Windsurf AI as a full IDE is the more interesting story than the VS Code plugin
The Cascade Agent with model selection between free Claude-based Cascade Base and premium GPT-4o/Claude 3.5 Sonnet is the right architecture for mixed-importance task routing. Not everything needs the most capable model and having the choice built into the chat interface rather than requiring separate tool switching is the right design.
Write and Chat mode distinction matters for how you interact with it. Write mode for direct code modification. Chat mode for explanation and exploration before changing anything. The discipline of using the right mode reduces unintended modifications significantly.
Tab Smart Suggestions for context-aware code scaffolding is the feature that makes Windsurf feel like a collaborator rather than a typing assistant. Suggestions calibrated to what the surrounding code is doing are a noticeably different quality.
The free tier being functional rather than crippled is the reason I am recommending it for evaluation. The main limitation being potential hallucinations with the free Cascade Base model is the honest caveat.
Has anyone switched from Cursor to Windsurf or vice versa? What was the deciding difference?