Cursor announced the third era of AI software development — and backed it up with data

When a company whose product is used daily by hundreds of thousands of developers worldwide publishes a manifesto about a paradigm shift in programming, it's worth paying attention to. Last week, Cursor's engineering blog published a text titled "The Third Era of AI Software Development" — a short but concentrated document backed not by abstract reasoning, but by real data on how developer behavior is changing right now.

Cursor is one of the most successful AI-oriented code editors, built on VS Code and deeply integrated with large language models. The company is in a unique position: it sees in real time how millions of engineers interact with AI in the process of writing code. This perspective makes their classification particularly valuable.

According to the manifesto, the first era of AI development is the autocomplete era. It began with GitHub Copilot and similar tools that offered continuations of lines or blocks of code based on context. The developer remained fully in control, while AI played the role of an advanced autofill — useful, but fundamentally not changing the process. Programmers thought, wrote, and machines sometimes guessed the next few tokens.

The second era is chat assistants. ChatGPT, Claude, and built-in IDE chats transformed interaction with AI into dialogue. A developer could ask a question, request an explanation of a code fragment, or generate a function from a description. This was a qualitative leap: AI became a conversation partner and co-author. But the initiative still remained with the human — each action required an explicit request, each result needed manual integration into the project.

The third era, which Cursor proclaims has arrived, is agentic development. Here AI stops being a tool that answers questions and becomes an autonomous agent capable of independently exploring the codebase, planning changes, making edits across multiple files, running tests, and iterating until achieving a working result. In this model, the developer shifts from the role of author of every line to that of architect, task setter, and reviewer.

The most important thing about Cursor's publication is not the classification itself, but the data behind it. The company's internal statistics show rapid growth in the share of agentic sessions in overall product usage. This is not a theoretical prediction or marketing narrative — it's a measurable shift in the behavior of real engineers who are voting with their working hours for a new way of interacting with code.

This trend aligns with what is being observed across the industry. Anthropic is developing agentic capabilities for Claude, OpenAI is investing in Codex and agentic frameworks, Google DeepMind is experimenting with autonomous coding agents. The largest market players simultaneously arrived at one conclusion: the future of development lies with systems that don't just help write code, but take on entire tasks from problem statement to implementation.

The consequences of this transition go far beyond the convenience of an individual programmer. If the third era is indeed arriving, it means a fundamental reconsideration of what it means to be a developer. Skills that have been considered key for decades — coding speed, syntax memorization, the ability to keep complex data structures in mind — give way to other competencies: the ability to formulate tasks clearly, decompose problems, assess the quality of generated code, and see the architectural picture as a whole.

The economics of development are changing too: if one engineer with agentic tools can do the work that previously required a team, this will inevitably be reflected in the structure of the job market and in which projects become economically viable.

That said, it's worth maintaining healthy skepticism. Cursor is a commercial company interested in promoting a narrative about the indispensability of its product. Agentic development is still far from perfect: autonomous agents often get stuck in dead ends, generate fragile code, and require significant oversight. The third era may have already begun, but it's still far from maturity. Nevertheless, the direction of movement is clear — and those building careers in development should start adapting now.