Cursor launches Agent Mode — multi-file edits with terminal access
Cursor's new Agent Mode can plan and execute multi-step coding tasks across files, run terminal commands, and iterate based on errors — all from a single prompt.
Original sourceCursor's Agent Mode transforms the AI code editor from an assistant into an autonomous coding agent. Give it a task like 'add authentication to this app' and it will plan the changes, create files, modify existing code, install packages, and run tests.
The agent has full terminal access (with user approval), can read error output, and iterate until the task is complete. It maintains context across files and understands the project structure.
Early users report it handles routine tasks like adding CRUD endpoints, setting up testing, and implementing UI components with minimal intervention.
Panel Takes
The Builder
Developer Perspective
“Agent Mode handles the 80% of coding that's boilerplate. The 20% that requires real thinking still needs a human. That's the right split.”
The Skeptic
Reality Check
“Terminal access for an AI agent is powerful but risky. The approval step is critical — never let it run unsupervised on production systems.”
The Futurist
Big Picture
“We're watching the IDE evolve from a text editor to a project manager. The editor doesn't just help you write code — it plans and executes.”