Give Us Control Over Subagents — Stop Forcing Composer on Paying Customers

I’ll be direct: Cursor is losing users over this. I might be one of them.

The Problem

When working in Agent mode, Cursor silently forces subagents into Composer — a tool that consistently produces worse results than the parent agent. We have zero control over this. No toggle. No model selection. No opt-out. Nothing.

I discovered this the hard way while building a 3D-Agent with Blender-MCP. My solution? I wrote a workspace rule that literally tells the AI: “NEVER use subagents.” The results immediately improved. Think about that — the product gets better when you disable a core feature. That should alarm someone at Cursor.

What We Need (None of This Should Require Max Mode)

  1. Enable/disable subagents — A simple toggle. Let us decide if we want them at all.

  2. Choose the subagent model — Maybe I want the same model as the parent. Maybe I want a cheaper one. Maybe I want a faster one. That’s my decision, not yours.

  3. All of this available on the standard plan — Cursor is already more expensive than Claude Code. Locking basic agent controls behind Max Mode isn’t a feature gate — it’s a paywall on usability. This is the kind of thing that pushes users over the edge to switch permanently.

The Real Issue: Stop Pushing Composer

I need to say this plainly because I don’t think the message is landing internally:

The constant, silent push to funnel users into Composer is adversarial UX.

We notice. Every time a subagent quietly routes through Composer. Every time the UI nudges us toward it. Every time agent behavior degrades because Composer was injected into the pipeline without our consent. We notice all of it.

This isn’t a growth strategy. It’s a dark pattern. And it’s costing you customers in real time.

If Composer is good, let it stand on its own. Offer it freely. Price it competitively. Let users choose it because it’s genuinely better — not because you’ve shadow-routed their workflows through it.

Every internal meeting where someone pitches “let’s push more users toward Composer” needs a follow-up question: “How many users did we lose this month because of exactly that?”

Bottom Line

I pay for Cursor. Many of us pay premium prices. We chose this tool because the agent experience was best-in-class. But the trajectory is going the wrong direction:

  • Features locked behind Max Mode that should be standard
  • Subagents forced into an inferior pipeline with no user control
  • A pricing model that’s already above competitors like Claude Code

You are one or two bad decisions away from a mass migration. This post is me telling you before I become part of it.

Give us the controls. Stop the dark patterns. Respect the users who are keeping the lights on.

4 Likes

Give us the controls. Stop the dark patterns. Respect the users who are keeping the lights on.

Agree with your thoughts on giving us the control and respecting users!

But FYI when sub-agents use a less premium model - is an industry pattern called “Speculative Decoding” - where a smaller model does the grunt work and a bigger model oversees it and synthesizes its output. It’s a way to keep costs low. I would not categorise it as a dark pattern.

The dark pattern is forcing Composer as the sole subagent runtime — not merely defaulting to a weaker model. If this were about user choice, I’d be able to route subagents through Haiku or Gemini Flash. I can’t. And also, it is hidden and hard for new users to check which model the subagents are using.

That’s the tell.

1 Like

Yes - they should give an option to select other models for doing this.

Cursor was created so that developers have an option to work with whatever models they want. When that option is taken away - it signals that they are diverting away from their mission of diversity towards capital control.

Hi @Colin how are you?

Could you please provide some clarity on these issues? We are currently lacking an official stance on the situation and would like to know if there is a timeline for a fix*, or if this is something that may not be addressed.

2 Likes