The AI coding assistant wars just got more interesting. On April 20, startup Cursor announced a major upgrade to its AI agents, a move it says is necessary to keep pace with heavyweights like OpenAI, Anthropic, and Microsoft.
The core of the update is boosting what Cursor calls "autonomous execution." The new agents don't just generate and modify code—they can test their own changes, documenting the entire process with video, logs, and screenshots. The goal is to make the development workflow more traceable and verifiable, letting you see exactly what the AI did and why.
How it works: A parallel, cloud-based architecture
To pull this off, Cursor introduced a parallel execution architecture built on virtual machines. Each AI agent runs in its own isolated cloud development environment. This means it doesn't hog your local computing power, and you don't have to waste time setting up the same environment over and over.
The design allows multiple tasks to run simultaneously, which Cursor claims significantly boosts development throughput. You can hand off more complex jobs to the agent, and it'll keep iterating and optimizing until the task is done.
In this model, the developer's role shifts. You're less of a hands-on coder and more of a project manager and aesthetic judge, making high-level decisions while the agent handles the implementation details.
The bigger picture: AI agents go mainstream
AI agents—software tools that can act on a user's behalf—have exploded in popularity over the last year as the underlying models have gotten smarter. Software development has been one of the first major use cases to scale.
Every big player is now pushing hard in this space. Anthropic has Claude Code, OpenAI has Codex, and Microsoft is leveraging the entire GitHub ecosystem with products like GitHub Copilot. The numbers show just how crowded the field is: Anthropic's tools reportedly generate around $2.5 billion in annualized revenue, OpenAI's Codex boasts over 1.5 million weekly active users, and GitHub Copilot has a staggering 26 million users.
Against that backdrop, Cursor, founded in 2022, faces increasingly fierce competition despite its early start and strong following in the developer community. The company recently faced community backlash after its latest agent was accused of 'shelling' Anthropic's Claude Code, a practice where a company essentially wraps another's core product with minimal changes.
Cursor says its approach is already showing results. The company reports that about 35% of code change requests (pull requests) are now generated by AI agents running in its virtual machines, a sign that automated development is moving from concept to reality. This push for automation comes as the entire category heats up, with Elon Musk's xAI preparing to enter the coding assistant arena and Anthropic continuously rolling out smarter coding models.