Reducing the YOLO of posit assistant

[Philip-Leftwich]

I understand that posit assistant is designed to be more agentic by design than the older positron assistant.
But I miss the ability to engage in quick chats to discuss very small/specific coding issues. I will find that the response can sometimes be overengineered, or start initiating a lot of code running.
Are there community recommendations on how to tweak the level of investment it puts into an answer?
I am thinking along the lines of structure prompts or agent.md files more clearly, settings I can alter, switching into plan mode?
Open to thoughts and recommendations from developers and users alike!

[juliasilge]

Thank you for this feedback! We've definitely heard that the highly agentic approach might not always be the best fit for all kinds of work.

• One thing you might try out is the brand new "Ask" mode that shipped with the latest version of Posit Assistant. It is a mode you switch into, similar to "Plan" mode. You can try it out (together with other experimental features) in Positron with this in your settings.json (it will look grayed out in the file, because it is not a regular setting):

"assistant.experimentalFeatures": true

• In my own work with agents, I find that how I start out the chat makes a big difference, and I am often happier if I start out saying what I am looking for, something like, "I'm not ready to implement anything, but I want to know..."