Cursor enters AI agent race against Codex and Claude Code

AI used to help you write code; now it is starting to write it for you.

Now we can see that shift in Cursor’s latest update. The AI code editor, built by Anysphere, has introduced an agent-first experience that moves beyond autocomplete and chat into full task execution.

Launched on 2 April, Cursor 3 positions the editor as a control layer where developers can assign work to AI agents, monitor progress, and review outcomes before committing changes.

Coding is becoming a delegation

Cursor’s update reflects a deeper shift in developer workflows. Instead of writing code line by line, developers can now describe a task in natural language. The system then spins up an agent that plans, executes, and delivers results.

This reframes coding as delegation. The developer defines intent and evaluates output, while the agent handles execution. It is a subtle change, but one that could reshape how software is built.

Your IDE is now a command centre

The biggest transformation is in how work is organised. Cursor 3 introduces an interface where developers can run multiple agents at once. A central task box acts like a command layer, while a sidebar tracks all active agents and their progress.

Each agent works in the cloud but returns results locally, allowing developers to inspect and edit code in their familiar environment. This makes parallel development far more practical, with multiple tasks progressing simultaneously.

From prompt to pull request

In practice, the workflow is designed to feel seamless. A developer describes a task. The agent interprets it, breaks it down, and executes it. The output is then surfaced locally, ready for review, testing, and modification.

Nothing is automatic by default. Developers remain in control, with the ability to pause, stop, or refine agents at any stage. This ensures that automation does not come at the cost of visibility or accountability.

The real battle is workflow, not models

Cursor’s move puts it in direct competition with Anthropic and OpenAI, which are building their own agent-based coding systems. But the competition is now becoming about how developers actually work.

The AI company is betting that tighter integration inside the editor, along with better task management, will matter more than raw model capability. At the same time, pricing and reliability remain critical factors influencing adoption.

Building beyond

Cursor is also investing in its own models. With the introduction of Composer 2, built on top of an open-source foundation, the company is starting to control more of its stack. Plans include training models from scratch, a move that could help optimise performance and costs. This signals a shift from being just a tool to becoming a platform.

Will developers become reviewers?

As tools like Cursor evolve, the role of developers is changing. Instead of focusing solely on writing code, they are increasingly reviewing, guiding, and refining AI-generated output. This requires a different kind of expertise, one that combines technical understanding with judgement.

Companies now expect developers to focus more on thinking through problems and checking results, while AI handles much of the coding, according to news reports.

This transforms what it means to be a good developer. It is less about typing code and more about judgment, clarity, and decision-making. AI can speed things up, but it also means developers carry more responsibility to make sure everything actually works, is secure, and makes sense.

The bottom line

Cursor 3 is changing how coding happens. By turning the editor into a space where tasks are assigned and managed rather than manually executed, Cursor is pushing towards a future where developers orchestrate systems instead of writing every line. Whether that future works will depend on one thing: how reliably these agents deliver.