Cursor

Cursor took the opposite route to the terminal agents. Instead of handing a model a command line, it rebuilt the code editor around one, and kept the developer in the picture: you can still see every file, every diff and every suggestion. That has made it the strongest in-editor agent, and the default choice for engineers who want AI help without giving up the interface they know.

What it is

Cursor is a code editor with an AI agent built into its foundations rather than bolted on. The agent takes a task and works across the project, editing multiple files while the developer watches the changes appear as reviewable diffs, and it can run on Cursor’s own in-house model, Composer, built for long-running agent work at lower cost than frontier models. Beyond the editor sit a set of connected pieces: Bugbot for automated code review on pull requests, cloud agents that run on their own virtual machines rather than the laptop, and an SDK that lets other companies embed Cursor’s coding agent inside their own products.

How it works

Recent versions have pushed hard on doing several things at once. The agent can run tasks in parallel, and cloud agents let long jobs continue on remote machines while the developer carries on locally. Side chats and conversation search, added in version 3.11 in July 2026, exist because engineers now run enough concurrent agent conversations to need a way to find an old one.

Design Mode, shipped with Cursor 3.0 in April 2026 and expanded since, addresses a different problem. Describing a visual change in words is slow and imprecise, so Design Mode lets a developer point, draw or narrate the change they want on the interface itself while the agent edits the code underneath. Like the rest of the field, Cursor connects to outside systems through Model Context Protocol servers, which teams can share through an internal marketplace.

Where it fits

The usual comparison puts Claude Code as the deepest programmable harness and Codex as the broadest in reach, with Cursor strongest where the developer stays in the loop and wants to see the work happening. Bugbot competes in a different lane again, against automated review tools rather than coding agents; a June 2026 upgrade made it over three times faster while cutting cost and catching more bugs.

Where it stands

Cursor’s trajectory through 2026 has been away from the single editor window: cloud agents, an iOS app in public beta since June 2026, an SDK for embedding the agent elsewhere, and team features for organisations. The strategic question is whether the editor remains the centre of the workflow as more work moves to agents running somewhere else entirely.