A massive opportunity for Cursor is to allow users to introduce their own custom composer agents for specific tasks that include eval loops. This is beyond just a tool that the composer agent would use but a custom agentic workflow build on top of your current composer agent.
A nice alternative to this would be to expose some of the magic of cursor via public APIs, for example context retrieval or code/file updates.
With these features Cursor would effectively remove all of my dependence on other libraries like Langchain, Autogen and Streamlit (Cursor UI is sufficient for most Agentic coding tasks).