Not op, but Kagi user. Also have perplexity but usually use kagi.

I would say: 1) The UI. You’re still performing normal searches in Kagi. But if you hit q, or end your query with a question mark, you get an llm synthesized answer at the top, but can still browse and click through the normal search results.

2) Kagi has personalization, ie you uprank/downrank/block domains, so the synthesized llm answer should usually be better because it has your personalized search as input.