I don't think the only drive is top-down. I work in a place where devs get a lot of say in which tech we push (other than core languages; more just for hiring/market/support purposes) and there are a lot of devs that seem all about serverless .. and then are on 3 hour support calls with me because they can't get some set of A + B + C components working with lambdas.
I haven't done direct development with lambdas, but I've had to help other devs debug issues (currently in a devops role, but I was mostly a dev for like 15+ years before) and been to a few meetups where I watched people advocate, write and show severless deploys on AWS and Google.
I can see some use cases for it, but I honestly hate losing a lot of the introspection and debugging capabilities with them. I'd personally have to deal with setting up my EC2 instance or K8s/maraton deploy and being able to get to the underlying system if I need to. If things go well with a lambda, then you're all set, but the moment you need to start debugging anything serious, I feel like you're going to be in a world of pain.
In one specific instance, we had a developer who set up a lambda and also setup nginx to route to it because he didn't want to use API Gateway, which he got working, painfully after a lot of work. He should probably put up a blog post on it honestly, as I don't think it's officially supported.
Of course, this is all anecdotal. It's really difficult to measure things like speed and cost objectively with this stuff since everyone's stack and needs and uses are just a little bit different.
API Gateway gets expensive when things get busy. App load balancers also let you point methods at Lambdas though, so nowadays that's likely a better idea (and certainly more serverless) than doing the routing in nginx.
I haven't done direct development with lambdas, but I've had to help other devs debug issues (currently in a devops role, but I was mostly a dev for like 15+ years before) and been to a few meetups where I watched people advocate, write and show severless deploys on AWS and Google.
I can see some use cases for it, but I honestly hate losing a lot of the introspection and debugging capabilities with them. I'd personally have to deal with setting up my EC2 instance or K8s/maraton deploy and being able to get to the underlying system if I need to. If things go well with a lambda, then you're all set, but the moment you need to start debugging anything serious, I feel like you're going to be in a world of pain.
In one specific instance, we had a developer who set up a lambda and also setup nginx to route to it because he didn't want to use API Gateway, which he got working, painfully after a lot of work. He should probably put up a blog post on it honestly, as I don't think it's officially supported.
Of course, this is all anecdotal. It's really difficult to measure things like speed and cost objectively with this stuff since everyone's stack and needs and uses are just a little bit different.