We need agents with parallel tool calling

Ideas Feature Requests
1

Feature request for product/service

Cursor IDE

Describe the request

I have been using Cursor AI for a while now and I quite like it. However, one frustrating point is seeing how painfully sequential most agent tool calls are. There could be a scenario where agent needs to do a simple “find and replace all” code edit, and it would sequentially do them one by one, accumulating context and billing costs.

There is a lot to be gained by implementing some kind of task pooling workflow where agent can queue up multiple tool calls, then review their results after they all finished in a single chat request.

If you actually look at history of many chat sessions that involve 10+ tool calls, you will see plenty of opportunities for consolidating them. Even if tools themselves require sequential execution, such as file edits, the model does not need to examine result after every single call.

4 Likes
2

would be cool and fast af, even if all models used tools as fast as Composer

3

Other IDEs already support it.

4

Actually, I’m curious to know if the Cursor team has actually done some process optimization. I asked Cursor to perform some advanced substitutions, such as treating a first-person short story into the third person. It always can’t be executed, it likes to replace it one by one, and then when it gets annoyed, it starts writing scripts, and then executes it directly without testing, and finally changes the data source in a mess, and finally proudly says that it has succeeded.