For Kagi, there is a big question mark around the availability of the third-party search API endpoints they rely on. If search evolves further and further away from textual matching, that will impact them as well. (For search providers, it could be commercially reasonable to replace more or less traditional document indexing with a way to synthesize responses—at a very high level similar to what some analytical tools for time series do.)

I hope that by then, they have deployed their own crawlers and indices, so that there's still an alternative, rather than merely adopting the latest industry trends.