At the same time, apps are also a bit in decline. People still make them but the whole race for making it to the top 10 in the app stores seems to have faded away. And a lot of them are simple web page wrappers. People still install some apps but more on a need to have basis than that they are constantly adding/removing apps. So, I don't buy this "the web is in decline" framing.

Change is a constant on the web. Things were very different in 1995 (plain html, no good search engines), 2005 (no widespread web capable smart phones usage yet, Google, AJAX), 2015 (peak social media and app hype), and 2025 (social media has shifted to new apps and lots of people are disengaging entirely, AI is starting to threaten Google, content aggregators serve most web content).

For 2035, I would predict that AI will drive a need for authenticity. Existing platforms don't provide this because they lack content signatures. We've had the tools to reliably sign content for decades. But we don't use those a lot except for DRM content behind paywalls (for commercial reasons). So, you can't really tell apart the AI generated propaganda, marketing, misinformation, etc. from authentic human created content by individuals you care about. And that might be contributing to people disengaging a bit. But you can see the beginnings of this on platforms like bluesky and signal which push end to end encryption and user verification. People might share AI nonsense via these platforms. But they seem to be less about that as say X, Tik Tok or Instagram are. We sometimes watermark our images. We don't digitally sign them. Why is that?

Just speculating here but the web could use a big upgrade here and do more than just certify domain name ownership. Which is fairly meaningless if the domain is some big network with many millions of users. What about certifying content itself? Reliably tie content to their creators in a way that can't be forged. IMHO this is long overdue and the related UX challenges are there but solvable in principle. DRM is a prime example of a fairly usable implementation. Just works if you paid for the content. Signed content would make it very challenging to pass off AI gibberish as authentic if it's not signed by a reputable private key. And if it happened anyway, that would damage the reputation of that key. I don't exclude the possibility of reputable AIs emerging. How would you tell those apart from the disreputable ones?