Cursor Agentic Platform Faces Orchestration Hurdles
The agentic coding platform Cursor is facing real-world engineering challenges with its multi-agent infrastructure. Recent user reports indicate that its cloud agents cannot yet invoke subagents, limiting complex delegation. Another update removed the user review step for file changes, applying them automatically and raising questions about user control.
- The parent company of Cursor, Anysphere Inc., was founded in 2022 by four MIT graduates: Michael Truell, Sualeh Asif, Arvid Lunnemark, and Aman Sanger. - The company has experienced rapid financial growth, reaching an estimated $1.2 billion in annual recurring revenue in 2025, a 1,100% increase from the previous year. - This growth is backed by significant venture capital, including a November 2025 Series D round of $2.3 billion, which valued the company at $29.3 billion with investors like Thrive Capital, Andreessen Horowitz, Accel, Google, and Nvidia participating. - The platform's architecture is designed around an orchestrator that coordinates multiple specialized components, including a router for determining execution paths and a "Composer" model to select the best frontier coding model for a given task. - The challenges Cursor faces with agent coordination are common in multi-agent systems, which often struggle with communication overhead, resource allocation, and unexpected emergent behaviors as they scale. - With its Cursor 2.0 update in October 2025, the platform shifted to a "true agentic architecture," designed to let the AI independently plan and execute complex tasks, moving beyond the capabilities of simpler coding assistants. - Subagents, which are currently facing invocation issues, are intended for complex, isolated tasks like generating an implementation plan or analyzing an entire codebase before a refactor, running in their own context to avoid token limits. - The move toward removing manual review of file changes is part of a broader industry trend where AI agents are shifting from assistive roles to more autonomous systems that manage entire workflows without constant human oversight.
