The IDE That Quietly Rewrote the Rules
Cursor is not a plugin. That distinction matters more than it might initially seem. While GitHub Copilot built its dominance by slotting into existing workflows – attaching itself to VS Code and asking developers to keep doing exactly what they were already doing – Cursor took the opposite approach. It built its own editor from scratch, forked from VS Code, and made AI the structural center of the experience rather than an add-on floating at the margins. The result is a product that does not just assist developers; it reasons alongside them, editing across multiple files, holding context through long sessions, and anticipating what comes next in ways that a sidebar suggestion tool simply cannot replicate.
The shift in developer sentiment has been noticeable. Forums, developer Slack groups, and X threads that once treated Copilot as the obvious default are filling up with conversion stories – engineers who tried Cursor for a week and stopped going back. The churn is not loud or dramatic. It is the quiet kind, where a tool stops being mentioned because something else has taken its place in daily practice.
Cursor is built by Anysphere, a small San Francisco-based startup that raised a reported $60 million Series A before growth pushed its valuation considerably higher in subsequent rounds. The company has operated with a level of product focus that larger organizations rarely sustain, shipping updates rapidly and keeping the feature roadmap tightly aligned with what working developers actually need, not what makes a good press release.
What Copilot Built, and Where the Cracks Appeared
GitHub Copilot was a genuine breakthrough when it launched in 2021. It turned natural language and partial code into functional completions at a speed and accuracy that had not been seen at scale before. Microsoft’s backing through GitHub gave it instant distribution, and its integration with VS Code meant millions of developers could try it without changing a single habit. At its peak, Copilot reported over a million paying subscribers, a number that represented real product-market fit in a category that had not previously existed.
But distribution is not the same as loyalty. Copilot’s architecture kept AI in a supporting role – it could complete a line, suggest a function, and occasionally generate a block of boilerplate. What it could not do well was hold a conversation about a codebase, make changes across ten files simultaneously, or understand why a piece of logic was wrong rather than just what it looked like. Developers who needed that level of collaboration found themselves context-switching between the editor and ChatGPT, pasting code back and forth in a process that was functional but deeply inefficient.
How Cursor Closed the Gap
Cursor’s core advantage is architectural. Because Anysphere controls the entire editor environment, the AI can operate with full visibility into the project – not just the file currently open, but the relationships between files, the history of changes, and the intent behind the structure. When a developer asks Cursor to refactor a function, it does not guess at what other files might be affected. It knows, because it has access to the full context. This is not a feature Copilot can simply add through a plugin update; it would require rebuilding the product from a different foundation.
The model flexibility Cursor offers has also become a meaningful selling point. Rather than locking users into a single underlying AI model, Cursor lets developers switch between providers – Claude, GPT-4, and others – depending on the task. Some models perform better on certain languages or reasoning problems, and giving developers that choice feels respectful of their judgment in a way that a closed system does not. It is a small detail that signals a larger philosophy: trust the developer to know what they need.
Cursor’s “Composer” feature – which allows multi-file editing through a natural language prompt – is the capability that comes up most often in developer discussions. The ability to describe a change in plain English and watch it propagate correctly across a codebase is qualitatively different from receiving a line completion. It removes a category of cognitive overhead that developers previously had no way to offload. Engineers working on large projects with complex interdependencies have described it as the first time an AI tool actually saved them hours rather than minutes.
Pricing has not hurt either. Cursor’s Pro plan, currently priced at $20 per month, sits in the same range as Copilot’s individual tier, which removes cost as a reason to stay with the incumbent. When two tools cost roughly the same and one does substantially more, the switching math is straightforward. Corporate procurement is slower to move – enterprises have existing Microsoft agreements, compliance requirements, and IT approval chains that create real friction – but individual developers and small teams have been making the switch without waiting for organizational permission.
Microsoft’s Response and the Limits of Platform Power
Microsoft has not stood still. Copilot has received consistent updates, including expanded context windows and deeper integration with GitHub’s pull request and code review workflows. The company has also positioned Copilot as part of a broader Microsoft 365 and Azure ecosystem play, which matters enormously for enterprise sales where bundling and existing relationships do the heavy lifting. For a company deciding between tools at the procurement level, Copilot inside an existing Microsoft contract is often the path of least resistance.
That enterprise stickiness is real, and it is probably where Copilot’s subscriber numbers hold up best. But the developers who actually write code every day are not always the ones signing enterprise agreements. There is a growing split between what IT departments buy and what engineers actually run on their machines, and Cursor is winning the latter group. The practical consequence is that even in organizations nominally committed to Copilot, individual contributors are often using something else for their actual work.
The competitive pressure is also not coming only from Cursor. Replit, Windsurf, and a cluster of other AI-native coding tools are each capturing segments of the market that Copilot once held by default. The category has fractured, and the question for GitHub is not just how to win back developers who left for Cursor – it is how to maintain relevance when the definition of what an AI coding tool should do is being rewritten by competitors who started with fewer constraints.
Cursor’s next real test is whether it can hold its edge as the models themselves keep improving. The gap between a well-integrated AI editor and a plugin-based one narrows if the underlying model becomes smart enough to infer context that it cannot directly observe. Anthropic and OpenAI are both developing tools of their own, and either could decide that the developer market is worth entering more directly. Cursor’s bet is that integration depth and editor quality will matter even as raw model capability improves – and for now, the developers making that bet alongside it have not been given a reason to stop.
