AI Is Eating SaaS: Building an IP Geolocation API in Two Hours in Rust
vpetersson.com
vpetersson.com
I feel like good engineers have always been able to write like 1000’s of line PR’s,
But I don’t outsource postgres to RDS because I can’t find enough devs to write a database.
It feels like we outsource a lot of stuff because of the mental load and overhead of running it in production.
But it seems like there’s this assumption in the thought leadership that says all software written by LLM’s will just work.
Maybe this makes sense more at small companies where like the limiting factor is the code? You have low usage of the service, don’t really need it to be HA(or can package it on the server in such a way that it’s tied to the lifecycle of the monolith)
Was there no open source ip geolocation they could self host before?
Well, certainly not all software. But a static, no deps, no external loading, no db, no dynamic anything binary, written in rust and running on a hardened pod? Yeah, that'll likely "just work".
Note the many good choices made by the person who implemented this. They're not vibe coding stuff straight to prod. They had a hunch it could be done, and tried it out. When it seemed to work, they took the time and effort to think about deployment, make more good choices and so on.
> Was there no open source ip geolocation they could self host before?
Probably, yeah. Would running those have taken less time to audit? I don't think so. Plus they'd probably come with 25 other features that OP didn't need/want. And someone would have still have to spend time integrating it. What's the point then, for such a limited scope app?
It’s the hidden costs of maintaining this quick little thing someone wrote that add up.
The point being, in this new age where code is free, it is the ability to be inhibitory or tasteful that truly defines a successful software engineer.