I find the opposite approach of local-first (database in the client).
It fits more solutions and solves many more things (free offline, syncing, undo support) and getting rid of caching, retries, and latency.
Of course the industry wants you to want more on the server though.
Same view applies to RSC (react server components).
Embrace CRDTs.
local-first is more decentralized as your clients can sync directly with each other (or through a dumb proxy for NAT punching)
I see no need for servers to be anything but dumb proxies for clients. Not the opposite of making clients dumb terminals.
of course, it's not right for everything and there are alternative approaches that work better or worse, depending on context, but i'd say it's hard to not call it a pretty successful approach to distributed systems
seems more like a successful escape hatch. yes it's simpler, but it's antithetical to the goal.
it seems to have been fairly successful to me
i think you are associating 'distributed' w/peer-to-peer models, whereas i'm using it in the broader sense of networked software that can interact (link) to one another
i have no beef w/ peer-to-peer and it would be interesting to think about how hypermedia could layer on top of that model, but i'm not an expert on that sort of stuff so I can't have particularly informed opinions on it