Cursor Agents Automate Testing, Ship Code Visually

Cursor has unveiled a significant update to its AI agents, introducing "Computers for Agents" that equip them with autonomous cloud testing environments. This fundamentally shifts the developer workflow, moving beyond mere code generation to full-stack validation. This feature, known as Cursor Agents autonomous testing, is set to redefine how software is built and reviewed.

StartupHub data

Companies working on this

Profiles of the companies named in this story, with funding and a one-liner from our database.

Cursor (Anysphere)

$29.3B

Cursor is an AI code editor built to make you extraordinarily productive by coding with AI.

When assigned a coding task, a Cursor Agent no longer just writes code. It now spins up a dedicated cloud environment, deploys the new code, and autonomously tests the changes. Crucially, the system records the agent's interactions, simulating human inputs like mouse clicks and keyboard typing, to visually demonstrate functionality.

Visual Proof, Faster Decisions

The deliverable isn't just a pull request; it's code accompanied by a video recording. This visual proof confirms the feature works in a live environment. Engineers Jonas Nelle and Alexi Robbins from Cursor demonstrated this with two examples.

First, an agent built a "long-running time picker" UI and its backend logic. The agent then delivered a video showing a simulated mouse interacting with the new dropdown, proving its functionality. Alexi noted this allowed him to evaluate the work in seconds, bypassing line-by-line code review.

In a second demo, engineers tasked multiple AI models with adding a 3-second brand tag to videos. By comparing side-by-side video outputs, they quickly identified a successful implementation from Codex 5.3, while Opus produced a glitched, unusable result. The decision to ship was immediate, based purely on visual evidence.

From Coder to Manager

Beyond the technical features, this tool redefines the software engineering role. Developers operate at a higher level of abstraction, acting as managers of AI workers. Jonas highlighted that his manual coding "to-do list" has shrunk, replaced by a longer list of "Agents to Review."

Engineers now focus on strategic thinking, the "what" and "why" of product development, rather than the tactical "how." As Alexi summarized, this new workflow means engineers spend their time "reviewing ideas, rather than reviewing code," making development more efficient and enjoyable.