Ever tried making large changes in a codebase using AI? If you are like me, you probably were very impressed – at first. The initial bliss quickly fades away when you realise the code will be very hard to maintain, or is flat out wrong.
Aide proactively proposes fixes or asks to include files that may be missing in the context. Our agent can do so by iterating on linter errors and pulling in relevant context using LSP tools, like “Go to references”.
Go ahead, do AI-edits on top of your coding session. We keep slim, VS Code-native checkpoints (we don’t use git) to easily roll back to previous states, in case the agent made any mistake.
We try to make Aide feel like a real engineer to pair-program with. Chat about a problem by @’ting the file(s) and then jump into edits, or go from a smaller set of edits and discuss their side-effects.
Taking inspiration from MacOS spotlight, we created a floating widget you can invoke with CMD + K. If you have a text selection active, you quickly prompt a change for it.