Hey, thanks for the detailed feedback. This is a topic we hear about regularly, and it’s covered in a few related threads:
- Let us disable composer-2-fast, it's 3x the price and forced on subagents
- Subagents are not useful if we cant select their model
- Feature Request: Dynamic model assignment for sub-agents (similar to Copilot)
- Disabling subagents should actually disable Composer 2 usage legacy plan users should not be forced into Composer 2 Fast
Here’s what I can share point by point:
-
Model selection for subagents: We’re tracking this issue and work is in progress, but I can’t share a specific ETA. Some related bugs, like silently swapping
composer-2tocomposer-2-fast, have already been fixed in recent releases. -
Fully disabling subagents: Right now the only reliable way is a workspace rule, like you did. I get that this is a workaround, not a real solution.
-
Gating behind Max Mode: I’ll pass along the feedback that basic control shouldn’t be locked behind Max. Pricing and gating decisions aren’t on my side, but I’ll log it as a clear signal.
On the “dark pattern” point: @valentinoPereira is right that using a smaller model for subagents is a common pattern due to cost and latency tradeoffs. I agree with you that the issue isn’t the idea itself, it’s the lack of control and transparency, and that’s what we’re trying to fix.