Microsoft officially announced All in Intelligent Body, SWE Agent first exposed! Ultraman warns of a programming revolution in 2025

Ultraman predicts that software engineering will undergo a major change in 2025. The explosion of intelligent agents at the beginning of the year has made AI automated software engineering an undeniable fact. Today, Nadella officially announced that GitHub Copilot will go all-in on intelligent agents, with Microsoft's self-developed SWE intelligent agent making its debut. GitHub's current CEO Thomas Dohmke stated that the self-developed SWE intelligent agent (project codename Padawan) will also be integrated into the GitHub user experience, but it will not be released until later this year. "It can elevate the GitHub Copilot experience from a partner to the level of a human programmer." Whether it's fixing bugs or developing new modules, there's no need to specifically explain the relevant code; the SWE intelligent agent will proactively find the appropriate code and solve the problem, just like hiring an engineer. Some netizens commented that basically, every repo will have an AI contributor. Software Engineering AI Intelligent Agent Awakens In 2021, GitHub Copilot was launched and became a powerful assistant for developers. After four years, Copilot has finally welcomed a major upgrade. As mentioned earlier, the biggest highlight of this update is the introduction of the Agent mode (preview version). In this mode, Copilot demonstrates astonishing autonomy. Autonomously iterates code, identifies errors, and automatically fixes them Proactively suggests terminal commands and requests execution Identifies runtime errors and proactively fixes them In agent mode, Copilot not only iterates on its own outputs but also continuously improves the output results until all sub-tasks are completed and the developer's requests are met. What's even more powerful is that it no longer just executes the requested tasks; it can also infer additional necessary tasks to ensure the complete execution of the request. During Copilot's self-correction process, developers do not need to manually copy and paste content back to the chat window from the terminal, which is quite convenient In terms of model invocation, developers have options for models from Anthropic and the OpenAI series. In the brand new VS Code intelligent mode, Copilot iterates on its own code, proposes and guides terminal commands, analyzes and solves problems. For example, building a web application with GitHub Copilot to track marathon training. Here is a Runner Tracks website that displays results from some races. Now, if you want to improve the pagination of this table and add more participant data, just enter the request directly in Copilot Chat. ——Update the pagination button to make it look more aesthetically pleasing. Then, selecting the o3-mini model, Copilot directly starts outputting code, and returning to the Runner Tracks website, you can see the "button" has turned blue. Next, continue iterating on this website—allowing users to choose the number of rows displayed per page. At this point, changes involve more than one file. For a more challenging task, there is a page on the Runner Tracks website that lists races, and if you want to search for races by name, this will require modifications to the project’s server-side code UI and many files in testing. In this case, it’s time for the Agent mode to come into play. It excels at reasoning and iterating over the entire project and executing repetitive tasks. It can be seen that Copilot Agent autonomously updated the server backend and UI, actively fixing errors. After completing these basic items, it then turns to service unit testing, page testing, and end-to-end testing. Until the testing updates are complete, the system will actively prompt developers to run unit tests. GitHub Copilot Edits This time, a multi-file editing tool GitHub Copilot Edits is also launched simultaneously. Copilot Edits can handle multiple files in the workspace at once and provide suggestions for code changes. All of these processes can be completed directly in the editor, making it very convenient for quick code reviews while understanding the complete context of the surrounding code. During the operation, users can choose to accept or reject each edit, allowing for more flexible control. Another new change is that when switching to side-by-side view, the control options for editing remain visible. This helps users understand the broader changes. Finally, Copilot Edits has added a new setting to automatically accept editor suggestions after a timeout. This new setting is named "chat.editing.autoAcceptDelay," which specifies the number of seconds to wait before accepting Copilot Edits' suggestions. The countdown stops when the user clicks the accept button or starts reviewing changes. This setting should be familiar to those who enjoy binge-watching shows on weekends. Project Padawan "Project Padawan" can further integrate the SWE intelligent agent into the GitHub user experience. Let's look at an example: the GitHub CLI library is used millions of times a day, and despite having hundreds of contributors, there are many issues piled up. One of the bug reports shows that the command for renaming GH reports is missing validation, which would require a lot of effort if done manually. It's worth noting that this codebase has 700 files and about 200,000 lines of code. Now, with the "autonomous SWE intelligent agent," you can completely trust this AI assistant. We can assign the issue to Copilot, and it will start processing immediately. A few seconds later, it links directly from the draft PR to the issue created by the developer Next, Copilot will continuously update the PR description and submit autonomously, allowing developers to clearly see the progress of changes while sharing real-time plans. After the PR is completed, it will also push the final changes, autonomously request code reviews, and mark the PR as "ready for review." Developers also submitted their review requests here, and Copilot immediately began making changes upon receiving the task. Additionally, it proactively replied to comments on the developer's changes and pushed the latest progress to the PR. All of the above processes demonstrate that Copilot has become a true "AI engineer," capable of collaborating with developers to complete coding tasks. In response, netizens