Feature request for product/service
Cursor IDE
Describe the request
Hi everyone,
After using Cursor for over a couple of months to build a personal project that will soon go live, I have realized that what I need from the tool can’t currently be delivered flawlessly in the pipeline.
There isn’t a single AI Model capable of articulating complex workflows, understanding the pipeline, and following a predefined set of “code development” rules at every turn, losing memory and hallucinating the more the conversation progresses.
I have tried so many different approaches, but there is a clear constraint in the ability of these models to abstract themselves from the task to “grasp the bigger picture” before actually doing the code, and I have experienced this with EVERY model, regardless of the rules, regardless of the markdown files attached, regardless of the clarity of the prompt.
To make it shorter, my idea is the following:
- Allow an external Agent (that is consistently provided with the engineering rules) to overlook and judge the output of the Agent that is tasked with code changes.
- This will allow for an additional step: “Review” => Linked to a specific Agent Chat.
- The new workflow should look something like this:
=> Ask Mode (with instructions on task and development rules)
=> Review:
A. Is it following the rules correctly? [Move on]
B. Do we need to interrupt the Agent for X, Y or Z? [Back to Ask with prompt]
=> Agent mode and perform the task
=> Review again to evaluate
=> Repeat || End.
(The more autonomous the loop the better it is)
I find myself performing this operation manually almost all of the time, and countless errors are always spotted.
Current models lack the ability to be good engineers for production-grade workflows.
I believe that the introduction of an intermediate node performing this review could represent a step forward into that direction.
I hope I have made myself clear enough.
Wish you all a good day.
Cheers,
Giovanni
