Original Title: Inside OpenAI's Race to Catch Up to Claude Code
Original Author: Maxwell Zeff, Wired
Original Translation: Peggy, BlockBeats
Editor's Note: In the rapid rise of AI programming agents, OpenAI, which once led the generative AI wave with ChatGPT, has unexpectedly become a 'follower' in this critical race. In stark contrast is Anthropic, founded by former OpenAI members, which quickly rose to fame in the developer community and enterprise market with Claude Code, becoming one of the important leaders in the field of AI programming tools.
This article reveals the true process behind this competition through interviews with OpenAI executives, engineers, and several developers: from the early OpenAI Codex project being split, resources shifting to ChatGPT and multimodal models, to internal teams restructuring and accelerating the launch of AI programming products, OpenAI is experiencing a shift from strategic neglect to a full chase. In a sense, this is not a lag in technical capabilities, but a misalignment of strategic rhythm: the explosion of ChatGPT changed the company's priorities, the partnership with Microsoft limited product pathways, while Anthropic bet on the AI programming track earlier.
Behind this competition, deeper issues are gradually emerging: as AI agents begin to take on more cognitive work, the software development process and even white-collar labor itself may be redefined.
The original text is as follows:
OpenAI CEO Sam Altman sat with his legs crossed in an office chair, staring at the ceiling as if contemplating some yet-to-be-formed answer. To some extent, this also relates to the environment.
OpenAI's new headquarters in Mission Bay, San Francisco, is a modern building made of glass and light wood, almost like a 'temple of technology.' The display shelves behind the reception desk are filled with brochures introducing the 'Eras of AI,' as if depicting a path to technological enlightenment. The walls of the staircase are plastered with milestone posters documenting the development of artificial intelligence, one of which records a moment: thousands of viewers witnessed live as a machine defeated a top esports team in a (Dota 2) match. In the hallway, researchers pass by wearing team gear with slogans, one of which says, 'Good research takes time.' Of course, ideally, it shouldn't take too long.
We sat in a large conference room. The questions I posed to Altman were related to the AI programming revolution sweeping the industry and why OpenAI seemed not to occupy the leading position in this wave.
Today, millions of software engineers have begun to delegate part of their programming work to AI, which has made many in Silicon Valley face a reality for the first time: automation could touch their own jobs. Programming agents have thus become one of the few application scenarios for which businesses are willing to pay a premium for AI. Logically, such a moment should ideally become the next 'victory moment' in the posters on OpenAI's staircase wall. But now, the name taking the headlines is not it.
The company's rival is Anthropic, an AI company founded by former OpenAI members. With its programming agent product Claude Code, Anthropic achieved explosive growth. The company disclosed in February that the product had contributed nearly one-fifth of its business scale, corresponding to an annualized revenue of over $2.5 billion. In contrast, according to an insider, as of the end of January, OpenAI's own programming product OpenAI Codex had an annualized revenue of just slightly above $1 billion.
The question is: why is OpenAI lagging behind in this AI programming race?
“The value of first-mover advantage is immense,” Sam Altman said thoughtfully, “we have already experienced this with ChatGPT.”
However, he believes that now is the time for OpenAI to fully commit to AI programming. He thinks the company's existing model capabilities are already strong enough to support highly complex coding agents. Of course, this capability was not accidental; the company invested billions of dollars in model training for this.
“This is going to be a huge business,” Altman said, “not only because of the economic value it brings but also because of the general productivity that programming can unleash.” He paused and added, “I rarely use this term lightly, but I think this could very well be one of those markets that reaches trillions of dollars in scale.”
Furthermore, he believes that OpenAI Codex might be the 'most likely path' to general artificial intelligence (AGI). According to OpenAI's definition, AGI is an AI system that can outperform human performance in the vast majority of economically valuable tasks.
Sam Altman, CEO of OpenAI. Photo: Mark Jayson Quines.
