AI-assisted programming is undergoing a fundamental paradigm shift.Recently, Cursor co-founder Michael Truell posted on the X platform, stating thatThe company has officially entered the "third era" of AI programming.The core of this new eraIt is driven by cloud agents capable of independently handling complex tasks and operating autonomously over long periods of time.This means that Cursor's positioning has undergone a fundamental shift, evolving from a "tool for writing code" to a "platform that helps developers build software factories."The data is confirming this assessment: currently, among the code commits (PRs) being merged within Cursor,35% of this work has already been completed by autonomous agents running in cloud virtual machines.More noteworthy is that...The company anticipates that within a year, the vast majority of software development work will be undertaken by these intelligent agents.This trend will not only reshape the competitive landscape of the AI programming tools sector, but will also have a profound impact on the business model of the entire industry.From Tabs to Agents: A Rapid Reversal of User BehaviorThe evolution of AI programming is exceeding everyone's expectations. In the article, he reviews the key turning points in this field, clearly dividing the development of AI programming tools into three stages:Phase 1: The era of Tab auto-completion.Tab not only completes the current line, but also intelligently predicts and completes the next line of code, even across multiple changes in different files, freeing developers from tedious code. This phase lasted nearly two years.The core idea is to automate low-entropy, repetitive tasks.Phase Two: The Era of Synchronous Agents. The core function of this era is conversational programming. Developers describe their needs to agents using natural language, and the agents generate code and respond in real time, forming a rapid interactive loop of "hint-feedback-correction."He predicted that this phase might last less than a year, and the speed of the transformation had far exceeded previous expectations.Phase Three: The Era of Cloud-Based Intelligent Agents. After developers deliver the task, the intelligent agent runs independently in a cloud-based virtual machine—autonomously completing code writing, debugging, testing, and iteration.Developers have shifted from being "people who write code" to "people who command intelligent agents."User behavior data confirms the dramatic nature of this paradigm shift: In March 2025, the number of Cursor's Tab users was 2.5 times that of Agent users; now, that ratio has completely reversed—the number of Agent users is twice that of Tab users, and usage is still surging. He revealed that many Cursor users no longer use the Tab key at all.The core advantages of cloud-based intelligent agents: parallelism and asynchronicityThe limitation of synchronous agents lies in their dual binding: they need to interact with developers in real time, but they must also compete with the local machine for computing resources. This means that the number of synchronous agents running simultaneously is extremely limited.Cloud-based intelligent agents, on the other hand, fundamentally remove these two constraints.Each agent runs in an independent cloud virtual machine. Once the developer delivers the task, they can move on to other tasks without needing to monitor it in real time. Over a period of several hours, the agents autonomously complete code iteration and testing, ultimately delivering results in the form of logs, videos, and real-time previews, rather than showing code differences line by line.This delivery method makes it possible to run multiple agents in parallel.Developers can quickly evaluate the output quality of multiple tasks without having to rebuild the context of each session from scratch. This fundamentally changes the role of humans:They evolved from people who "guided the code line by line" to people who "define problems and set review standards".Internal practices reveal new developer working modelsCursor, using its internal practices as an example, describes the specific form of this new working model. Developers adopting this new approach exhibit three common characteristics:The agent is responsible for writing nearly 100% of the code; developers focus their time on breaking down problems, reviewing artifacts and code, providing feedback, and launching multiple agents in parallel rather than guiding them step-by-step to completion.He also admitted,This model still faces challenges in achieving widespread adoption across the industry.At an industrial scale, unstable testing or a corrupted environment that a single developer can bypass can evolve into a systemic failure that disrupts every instance of an agent's operation. Furthermore, ensuring that agents have access to the complete tools and context they need remains a critical unresolved issue.At the same time, he said that Cursor's recent new feature releases are "an initial but important step" in this direction.