Even though Windsurf has a good understanding of the codebase, sometimes when you ask it to do a complex task, it may complete it in the first try but may need improvements. When you prompt it again repeatedly, after the sixth or seventh prompt, it may start performing poorly. That is one of the key things I have noticed—its performance is not very consistent with continued prompts. Additionally, it sometimes hallucinates things. Although this previously happened a lot, it has improved, and now these checkers and linter functions ensure the code it writes is correct, reducing those hallucinations, but the performance inconsistency is still something I have noticed. The UI is good as it is basically a Visual Studio fork, so it has a good user interface. Moreover, every Visual Studio integration works here, so all of those are good. One of the main issues is the agent hallucinating and the generations not being good; those are two main points.
