Question: Intended workflow

[illumineawake]

I just wanted to gather the communities thoughts on intended/best workflow. In particular when initiating a chat in VS Code (for e.g. with AI Studio) it will open a new chat in AI studio and when the response is finished you apply the changes and all good.
Then if you have a follow up task is the intention you go back to VS Code with the new task in web chat and press send which will open a new chat in AI Studio?
This workflow has been working well for me but obviously you end up with tonnes of chats doing it this way.
Do users continue the discussion in AI studio even after applying the initial change and work that way or is starting a new chat every time most effective?

Lastly in terms of most effective FREE model for applying a change in VS Code, I've been using Flash 2.0 on Openrouter and it seems to be working well without running into limits, any other insights on best model for this task?

[robertpiosik]

Your workflow is perfect. What you're doing is a single-turn prompting, which is great. Model's performance degrades with each turn and the best approach is to, even with the same task, work on initial instructions and re-send until you have expected results.

Lastly in terms of most effective FREE model for applying a change in VS Code, I've been using Flash 2.0 on Openrouter and it seems to be working well without running into limits, any other insights on best model for this task?

I'm rocking Flash 2.0 thinking via Gemini API. It has 500 free calls a day:-)

[illumineawake]

Beauty! How have you been handling the insane amount of various chats you end up with using this method? Do you delete them after it's applied? I was also considering making a new account so they all sit separately in that, but I was a bit paranoid it would be against TOS

[robertpiosik]

In AI Studio you can disable auto-save.

[djex]

Beauty! How have you been handling the insane amount of various chats you end up with using this method? Do you delete them after it's applied? I was also considering making a new account so they all sit separately in that, but I was a bit paranoid it would be against TOS

Like @robertpiosik mentioned you can turn off auto save. However any chats that are currently saved you can delete. They are stored on your Google Drive for that account. If you go into Drive you can deleted the saved chats.

[illumineawake]

Have you guys found much of a preference in edit format for AI studio? Just noticed there's 3 different options, for now I've been using default and only run into the rare issue.

[robertpiosik]

Every format has its pros and cons. Switcher is placed in a prominent place to allow quick selections depending on what you do.

Use "truncated" edit format when:

• you're iterating over your instructions doing single-turns (best results) and want to clearly see proposed solution
• you code manually being just guided by LLM
• you're OK with calling API with refactoring tool for each modified file when integrating changes intelligently
• you want to do more work until chatbots throttle you

Use "whole" edit format when:

• you don't want to use API for changes integration
• you want reliable results but are wary of hitting output token limits or being throttled by chatbots sooner than later

Use "diff" edit format when:

• you don't want to use API for changes integration but you're OK using it when DIFF code is incorrect
• you want quick results at a cost of readability

[robertpiosik]

Generally "truncated" will always use API for changes integration for code blocks with truncation comments, e.g. "// ...".
"Whole" will never use API - will always just replace target files. "Diff" will never use API when patches are valid.