Hi Cursor team,
I’m a solo operator building production systems with Cursor. I spend well over $1,000/month, almost exclusively on top-tier models (Claude Opus 4.6, GPT 5.4 Extra High). I wanted to share some thoughts, not as a pricing complaint, but as strategic feedback from someone who has tested the alternatives and made a deliberate choice to stay.
I recently subscribed to Claude Code Max ($200/month) to give it a fair shot. I’ve already cancelled. The core issue isn’t quality, it’s control. Claude Code optimizes for autonomy: you hand off a task and hope for the best. That doesn’t work for production systems. I need to steer the process, step by step: understand the current state, evaluate options, then execute. Cursor’s entire platform layer maps directly to how I work: interaction modes (Plan, Ask, Agent), project-level rules and skills, @-references for precise context injection, multi-file orchestration, notepads, custom commands. It’s not one feature, it’s the ecosystem that makes structured, human-steered development possible.
But here’s what I actually want to say: the reason I stay isn’t the model. Models are interchangeable, and the pricing pressure they create is, in my view, a temporary problem. Open-source models and Composer will reach “good enough” quality within the next 6 to 12 months. Once that happens, the margin problem solves itself.
What won’t solve itself is the platform question. And that’s where Cursor is already winning, even if it might not be obvious from the outside.
I’ve built an entire working system on top of Cursor’s platform layer: project-specific rules, custom skills, architecture decision records, feature documentation per domain, persona files for context. This isn’t a nice-to-have setup. None of this is portable to Claude Code or any terminal-based tool. That’s your moat, not the model.
There’s a competing philosophy right now, championed by people like Peter Steinberger, that says code should be optimized for the AI, not for humans. Let the agent name things, let it structure the project, minimize human oversight. That works for prototypes. It doesn’t work when someone has to take responsibility for what the system does, when there are tax audits, legal requirements, or real customers depending on correct behavior.
Cursor’s approach, human steers, AI executes, is the right architecture for anything that matters. My suggestion would be to double down on that. The platform features (rules, skills, context management, structured modes) are what make Cursor irreplaceable, not access to Claude or GPT.
The pricing gap to Claude Code is real today. But it’s not what will decide where power users end up in 12 months. The depth of the platform will.
Thanks for building a tool that takes orchestration seriously.
Michele