commented, "I have been using the agent for the past month, and it feels similar to Karpathy's coding style. Once you establish a feedback loop of building, testing, and iterating, you and the agent can immediately get into the groove." The developer leads the SWE agent, working with the project manager to write detailed work orders, review work, and integrate when necessary. This is the future of coding. How to use it? To use the new features of the Copilot agent, you need to download VS Code Insiders and then enable the GitHub Copilot Chat agent mode settings. Then, in the "Copilot Edit" panel, switch from "Edit" next to the model selector to "Agent." The introduction of intelligent agents will change the way developers work in the editor. To ensure seamless integration, Microsoft has simultaneously embedded it into all integrated development environments (IDEs) that support Copilot. Major Changes in Software Engineering by the End of 2025 On February 5, OpenAI CEO Sam Altman had a conversation with India's Minister of Electronics and Information Technology Ashwini Vaishnaw. During this interview, Altman emphasized the crucial role that intelligent agents and software engineering will play in future transformations. When discussing how AI will change software engineering, Altman provided an exciting outlook. He stated, "By the end of 2025, software engineering will undergo a tremendous change. This not only means a significant improvement in development efficiency but may also have profound implications for cybersecurity." AI will become a powerful assistant in software engineering, with tasks such as automated testing, code optimization, and vulnerability detection no longer being the sole responsibility of human engineers, but rather the result of close collaboration with AI. The application of AI in software engineering can not only improve development efficiency but also help solve some long-standing problems that developers have faced. For example, AI can quickly identify potential issues in code and propose solutions, greatly shortening development cycles. At the same time, AI will play an important role in cybersecurity, although this process needs to be approached with caution, as the proliferation of AI may also bring new security threats. Altman mentioned, "By the end of 2025, AI will become smarter, especially in the field of software engineering. We will see a more intelligent programming environment, with AI playing a significant role in solving complex problems." Users on social media have already begun to look forward to the arrival of the software intelligent agent era. Some users have even started to fantasize about using AI intelligent agents to establish million-dollar companies. Intelligent AI software engineering can not only enhance productivity but also enable more efficient system development and optimization, potentially changing the entire landscape of the software development industry in the future. This transformation is akin to the automated production lines of the industrial revolution, where AI will help software developers free themselves from tedious detail work, allowing them to focus their energy on higher-level innovation and design At the same time, ensuring the safety of AI technology in application and preventing its misuse will be an important issue to address in the future. Regarding these issues, Altman pointed out: "We must ensure controllable safety while developing technology. The impact of AI is profound, and we must ensure it brings positive change to the world." Deep Research Supports Research During the interview, the host posed a key question to Altman: "In the current deep research environment, is AI mature enough to play a role in some key areas?" In response, Altman confidently stated: "The underlying technology has reached a threshold, especially in the fields of disease diagnosis and education, where we have already seen amazing results. In the coming months, we will release models that can solve real-world problems." He specifically mentioned that the progress of AI is not limited to the research field; more application scenarios have quietly emerged, from medical diagnosis to educational tutoring, and the potential of AI is gradually being revealed. For example, in the medical field, AI research assistants can not only help scientists efficiently review existing literature but also find potential connections within vast amounts of data, providing strong support for scientific discoveries. However, despite AI's ability to help improve efficiency, it is not omnipotent. Altman emphasized: "Deep Research can help us improve efficiency, such as in literature review and data organization, which are heavy low-level tasks. But it cannot independently complete a complex task. Just like in cancer research, AI does not directly cure diseases, but it can help researchers find solutions more quickly." This efficient support can significantly accelerate the progress of scientific research, freeing researchers from tedious tasks and allowing them to focus more on innovation and breakthroughs. Conclusion Whether it's Microsoft CEO Satya Nadella announcing that GitHub Copilot fully embraces intelligent agents, or OpenAI CEO Altman envisioning the future of software engineering, it is clear that AI intelligent agents are leading the software engineering field into a new era. AI intelligent agents will not only become powerful assistants for developers but may also become the core force driving transformation across the industry. From code writing, error fixing, project development, to deep scientific research, the application potential of AI is being rapidly unleashed. This transformation also brings a series of thought-provoking questions: How will developers adapt to the new model of working collaboratively with AI? How can we ensure that AI applications in cybersecurity enhance defense capabilities while avoiding new risks? How do we balance the efficiency gains brought by AI with potential employment impacts? Although AI is not omnipotent, we can expect to witness a smarter and more efficient future in software engineering by the end of 2025 Source: New Intelligence, Original Title: "Microsoft Officially Announces All in Intelligent Body, SWE Agent First Exposed! Ultraman Warns of Major Programming Changes in 2025" Risk Warning and Disclaimer The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at one's own risk