For all the AI editors, there are too much attention to the so called vibe coding or agent.
Given that if we could achieve everything with a simple prompt, this must happen within the LLM vendor like OpenAI, Claude or others, no in the so called AI editor.
Most of the time the AI editors make some big progress, the underlying reason is the enhancement of the model, not the editor itself.
The engineers thrived to fine tune the agent or something else to make the LLM more effective, but in the same way, more complexity brings more unstability as there is a ceiling belongs to the LLM itself like context size.
Vibe coding can make us happy, or make some demos, but still not so good in the real project.
It’s time to come back the editor itself, it is an editor first, and then AI enhanced. If it’s an AI first, any LLM would replace the editor?
Focus on the editor’s functionalities(like code auto completion), develop some domain specific models, to make it a real editor.