On Friday, May 16, OpenAI announced the launch of the Codex research preview, which will be available to users of ChatGPT's Pro, Enterprise, and Team versions starting Friday, and users of the Plus and Edu versions will soon be able to use Codex's services.

The competition for AI programming tools is heating up. The rumored $3 billion acquisition of the programming artifact Windsurf has not yet been confirmed, and OpenAI has first configured ChatGPT with the strongest self-developed AI agent for writing code.

The AI programming market is rapidly expanding to billions of dollars. For investors, this marks the shift of AI application scenarios from "toys" to real productivity tools, which will reshape the value chain and investment landscape of the software industry.

Codex has been upgraded, and this time it's a little different

OpenAI launched a brand new Codex on Friday, and compared to the old version in 2021, it is not just a small change, but a comprehensive upgrade!

This new version of Codex is based on the latest codex-1 model, which is an optimized version of OpenAI's o3 model in software engineering.

To put it simply, it's more like a "trained" programmer's assistant who not only understands the technology, but also knows how to write code that conforms to the human style and even the company's PR standards.

What the hell is it capable of?

The new version of Codex can accomplish a variety of software development tasks, such as:

Write functional code

Answer questions about your project's code

Fix bugs automatically

Submit a pull request (Pull Request）

And each task runs in a separate cloud sandbox environment, pre-loaded with your codebase, like a dedicated laptop for the AI, so it can work without disturbing the main system.

Depending on the complexity of the task, the processing time can range from 1 minute to 30 minutes, and users can also see the progress in real-time – just as clear and transparent as watching a food delivery.

The best thing about this version of Codex is that it doesn't rely on rote memorization to generate code, but through reinforcement learning, it "practices" in a variety of real-world coding tasks, and finally learns to write high-quality, human-like code.

Not only does it execute instructions with precision, but it can also run tests and iterate on its own until the results are up to standard – a self-correcting "smart coder".

OpenAI CEO rave: One person can make a bunch of software!

Sam, CEO of OpenAI Altman personally called Codex on social media, saying, "It's a software engineering agent that runs in the cloud and can handle multiple tasks at once. It's amazing and exciting to see how much software one person can create with the help of such a tool. ”

He also made a special mention of a very interesting sentence: "You." can just do things", this sentence was originally a joke he used to ridicule the entrepreneurial circle, but I didn't expect it to be so appropriate for AI now.

Not only can it write code, but it can also "verify" that it's doing it right!

You don't need to download complicated software or write a bunch of configuration files. As long as you know how to type, it's easy to get started with Codex.

How? Take a look: Users can access Codex through ChatGPT's sidebar; Want it to code? Click the "Code" button and enter your needs; Want to ask it a question about your project? Click the "Ask" button and just ask a question.

Each task is handled in a self-contained environment that is pre-loaded with your codebase, like giving the AI its own workspace to work without disturbing your main system.

Codex doesn't just end up writing code, it also: reading and editing files; Run various testing tools, code checkers, type checkers, etc.; When the task is completed, the modifications are automatically committed, and terminal logs and test results are generated as evidence.

In other words, it not only does the work, but it also gives you "homework" so you can clearly see how it was done and if there were any mistakes – complete transparency!

The most reassuring thing is that Codex works without an internet connection. It only relies on user-provided code and related dependencies, and does not sneak online to find information or leak your data.

This design has two benefits: high security: there is no need to worry about the code being uploaded to an external server; Low risk of abuse: Because it doesn't have free access to the network, it's less likely to be exploited.

Alexander Embiricos, Head of OpenAI's Desktop & Agent Team, also emphasized: "This is more than just a model API. Because it's running in an isolated environment and has human review, we can feel more comfortable giving the model more freedom. In other words, Codex is powerful, but it is a "caged" tiger that does things with boundaries and oversight, which is both efficient and controllable.

OpenAI's launch on Codex: The timing is just right

AI coding tools are now one of the fastest-growing areas of technology. The CEOs of Google and Microsoft both mentioned that about 30% of their company's code is now written in AI. What does this mean? It shows that the role of AI in software development cannot be ignored!

Just a day before OpenAI released Codex, an AI coding startup called Windsurf launched its own series of coding foundation models, SWE-1. According to media reports, OpenAI is discussing the possibility of acquiring Windsurf for about $3 billion.

Not only that, but OpenAI had previously negotiated an acquisition with Cursor, another fast-growing AI development tool, though Cursor is said to have eventually pulled out of the talks. Still, Cursor's performance has been impressive: in April, Cursor achieved annualized revenue of about $300 million; It was recently reported that Cursor's development arm, Anysphere, is in the midst of a new round of funding that has reached $900 million and values the company at a whopping 90billion dollars.

According to SimilarWeb, traffic to AI tools for developers has increased by 75% in the last 12 weeks. This shows that the demand for coding assistants is growing rapidly, and developers are increasingly relying on these smart tools to increase efficiency and reduce errors.

Jingtai believes that for investors, the opportunities in the field of AI programming are not small. Why? Significant revenue: Tools like Cursor have achieved $300 million in annualized revenue, which is a testament to the real need for the market.

Changing the development process: These tools aren't just ancillary tools, they're reshaping the entire software development process. More and more technology companies rely on these tools to accelerate development and reduce costs.

Continued growth potential: As more businesses recognize the value of AI programming tools, companies that invest in this space or provide related services are likely to see continued growth.

From an investment perspective, there are a few key highlights of a product like Codex that deserve our attention:

Developer efficiency revolution: In the future, enterprises will develop software faster and at a lower cost. On-premise + cloud hybrid architecture trend: AI tools are beginning to develop in the direction of "local security execution + cloud collaboration".

There is great potential for the commercialization of AI programming assistants: like GitHub Copilot and Amazon CodeWhisperer and others are already charging for monetization, and the launch of Codex has undoubtedly ignited the enthusiasm of this track again; Design concept of both compliance and security: For sensitive industries such as finance and healthcare, this "no network, no leakage" feature is especially important.