I’ve observed that the composer’s agent
mode shows varying levels of intelligence in different situations. While some users have suggested adding normal mode tags like folder
and web
to the agent
mode, I’d like to propose a simpler approach:
- We should only have 2 modes:
chat
andagent
chat
mode would not modify code automatically, whileagent
mode would have this capability- With just these 2 modes, users can easily start conversations in
chat
mode and seamlessly transition toagent
mode for code refinement while maintaining context - Under
agent
mode, we would retain all features from the previousnormal
mode. Additionally, when users trigger input viactrl+enter
, it signals the AI to intelligently add more relevant context beyond what was explicitly tagged by the user.
This simplified approach maintains functionality while improving usability and reducing mode complexity.