This is what Twitter does. It's possibly the least performant major website I've used.
Conditional loading certainly can improve perf. in theory. I've yet to see any evidence it does so in practice. The aggregate of bundle-size, bundle-parse, client-side execution resource-usage & added latency of the plethora of metadata normally bundled with API responses is more than enough to negate any actual perf. gains.
As for "easier to maintain", I've never seen anyone even try to make that argument in theory, nevermind practice. Pretty sure it's widely accepted even by advocates of this architecture that it's a trade-off of perf. gains for ease-of-maintenance losses.
Just because Twitter does that doesn't mean it is the case everywhere else.
It moves some of the rendering work from the backend (having to query the data and generate markup) to the browser (query the API and generate the content based on the responses).
At my current job, it's made a significant improvement. The server returns compact JSON data instead of HTML, so it's easier to generate the data and uses less bandwidth.
It also looks faster for the user, because they change search parameters and only part of the page changes, rather than reloading the entire page.
As for "easier to maintain", that may be subjective. Code to generate a simple HTML template from results is replaced by JavaScript code to hit the API and generate the DOM. Although HTML5 templates makes that much easier.
I'm not saying it's impossible - glad to hear you've successfully implemented it in your workplace. I'm just saying that by-and-large it has the opposite effect to the stated intent.
If most examples of a strategy make things worse, and only one person uses that strategy to improve things, then going around saying "everyone is doing it wrong" rather than questioning the strategy isn't particularly sound.
I've build plenty of (small) client-side rendered UIs myself that lazy-load content; I know the trade-offs and I even believe I can achieve a performant outcome on my own. But that's anecdotal. In the wild, I have not seen a single major website improve perf. via lazy-fetched content rendering.
Conditional loading certainly can improve perf. in theory. I've yet to see any evidence it does so in practice. The aggregate of bundle-size, bundle-parse, client-side execution resource-usage & added latency of the plethora of metadata normally bundled with API responses is more than enough to negate any actual perf. gains.
As for "easier to maintain", I've never seen anyone even try to make that argument in theory, nevermind practice. Pretty sure it's widely accepted even by advocates of this architecture that it's a trade-off of perf. gains for ease-of-maintenance losses.