However, despite Altman making confident judgments with a calm demeanor, the actual situation within the company over the past few years has been much more complex. To understand the fuller internal story, I interviewed over 30 insiders, including current OpenAI executives and employees who spoke with approval, as well as some former employees who described the company's internal operations under anonymity. Through these narratives, a rare situation can be seen: OpenAI is striving to catch up.
Back to 2021. At that time, Altman and other OpenAI executives invited (WIRED) reporter Steven Levy to their early office in San Francisco's Mission District to watch a demonstration of a new technology. This was a project derived from GPT-3, trained on a large amount of open-source code from GitHub.
During the live demonstration, executives showcased how the tool named OpenAI Codex could receive natural language instructions and generate simple code snippets.
“It can actually perform operations for you in the computer world,” explained Greg Brockman, then president and co-founder of OpenAI. “What you have is a system that can truly execute commands.” Even at that time, OpenAI's researchers generally believed that Codex would be a key technology in building a 'super assistant.'
During that time, Altman and Brockman's schedules were almost filled with meetings with Microsoft—the software giant that is OpenAI's largest investor. Microsoft planned to leverage Codex to provide technical support for one of its first commercialized AI products: a code completion tool called GitHub Copilot, which can be directly embedded into the development environments that programmers use daily.
An early OpenAI employee recalled that at that stage, Codex “could basically only do autocomplete.” But Microsoft executives still regarded it as an important signal of the arrival of the AI era.
In June 2022, when GitHub Copilot was officially released, it attracted hundreds of thousands of users within just a few months.
Greg Brockman, President of OpenAI. Photo: Mark Jayson Quines.
The team initially responsible for Codex was later reassigned to other projects. An early employee recalled that the company's judgment at the time was that future models would inherently possess programming capabilities, so there was no need to maintain a separate Codex project team long-term. Some engineers were reassigned to participate in the development of DALL-E 2, while others shifted to training GPT-4. At that time, this seemed to be the key path to bringing OpenAI closer to AGI.
Subsequently, in November 2022, ChatGPT was launched and gained over 100 million users within two months. Almost all other projects within the company were forced to pause as a result. In the following years, OpenAI effectively did not have a dedicated team for AI programming products. A former member involved in the Codex project stated that after ChatGPT became popular, AI programming seemed to no longer fall within the company's new 'consumer product priority' strategy. Meanwhile, the industry commonly believed that this field had already been 'covered' by GitHub Copilot, which is essentially Microsoft's home ground. OpenAI mainly just provided underlying model support.
Therefore, in 2023 and 2024, OpenAI's resources were more directed towards multimodal AI models and intelligent agents. These systems were designed to understand text, images, video, and audio simultaneously and operate the cursor and keyboard like humans. This direction seemed more aligned with industry trends at the time: Midjourney's image generation model quickly became popular on social networks, and the industry generally believed that large language models must be able to 'see' and 'hear' the world to truly reach a higher level of intelligence.
In contrast, Anthropic chose a different path. Although the company is also developing chatbots and multimodal models, it seemed to realize the potential of programming capabilities earlier. In a recent podcast, Brockman admitted that Anthropic had been 'highly focused on programming capabilities' from a very early stage. He pointed out that when training models, Anthropic not only used complex programming problems from academic competitions but also incorporated a large amount of 'messy' code issues from real code repositories.
“This is a lesson we realized later,” Brockman said.
At the beginning of 2024, Anthropic began training Claude 3.5 Sonnet using data from these real code repositories. When the model was released in June, many users were impressed by its programming capabilities.
This performance was particularly validated in a startup called Cursor. This company, founded by a group of twenty-somethings, developed an AI programming tool that allows developers to describe their needs in natural language, with AI directly modifying the code. After Cursor integrated Anthropic's new model, its user base grew rapidly, as revealed by a source close to the company.
Months later, Anthropic began internally testing its programming agent product, Claude Code.
As Cursor's popularity continued to rise, OpenAI once attempted to acquire this startup. However, according to several sources close to the company, Cursor's founding team rejected the proposal before negotiations had advanced significantly. They believed that the AI programming industry had massive potential and thus wished to continue developing independently.
Andrey Mishchenko, head of OpenAI Codex research. Photo: Mark Jayson Quines.
