I'm not sure if this is entirely related, but during COVID (in SF) I was always surprised to see my colleagues so adamant against RTO once the vaccine was widely disseminated. I always felt that if jobs were remote there would be no need to hire us specifically - they'd hire in the cheaper parts of the country remotely or worse yet, other countries. The usual retort was that folks in North Carolina, Canada, Eastern Europe, Bangalore or whatever weren't as good - but I always thought such rebuttals were arrogant at worst, and ignorant at best.
Sad to see that to some extent that's exactly what happened. My current tech take is that developers shouldn't allow AI and/or agents to do the entirety of their job, rather allow them to do more, and it should be framed as such specifically. e.g. don't use AI to write the entire feature, use it to make the feature better and drive more revenue, or more correct and result in less bugs, thus less wasted effort on ops/etc.
It's amusing to see some people excited for AI to do the entirety of the implementation and planning, as if there would be no impact on us. If AI is that good, you just need TPMs (to the extent it's even possible, anyway).
It's no surprise to see that jobs that cannot be done remotely are making a comeback.
seems outrageous to me. money would've been better spent removing all dry wall and existing insulation, spraying in closed spray foam, putting up the sheetrock and repainting, and then buying a single portable inverter air condition unit.
self hosting is obviously a spectrum - with building a physical building for space, getting power, equipment, techs, etc on one end, and something like Google Cloud Run or DynamoDB on the other, but...
I don't see why it's that hard these days. is it really so unreliable to go on something like Hetzner, install Ubuntu, Docker Swarm, and a clustered database like TiDB or CockroachDB both of which support functionality to make it easy to schedule backups to something like s3, or other hosts, or ftp, etc.
Even if you have updates, do it one machine at a time, swarm by default is load balancing your traffic, and you're using a clustered DB so it shouldn't be a problem.
and of course, put it all behind cloud flare, because why not.
I get how like in 2012 it was annoying. you had to use Postgres, which is a great database, but then you had to deal with backups and k8s and swarm barely existed, so you had to roll your own nginx or apache config for load balancing and was annoying. in 2025 seems crazy to not do it. going to the article, obviously you should as a sass support it, if my premise is accepted =)
and before someone talks about security - it's also very easy to set up service accounts or iam improperly and leave your rds, firebase, whatever thing totally open or on defaults...