Cursor AI is improving Sonnet 3.7 in version 0.47 to allow users to delegate more complex and demanding tasks to the AI model. The key is to utilize the model for long tasks, complex problems, and in-depth thinking. The update optimizes the AI for asynchronous work, which involves delegating tasks, and will be available with usage-based pricing due to the increased costs associated with the enhanced capabilities. A future post will detail the changes and how to enable them.
dont want agent mode at all tbh. anyone know of alternatives that are not going in agent direction. the normal mode review and apply performance is noticeably better. think its time to admit .46+ was a mistake and revert to .45 and start over
Interesting reading about what manus (new agent from China) tick. One of the team says:
“Multi-agent implementation is one of Manus’s key features. When messaging with Manus, you only communicate with the executor agent, which itself doesn’t know the details of knowledge, planner, or other agents. This really helps to control context length.”
I wonder if this is the way cursor will go longer term. I understand the new cursor will have much granular control over models - with a new UI for selecting and modifying models.
It does feel like it would be better to have agents tuned/prompted for different roles e.g. Project manager vs planner vs coder vs. tester vs debugger. And different LLM’s might best for each of these (though I imagine you can pretty far with Sonnet 3.7)
I notice in the latest update to 4.6 you can select agent, ask and edit, not sure when that appeared.
What is the difference between agent, ask and edit in the drop down menu?
Differences Between Agent, Ask, and Edit Options
The dropdown menu you’re referring to likely contains different interaction modes for working with an AI assistant. Here’s what each typically means:
Ask: This is the standard mode where you ask questions or request information, and the AI provides answers based on its knowledge. It’s a straightforward question-and-answer interaction.
Edit: This mode is designed for when you want the AI to help modify or improve existing text or code. You provide content that needs editing, and the AI suggests changes, corrections, or improvements.
Agent: This typically refers to a more task-oriented mode where the AI acts as an agent to accomplish specific goals on your behalf. It might involve more complex, multi-step processes or specialized capabilities beyond simple Q&A.
Each mode is optimized for different types of interactions, allowing you to choose the most appropriate way to work with the AI based on your current needs.
It started with 0.46.0 first launch. Surprised they didn’t mention it specifically in the patch notes, but I saw a dev post about it on here and it is in their documentation now: Cursor – Composer
good news
I really hope cursor’s positive future
cursor is too deeply inside of my develop life
one thing to suggest feature is “get longer time to thinking”
compare with o3-mini-high, sonnet-3.7-thinking’s "thinking"time is too short I feel.
some research says just give long time to think makes better respond
However last week I did finally experience the stuck terminal issue once. It happened in a long-running chat that had made a lot of terminal calls, and eventually one of them got stuck. After cancelling the stuck one, future terminal calls worked after that.