At that time, OpenAI was training its first so-called 'reasoning model,' OpenAI o1. These models can perform step-by-step reasoning on questions before providing answers. When OpenAI launched it, they stated that the model particularly excelled at 'accurately generating and debugging complex code.'
Mishchenko explained that a significant reason for the notable progress in programming capabilities of AI models is that programming is a 'verifiable task.' Code either runs or it doesn't, providing very clear feedback signals to the model. Once an error occurs, the system can quickly identify where the problem is. OpenAI has leveraged this feedback loop to continually train o1 on more complex programming problems.
“Without the ability to freely explore, implement modifications, and test its own results within the codebase—these are all part of 'reasoning' capabilities—today's programming agents could not reach their current level,” he said.
By December 2024, multiple small teams within OpenAI had begun to focus on AI programming agents. One of the teams was co-led by Mishchenko and Thibault Sottiaux. Sottiaux had previously worked at Google DeepMind and is now the head of Codex at OpenAI.
Initially, their interest in programming agents mainly stemmed from internal research and development needs, hoping to use AI to automate a large number of repetitive engineering tasks, such as managing model training tasks and monitoring GPU cluster statuses.
Another parallel attempt was led by Alexander Embiricos. He previously managed OpenAI's multimodal agent project and is now the product head of Codex. Embiricos had developed a demo project called Jam, which quickly spread internally within the company.
Thibault Sottiaux, head of OpenAI Codex. Photo: Mark Jayson Quines.
Unlike controlling a computer through a mouse and keyboard, Jam can directly access the computer's command line. The 2021 Codex demonstration only showcased AI generating code for humans to run manually, whereas Embiricos's version could execute that code by itself. He recalled being nearly stunned watching a webpage that recorded Jam's actions in real-time refresh continuously on his laptop.
“For a while, I kept thinking that multimodal interaction might be the path to achieving our mission. For example, humans share screens with AI and work together all day,” said Embiricos. “Then it suddenly became very clear: perhaps giving the models direct programmatic access to computers is the real way to achieve this goal.”
These scattered projects took months to gradually integrate into a unified direction. By early 2025, when OpenAI completed training on OpenAI o3, a model optimized further for programming tasks than OpenAI o1, the company finally had the technical foundation to build a true AI programming product. However, by then, Anthropic's Claude Code was already ready for public release.
Before the release of Claude Code (which was launched in February 2025 as a 'limited research preview' and went fully online in May), the mainstream mode in the AI programming field was still referred to as 'vibe coding.' Developers pushed project progress through AI-assisted tools, with humans controlling the direction while AI supplemented the specific implementations during the process. Such tools had already attracted hundreds of millions in investments.
However, Anthropic's new product changed this dynamic. Like the Jam demonstration, Claude Code can run directly through the computer's command line, which means it can access all of the developer's files and applications. Programming is no longer just 'AI-assisted,' but developers can hand over entire tasks directly to AI agents.
Faced with this change, OpenAI began to accelerate the launch of competitive products. Sottiaux recalled that in March 2025, he formed a 'sprint team' tasked with integrating multiple internal teams within weeks to quickly launch AI programming products.
Meanwhile, Altman attempted to achieve 'curve jumping' through acquisitions, offering $3 billion to acquire the AI programming startup Windsurf. OpenAI executives believed this deal would bring a mature AI programming product, an experienced team, and an established client base.
But the acquisition then stalled. According to The Wall Street Journal, the problem lay with OpenAI's largest partner, Microsoft. Microsoft wanted to gain access to the intellectual property of Windsurf. Since 2021, Microsoft has been using OpenAI's models to support GitHub Copilot, which has also become a highlight in Microsoft's earnings call. However, with Cursor, Windsurf, and Claude Code launching new AI programming agent experiences, GitHub Copilot began to seem like it was stuck in the previous generation of AI tools. If OpenAI were to release a new programming product, it might not be good news for Microsoft.
