Cursor

Most developers met AI coding through GitHub Copilot: autocomplete that lives inside your editor and waits for you to type. Cursor pitches itself as something different. Its homepage calls it "your coding agent for building ambitious software," and the emphasis isn't on finishing your line. It's on handing off whole tasks. The demos show agents that plan a feature from a PRD file, explore a dozen files, work for 14 minutes, then return a walkthrough video and a staging deploy for you to review. That's a different contract than inline suggestions. You stop being the typist and become the reviewer.

The product reaches well past the desktop editor. There's a CLI you install with a single curl command, a Slack bot you can tag in a channel to open a PR, GitHub PR reviews, automations, a marketplace, and cloud agents that "use their own computers to build, test, and demo features end to end." Model choice is wide in the demos, with Grok 4.7, Opus 5.5 and GPT-5.6 Sol all appearing in the model picker. The research timeline on cursor.com lists shadow workspaces, multi-agent collaboration, semantic search and secure codebase indexing, which explains why the agent can find its way around a large repo instead of guessing.

The catch is in the pricing. The current page bundles Cursor with Grok and what look like social features ("Less timeline ads," "Preferred @handle"), and the cheaper Plus plan is limited to Grok models only. If you want Claude or GPT models inside Cursor, you'll need Premium at $30/month or higher, and even then third-party models are billed separately by usage at API list price. Usage is described in multiples (4x, 15x compared to Plus) rather than hard request counts, so it's tough to predict your bill before you've used it for a week. If you're coming from Copilot's flat pricing, budget some time to figure out which tier actually fits how hard you'll push the agents.