[varsketiz]

I'm really curious in what problems the codebases of startups of today will have in a few years. The internet is already full of memes about working with legacy code. What will be the legacy codebases where half of the code is generated with AI tools?

[Quekid5]

Sooooo much boilerplate and pointless generated tests is my prediction.

EDIT: Oh, and as a sibling poster mentioned: A huge number of security vulnerabilities -- except now they can be purposefully injected by just posting random subtly-wrong code on the interweb. Not that you couldn't do that before, but reach would be much more limited unless you got your 'seemingly correct' code posted on SO.

[tom_m]

I think about this a lot. Then I realize many are already bad to begin with. So they may not be much worse as it turns out. We'll have to see.

One thing is certain though - it's an amazing time for security professionals. We already have careless developers without AI, but now? Oh boy oh boy.

[tough]

and 99% of them end up in startup cementery anyways, does it matter much if those codebases stink a lil more?

[tom_m]

True, but before they end up there, they can expose PII and all sorts of nasty stuff.

It does matter.

[trowawee]

They'll be trash, but after a decade bouncing around startups, that's not exactly a problem unique to LLMs. There's probably going to be more startups with more trash than there used to be, but hey: that's job security.

[jryle70]

Are you sure that, when it reaches that point -- half of the code being AI-generated, AI won't be as good as half of the developers?

[varsketiz]

Interesting question. Could you elaborate in the definition of "AI is as good as half of the developers"?

[layer8]

Are you sure that it will?