This acquisition negotiation coincidentally occurred during one of the most tense periods in OpenAI's relationship with Microsoft. The two parties were renegotiating their cooperation agreement, while OpenAI was trying to diminish Microsoft's control over its AI products and computing resources. Ultimately, the Windsurf acquisition became a casualty of this game. By July, OpenAI abandoned the deal. Subsequently, Google hired Windsurf's founding team, while the remaining employees were acquired by another AI programming company, Cognition.
“At the time, I certainly hoped this deal would go through,” Altman said, “but not every deal can be controlled.” He mentioned that while he initially hoped the acquisition of Windsurf “could accelerate our progress to some extent,” he was equally impressed by the momentum of the Codex team. While negotiations were ongoing, Sottiaux and Embiricos continued to develop products and roll out updates.
By August, Altman decided to accelerate progress comprehensively.
Alexander Embiricos, head of OpenAI Codex. Photo: Mark Jayson Quines.
Greg Brockman’s preferred way of measuring AI capabilities is a mini-game he designed himself, the 'Reverse Turing Test.' Years ago, he handwrote the code for this game and now entrusts the task to AI agents to reimplement it from scratch.
The rules of the game are simple: two human players sit in front of different computers, and each screen shows two chat windows. One window connects to another human player, while the other connects to AI. Players need to guess which window is AI while trying to make their opponent believe they are the AI.
Brockman said that for most of last year, OpenAI's strongest models took hours to build such a game and required a lot of specific human instructions and assistance in the process. But by December of last year, Codex was already able to directly generate a fully functional version through a well-crafted prompt, using the new GPT-5.2 model underneath.
This change has not only been noticed by Brockman. Developers worldwide have begun to realize that the capabilities of AI programming agents have suddenly surged. Discussions around AI programming, initially focused on Claude Code, quickly broke out of the Silicon Valley tech circle and became a topic of mainstream media attention.
Even some ordinary users without programming experience have begun to use AI to directly create their own software projects.
This surge in usage was not coincidental. During this time, both Anthropic and OpenAI invested heavily to acquire more AI programming agent users. Multiple developers told (WIRED) that their $200 monthly subscription plans for Codex or Claude Code could actually yield usage credits worth over $1,000. This relatively 'generous' limit is essentially a market strategy: first, let developers get used to using AI programming tools in their daily work, and then charge based on usage in enterprise contexts.
According to several insiders, by September 2025, Codex's usage was only about 5% of that of Claude Code. But by January 2026, Codex's user base had risen to about 40% of Claude Code's.
George Pickett, a developer who has worked in tech startups for 10 years, has recently started organizing offline meetups themed around Codex.
“I think it's clear that we are replacing white-collar jobs with AI agents,” Pickett said. “As for what this means for society, to be honest, no one can say for sure. It will definitely bring a huge impact, but I remain relatively optimistic about the future.”
Meanwhile, Simon Last, co-founder of efficiency software company Notion, valued at about $11 billion, said that after the release of GPT-5.2, he and the core engineering team of the company had switched to using Codex primarily due to better stability.
“I find that Claude Code often 'tricks me,'” Last said, “it says the task is running, but in reality, it isn't.”
Katy Shi, OpenAI researcher. Photo: Mark Jayson Quines.
Katy Shi, responsible for researching Codex model behavior at OpenAI, stated that while some describe Codex's default style as 'dry bread,' more and more users have come to appreciate this unpretentious way of communicating. 'A lot of engineering work is essentially about being able to accept critical feedback without taking it as an offense,' she said.
Meanwhile, some large enterprises have already started adopting Codex. OpenAI's application business CEO Fidji Simo stated, “ChatGPT has become synonymous with AI, giving us a huge advantage in the B2B market. Companies are more willing to deploy technologies that their employees are already familiar with.” She added that OpenAI's core strategy for selling Codex is to package it with ChatGPT and other OpenAI products.
Jeetu Patel, President and Chief Product Officer of Cisco, made it clear to employees that they need not worry about the costs of using Codex, as the key is to familiarize themselves with the tool as quickly as possible. When employees worry that “using these tools might cost them their jobs,” Patel's response is: “No. But I can guarantee that if you don't use them, you will lose your job because you will become less competitive.”
