I use both.
Both are working great.
My feeling in normal mode you have some more control.
When the project is getting bigger sometiems i prefer normal mode
If I know exactly what files need to be edited and I’m certain nothing else is required, I will usually use normal mode. This is particularly relevant when I only need a small part of a larger file.
The agent will see the whole file and can get confused by context that is not relevant. But if I know there’s only a certain part of a file that’s relevant, I find the normal composer does a much better job executing the command.
To synthesise, here’s when to use each mode:
Normal mode: When you know exactly what your context is and @ it in. Useful when only part of a large file is relevant.
Agent mode: When you’re unsure what your context might be, or when the context isn’t that long.
Just used agent to set up a new node project including firebase with auth and everything. Is an absolute beast, every linter and import error and other stuff like that would be a new composer prompt from me, instead if just kept going and fixed all of them. Set up tailwind, imported lucide and firebase functions and all of that stuff, i just had to press “Run command” at every step. So this process went probably 3-4x faster and took 0 mental energy than if I had did a composer prompt, then check the error, paste in the error, do another composer run, etc. etc
One request is that it doesnt seem to see my npm run dev terminal window. If it could see that then it would see the compile errors (that aren’t linter errors). Now I have to manually paste these in for it to keep going. Would be absolutely magical if it could see it.
But the agent in its current state at 0.45.11 is an absolute beast I have to say, super pleasant to use. Another minor thing that would be an easy fix and help a lot is if it knew exactly what path it was in, cause sometimes it runs a command in the wrong path if you understand how I mean. Or display to the user where it will run relative to the root