Cognition, a startup company that has been established for less than two months but has ten talented engineers, has made a big move to ignite the technology industry. They have launched an artificial intelligence (AI) assistant called Devin, which can assist human software engineers in completing many development tasks. Devin is different from other existing AI coders, as it can build websites from scratch, deploy applications on its own, fix vulnerabilities, learn new technologies, etc. Humans only need to play a role of giving instructions and supervision. This is the first truly autonomous AI software engineer, and upon his debut, he caused a huge uproar because people were worried: would human programmers really be unemployed? According to the Cognition official website, Devin can handle the entire development project end-to-end with just one command, surpassing its peers in test performance. According to user needs, Devin can execute a multi-step workflow simultaneously, while human engineers can observe its progress in real time, and when errors are found, jump out of instructions to correct them. This allows engineers to outsource most of their work to AI, allowing them to focus on creative work. In the SWE bench benchmark test, Devin's performance far exceeded that of players such as Claude2, Llama, GPT-4, and was able to handle 13.86% of the issues completely and correctly. By comparison, GPT-4 can only handle 1.74% of issues. More importantly, Devin did not receive any assistance during testing, while all other models require assistance, meaning that people need to accurately inform the model which files need to be edited. This progress marks significant progress in AI's ability to autonomously understand and solve software development problems. At present, Devin has successfully passed an interview with an AI company and completed actual work on Upwork. The strength of Devin's almost completely autonomous operation lies in its ability to plan and execute exceptionally complex engineering tasks, which typically require thousands of decisions to complete. In this, no matter at any step, it can recall all relevant contextual information to ensure the logical coherence of the task and facilitate calibration at any time. What's even more shocking is that Devin not only helps people solve code, but also encompasses the entire workflow related to it. For example, when engineers need to design a web game, Devin can first generate web pages, then complete server deployment, and finally directly publish and launch, saving a lot of manual operations in the middle. Even after discovering vulnerabilities, Devin will trace back to the location and corresponding data where the error occurred, analyze the cause, and provide a solution. In addition, as an AI versatile assistant, Devin can also assist humans in training and fine-tuning other AIs. For some common models, you only need to mention the name of the model, and Devin knows how to train it himself. What do people need to do? As envisioned, the ultimate Devin would allow humans to simply give orders and not do anything else. The impact on the software development industry is far-reaching. Although Cognition is only ten startups