Cursor 2.4 Adds Subagents—AI Becomes a Project Manager and Generates UI Mockups Instantly

Cursor 2.4 is officially released, with the headline change being the introduction of Subagents that turn the AI from a single‑threaded worker into a multi‑threaded collaborator.

1. Subagents: Let Specialists Do Their Jobs

Previously, all tasks—searching code, running terminal commands, writing code—were executed linearly in a single context, leading to context‑window bloat and confusion when switching tasks.

Subagents solve this by allowing the main Agent to create independent Subagents, each with:

Independent execution : they can run in parallel without interfering with each other.

Independent context : each maintains its own conversation context, keeping the main dialogue clean.

Specialized division : each can be configured with dedicated prompts, tools, and models.

Cursor ships with several default Subagents that handle:

Research codebase : combs through large repositories to find relevant information.

Run terminal commands : executes compilation, testing, and other shell operations.

Parallel workflow : processes multiple independent tasks simultaneously.

Analogy: before, asking the AI to implement a feature forced it to search docs, write code, and run tests all at once, causing chaos. With Subagents, one assistant fetches documentation, another runs tests, the main Agent focuses on coding, and the results are merged at the end, dramatically speeding execution and keeping the main context focused.

2. Image Generation: Sketch‑as‑You‑Code Tool

The new Cursor Agent can generate images directly, powered by the Google Nano Banana Pro model.

Typical use cases include:

UI mockups : generate placeholder or schematic UI images when asking the AI to create a web layout.

Architecture diagrams : describe a system architecture and receive a visual chart.

Asset filling : generated pictures are saved automatically in the project's assets/ folder.

While it does not replace dedicated AI drawing tools, it streamlines the workflow for developers by avoiding context switches between browser and editor.

3. Cursor Blame: Transparent Code Ownership

For enterprise developers, knowing who wrote a line of code is critical. Cursor 2.4 adds the "Cursor Blame" feature, which extends traditional Git blame with AI‑driven attribution.

When inspecting code, the tool can tell whether a line was hand‑written, generated by AI tab‑completion, or produced by an Agent during a conversation.

Was this line written by a human?

Was it generated by AI completion?

Did an Agent produce it in a prior dialogue?

If the line originated from AI, a clickable link jumps to the exact conversation that generated it, helping code review and bug tracing by showing not only who contributed but also why.

4. Proactive Clarification: From "Do as Told" to Collaboration

Earlier agents followed commands blindly and would guess or error on ambiguous instructions. In version 2.4 the Agent now asks clarification questions.

This capability works across all modes, not just planning or debugging. When the Agent lacks sufficient information, it pauses and asks, for example, "Do you prefer solution A or B?" The questions are non‑blocking: the Agent continues reading files or preparing work while waiting for the user’s answer, then integrates the response.

5. More Practical Details: Experience and Performance Boosts

Built‑in browser speed : up to 10× faster, more precise clicks, drag‑and‑drop support, and the Agent can lock the browser to avoid accidental interference.

PDF reading : drop a PDF into the Agent and it can reference its content directly in the conversation.

MCP protocol optimization : on‑demand loading of MCP tools saves tokens and keeps context focused.

Diff viewer upgrade : significantly faster diff inspection for smoother code reviews.

Hooks speed : hook command launch is 40× faster and now supports richer trigger moments such as the stop hook.

CLI enhancements : the agent command can start an interactive session and run under a service account.

These updates collectively make the development experience more efficient and collaborative.