Cursor 3 Isn't an IDE Anymore. It's an Agent Switchboard.

By John Davenport · Published on April 08, 2026

Cursor shipped version 3 on April 2, and I’ve been poking at it for a couple days now. This is not a point release. They rebuilt the entire interface from scratch under the codename “Glass,” and the result is a bet that you’re going to manage agents, not write code.

Here’s what actually changed and what it means for the rest of us.

They Ditched the IDE Layout

The old Cursor was VS Code with a really good AI chat panel bolted on. Cursor 3 flips that. The primary interface is now an “Agents Window” - a full-screen workspace for running and managing multiple AI agents simultaneously. The traditional editor? Still there, but it’s the secondary surface. You switch into the IDE when you need it, not the other way around.

This is a philosophical shift. The team said explicitly they rebuilt the interface “from scratch, centered around agents.” They’re positioning this as the “third era of software development” where agent fleets ship code autonomously.

Parallel Agent Fleets

The headline feature is running up to 8 agents in parallel across isolated Git worktrees. You can have Agent A refactoring your auth module, Agent B writing tests, and Agent C fixing CSS - all at the same time, all visible in one sidebar.

These agents run locally, in worktrees, in the cloud, or on remote SSH. And there’s a seamless handoff between environments - start a task locally, push it to the cloud when you close your laptop, pull it back when you’re ready to test. That’s genuinely useful for longer-running tasks.

There’s also a /best-of-n command that runs the same task across multiple models in parallel, each in its own worktree, then compares outcomes. That’s clever. Let Claude and GPT race on the same problem, pick the winner.

Multi-Surface Agent Control

Here’s where it gets interesting for the “IDE as orchestration platform” thesis. All local and cloud agents appear in the sidebar, including agents kicked off from mobile, web, desktop, Slack, GitHub, and Linear. You start an agent from your phone, it shows up in your Cursor sidebar. You trigger one from a GitHub PR, same thing.

This is Cursor saying: we’re not just where you write code. We’re the dashboard for all your coding agents, regardless of where you launched them.

How This Compares to Claude Code Agent Teams

Anthropic shipped Agent Teams in February 2026 as a research preview. The concept is similar - multiple agents working in parallel on shared tasks, communicating with each other, picking work off a shared list.

The difference is the surface. Claude Code Agent Teams are terminal-native. One session leads, teammates work independently in their own context windows, and they coordinate through a shared task list. Anthropic stress-tested this by having 16 agents build a C compiler - 100,000 lines of Rust across nearly 2,000 sessions.

Cursor wraps the same parallel-agent concept in a visual interface with a unified sidebar. Claude Code gives you raw power and flexibility. Cursor gives you a dashboard. The tradeoff is the usual one: visual control vs. composability.

Design Mode

Cursor 3 also ships Design Mode, which lets you annotate UI elements directly in a built-in browser and point agents at specific parts of your interface. It’s targeted at frontend iteration - click the thing that’s wrong, tell the agent to fix it. Not groundbreaking, but it reduces the friction of describing UI problems in text.

Pricing: No Change (Sort Of)

Pro is still $20/month. They didn’t raise prices with this release. But remember that Cursor switched to credit-based billing in June 2025, and that rollout was rough enough to warrant a public apology. The $20/month Pro plan now includes a $20 credit pool. Your actual costs depend on which models you use and how heavily you lean on agents.

The tier lineup: Hobby (free), Pro ($20/month), Pro+ ($60/month, 3x credits), Ultra ($200/month, 20x credits), and Teams ($40/user/month).

What This Means for the Landscape

The three lanes of agentic coding are now locked in:

  1. Terminal-native agents: Claude Code, Codex CLI, Gemini CLI
  2. Agent-first IDEs: Cursor 3, potentially Windsurf
  3. Multi-editor extensions: GitHub Copilot

Cursor at $2B ARR is the fastest-growing tool in this space, and this release widens the gap. One reviewer scored it 92/100. The IDE-as-agent-orchestration-surface is now a real product category, not just a concept.

The reality is that “where do you manage your agents?” is becoming the key question in developer tooling. Cursor is betting it’s a visual dashboard. Anthropic is betting it’s the terminal. Both are betting it’s not a traditional code editor.

What are you using to manage your agent fleet?

Sources