February 3rd news that artificial intelligence has had a destabilizing effect on the way software is developedprogrammingA great deal of heavy and trivial work is done by a group of intelligent and sub-intelligence. However, developers continue to explore new interfaces and product patterns of human collaboration, and even the most advanced artificial intelligence laboratories can hardly keep up with this iterative pace。
According to 1AI, the prevailing trend is the development of smart-body software, i.e., systems where artificial-intelligence intelligents can independently perform encoded tasks, and applications such as Claude Code and Work are typical of this model. In the meantimeOpenAI I'm working on it Codex The tool, which was launched in April last year in the form of a command line tool, will be available in a month。
Now OpenAI is officially taking the critical step of catching up. This Monday, local time, the company introduced a new Codex application to fit the macOS system, which integrates the various types of intelligent body-development logic that have been widely used over the past year. This new application supports multi-intellectual parallel operations, the ability to integrate different intelligent bodies and the current cutting-edge workflow. Less than two months have passed since OpenAI launched its strongest encoded large model, GPT-5.2-Codex, on which the company hopes to attract Claude Code users。
The CEO of OpenAI, Sam Altmann, said at the press conference: “GPT-5.2 is the most performing model of the day to deal with the sophisticated development work in a complex scenario. However, since it had previously used a high threshold, we believe that it would be of great value to encapsulate the robust capabilities of the model into a more flexible interface.”
Despite Altman ' s confidence in GPT-5.2, the results of the various coding benchmark tests show a more complex pattern. GPT-5.2 was at the top of the list as of the draft, in Terminal Bench (a test to measure the ability of artificial intelligence to program commands), although Gemini 3 and Claude Opus scored essentially the same smart body (a slightly lower value) and was within the margin of error of the benchmark test. Another benchmark test, SWE-bench, which tested the ability of artificial intelligence to repair real software loopholes, was similar, and GPT-5.2 did not show a clear advantage. However, applications of intelligent bodyization are difficult to assess effectively through existing benchmarks, and top-of-the-art models may vary significantly from user experience。
The Codex application also carries a number of new features that will help it to reach a level comparable to that of the various Claude applications, and in some cases even reverse. The application support setup automation task, which will automatically run backstage as planned, the results will be placed in the queue and will be looked at as one user returns. Users can also choose different interactive styles for intelligent bodies according to their working style, ranging from pragmatic rationality to inter se communication。
But for OpenAI, the biggest core selling point for this product is the high development efficiency of artificial intelligence. Altman says: “You can start from scratch, with this tool alone, develop a very sophisticated software within hours. As soon as I can enter new ideas, the speed with which software is developed is the only limit."