Today, the anxiety surrounding AI programming agents has far exceeded the Silicon Valley tech circle. The Wall Street Journal attributed part of a recent $1 trillion tech stock sell-off to Claude Code, as investors feared that software development might soon be massively replaced by AI. A few weeks later, after Anthropic announced that Claude Code could be used to transform legacy systems running COBOL (which are common on IBM machines), IBM's stock experienced its worst day in 25 years.
Meanwhile, OpenAI is also striving to push AI programming agents into the center of public discussion. The company even spent millions of dollars airing an advertisement about OpenAI Codex during the Super Bowl, instead of promoting ChatGPT.
At OpenAI's headquarters in Mission Bay, almost no one needs to be convinced to use Codex. Many engineers I interviewed said they now rarely write code themselves, spending most of their time just conversing with Codex. Sometimes, they even 'communicate collectively.'
At headquarters, I overheard a Codex hackathon. About 100 engineers crowded into a large room, each with four hours to create the best demo project using Codex. An OpenAI executive stood at the front, looking at his laptop while announcing team names through a microphone. Teams nervously walked up to the stage and introduced their AI projects in slightly trembling voices. The final winner received a Patagonia backpack as a prize.
Many projects were developed using Codex and aimed to help engineers better use Codex. For example, one team developed a tool that could automatically organize Slack messages into weekly reports; another group created an internal AI guide similar to Wikipedia to explain various internal services at OpenAI. In the past, such prototypes often took days or even weeks to complete, but now an afternoon is sufficient.
As I was leaving, I encountered Kevin Weil, a former Instagram executive who now heads OpenAI's 'OpenAI for Science' department. He told me that Codex was completing some project tasks for him overnight, and he would check the results the next morning. This way of working has become the daily routine for him and hundreds of OpenAI employees. One of OpenAI's goals in 2026 is to develop an 'automated intern' to research AI itself.
Simo stated that in the future, Codex will not only be used for programming but hopes to become the task execution engine within ChatGPT and all OpenAI products, completing various practical tasks for users. Altman also expressed that he hopes to launch a general version of Codex but still worries about safety risks.
He said that at the end of January 2026, a friend without a technical background had asked him to help install a viral AI programming agent called OpenClaw. Altman declined the request because, in his view, “it is clearly not a good idea right now,” as OpenClaw might delete important files by mistake.
Ironically, weeks later, OpenAI announced it had hired the developers of OpenClaw.
Many developers told me that the competition between Codex and Claude Code has never been so fierce. But as these tools' capabilities continue to grow and are increasingly integrated into workflows by business managers, the societal issues at hand are far beyond just 'which AI programming tool to use.'
Amelia Glaese, OpenAI's VP of research and head of alignment. Photo: Mark Jayson Quines.
Some regulators are concerned that in the race against Claude Code, OpenAI might let safety issues take a back seat. A non-profit organization called the Midas Project accused OpenAI of weakening its safety commitments when releasing GPT-5.3-Codex, failing to adequately disclose potential risks associated with the model in cybersecurity.
In response, Glaese countered that OpenAI did not sacrifice safety to advance Codex, and the company stated that Midas Project had misinterpreted its safety commitments.
Even Greg Brockman, who donated $25 million to a pro-AI super political action committee and an organization supporting Donald Trump last year, and still optimistically stated that 'we are on track toward AGI,' has complex feelings about this new reality.
In the Silicon Valley engineering circle, Brockman has long been known for his 'extremely dedicated' management style: the type of boss who would still be deeply checking code repositories for details the night before a product release. To some extent, this more 'hands-off' approach to work now makes him feel relaxed. 'You realize that your brain was occupied by many details that were actually unnecessary,' he said.
But at the same time, when one becomes the 'CEO of hundreds of thousands of AI agents,' executing your goals and visions through these systems, it becomes difficult to delve deeply into the specific details of every problem-solving.
“In a sense, it makes you feel like you are losing the 'pulse' of the problem itself,” Brockman said.
Original link