In that time, the service-oriented code which has significantly outlasted anything else, with nearly no maintenance, has maybe surprisingly or unsurprisingly been: AWS Lambda functions.
We have quite a few Lambda functions (written in Node) that I believe pre-date even my tenure at the company; the only modifications in the interim, maybe a few lines of obscure bug fixes, updating the lambda node version, we switched from JS to TS everywhere a few years ago, and keeping the CI updated on those repos (which is a rather ironic and wasteful investment of resources given they've only changed a half dozen times in something like six years).
Checking in on those dozens of functions handling millions of invocations every month, and seeing them just chug along with no maintenance, has convinced me, personally, that if a problem domain can be solved in a FaaS-like way, its the way to do it. Not every problem can, and I'm not specifically advocating for Lambda vs Azure FaaS vs an open source kubernetes-based solution.
More-so that, every deployment artifact has to draw an "operations" line, to the left of that is the rote, daily operations banality, and to the right is the ever-changing business problem space. The further right you can draw that line, to push as much as possible into operations, the better; it gives you optionality to outsource that left part, maybe to automated software, maybe to AWS, maybe to an in-house operations team. FaaS draws that line roughly as far right as I think a general purpose programming system can; further right would approach things like Excel, Retool, etc.