I actually really love Composer 2
By SwiftMaker899 · DISCUSSION · About Cursor
I know there was a lot of stink around Composer 2 and the lack of transparency around it basically being an optimised Kimi k2.5 but absolute shout out to the Cursor team on it.
It's incredibly fast compared to models like Sonnet 4.6, I've never hit a limit on it, and most of all it's very capable.
My workflow is basically this now:
* Plan mode in Opus 4.6.
* Refine the plan either manually or with more communication with Opus.
* Build with Composer 2.
* Switch to Claude Code for the \`/simplify\` command with Sonnet 4.6.
That last step is something I want to refine. I love the 3 subagents that Claude Code spins up with that built in skill. Does anyone know of something similar with Cursor or if I can just rip that functionality out of Claude Code directly and add it as a skill/command to Cursor?
And how is everyone else feeling about Composer 2 a few weeks post-release?
25 upvotes · 7 comments
Comments (7)
NeonPhoenix3047: It's my favorite model right now. It's just so fast.
DigitalHawk9933: You can easily build the simplify skill in Cursor!
ThunderGamer219: similar workflow here but I ended up going all-in on claude code since I'm mostly doing native macOS stuff in Swift. cursor's swift support never quite clicked for me. the plan-then-build approach is the key insight though, doesn't matter which tool you use for each step.
for the simplify step I bu
CosmicByte7712: Now that you mention Kimi 2.5.... hmmm, lemme get a few more crunches out..
HappyFox759: Very interested to also know if final step the author mentions can be ported over to Cursor.
ShadowVibe354: It’s doing all I need at the moment for the complex tax app I’m coding at the moment (~3k lines of python code) I spent an extra few hundred over the last few months on Opus prior to composer 2, since composer 2 I’ve not needed anything other than Composer 2… honestly I don’t care at all if it’s Kim
FrostNinja842: I don't understand the hate around it. I really like it as well. It has significantly reduced my token usage.
* I plan with the larger models (I really like GPT 5.4 atm) to get the architecture right.
* I'll run a sub-agent feedback loop on the plan with Composer 2 Fast to identify gaps.
* Then bui
More discussions about Cursor on HonestUse