Feature request for product/service
Cursor IDE
Describe the request
First of all, I want to say that subagents have been an amazing addition to Cursor. They’ve made it much easier to delegate exploration and keep the main agent focused.
One thing I’ve noticed, though, is that they can make some workflows feel slower than necessary. A large part of that comes from the fact that many subagent tasks are highly repetitive across different chats.
For example, I often see subagents being asked to:
Explore the frontend structure.
Explore a specific page.
Understand the project architecture.
Inspect a particular feature or module.
These exploration tasks are frequently repeated, even though the underlying code hasn’t changed.
A possible improvement would be to make subagents cacheable and discoverable.
Instead of always creating a brand-new subagent, the main agent could first search for previous subagent runs that are semantically similar. If a relevant result exists and is still valid (for example, based on the current commit or repository state), it could reuse that result instead of performing the same exploration again.
This could significantly reduce repeated exploration work. For larger repositories, I could easily see this saving anywhere from 2–5 minutes per request, while also reducing unnecessary token usage.
I realize this is a fairly niche feature, but I think it could make repeated workflows much faster for users who spend a lot of time working in the same codebase.