Looking at the code can tell you whether it was “implemented properly” - or at least, if it was done reasonably competently.
Relationships, though, can help you know if that properly implemented code was likely a fluke or not. Does the person ask questions? How do they ask questions? How do they communicate? How do they capture requirements? How do they push back when something seems unclear or unwise? etc.
Now, I've seen plenty of "demo-ware" software that has been presented as something that was much more than it actually was, but again, stuff like that always falls over, riddled with bugs and crumbling under load, when it gets substantial use.
Some things you really can't fake, and software engineering is one of them.
Input lag on devices has doubled or tripled from the 1980s while clock speeds have increased drastically. Reddit and Hackernews do essentially the same task, yet Hackernews' load times are almost instant while Reddit takes more than a second to load at times.
It's true that you can't fake a minimum viable product, but it seems to me there is a long way from that to actually making something good.
Noone wants to write low level code because learning it costs money, so instead we just throw more hardware at the problem until it goes away...
Human time is expensive, so it makes sense to throw anything cheaper at the problem first.