided by research firm Sacra to create a chart comparing the annualized recurring revenue (ARR) of mainstream AI programming products on the current market. It can be seen that Cursor's ARR has jumped significantly from 2023 to 2024, but the overall volume is still not comparable to GitHub Copilot.

▲The ARR of mainstream AI programming products on the market (Source: Twitter user auchenberg, Sacra)

Among them, the ARR of GitHub Copilot will increase from US$115 million in 2023 to 2024 Nearly US$400 million, a growth rate of 281%. By November 2024, Bolt.new's ARR will reach US$20 million, and the platform's average daily ARR will increase by US$500,000, with good growth momentum.

GitHub Copilot, Cursor and Bolt.new respectively represent different development directions of AI programming tools. GitHub Copilot focuses on large-scale popularization and multi-platform integration, Cursor is an intelligent upgrade based on the existing IDE, and Bolt.new is exploring a new Web development model.

▲Different types of code editors (Source: Cursor)

GitHub Copilot, as a product of Microsoft, directly integrates the two major platforms of Visual Studio Code and GitHub, providing powerful Code completion function supports more than dozens of programming languages. Through deep learning model training, Copilot is able to understand the context and generate reasonable code snippets, greatly improving development efficiency.

Cursor focuses more on traditional IDE(such as VSCode), it can help developers improve coding efficiency by providing intelligent code completion, automatic reconstruction and other functions. Its success lies not only in its efficient auxiliary functions, but also in its good user experience and ease of operation. Users can get real-time advice and support while writing code, reducing the learning curve for junior programmers while also providing efficient tools for senior developers.

Different from the previous two local IDEs, Bolt.new pioneered browser-based low-code development, providing a more intuitive and low-code development experience. Relying on WebContainers technology, it can be used while browsing A micro-operating system runs in the processor and the Claude 3.5 Sonnet model is integrated. In this way, users can code directly in the browser without configuring a virtual machine or logging in to an account, which simplifies the development process and is especially suitable for beginners and rapid development needs.

Cursor improves development efficiency through a global project perspective and intelligent task execution, and Bolt.new lowers the technical threshold by simplifying the development process. The success of these products reflects developers' strong demand for intelligent tools. They are no longer satisfied with the single function of traditional IDEs, but focus more on tools that are closely integrated with business needs.

04. Conclusion: AI programming development has entered the fast lane

AI programming tools bring advantages but also come with challenges. Code quality and security become important issues. AI-generated code may contain errors or inconsistencies and still require human review. In addition, data privacy and intellectual property issues also need to be governed by compliance. In the future, how to balance technological innovation with the guarantee of code quality and security will be the key to the development of the industry.

What’s more interesting is that Cursor publishes a blog post every year describing a list of problems it hopes to solve. Issues listed in 2023 include better context, editing Copilot, active error checking, larger editing scope, etc. Issues listed in 2024 include next action prediction, perfect editing, optimal context, error detection and debugging.

Entering 2025, the AI ​​programming track is expected to develop even more rapidly. After all, the Cursor team has already set a flag in an interview: "The Cursor one year from now will make today's Cursor look outdated."