Cursor Projects: Persistent AI Project Manager Replaces Chat-Based Workflow

Cursor has launched a new feature called Projects that fundamentally changes how developers interact with its AI coding assistant. Previously, each task required opening a separate chat window; closing the window meant losing the conversation context. Projects places a coordinating agent in a durable thread that persists across sessions. You give this agent a high-level goal, and it independently decomposes the goal into subtasks, dispatches sub-agents to execute them, and gradually improves its performance as it accumulates project history.

The agent behaves like an always-on @bot that proactively works without waiting for step-by-step instructions. While this reduces manual overhead, it introduces a new control challenge: the agent decides task priorities on its own, chooses which code to modify, and determines when to act. Developers must now understand how to interrupt or steer the agent at appropriate points. This shift mirrors the broader spread of autonomous agent paradigms (exemplified by projects like grokbot) and represents a fresh test of how much control developers retain when AI tools move from reactive assistants to self-directed project managers.