Atrioc

Last time we talked about Anthropic doing this study with programmers who learned a new Python library using AI.

And the ones who used AI didn't learn it as well.

Cool.

couldn't refer back to what they had learned.

Like the understanding was worse.

So this is really concerning in a world where we have AIs that are increasingly like, oh, I can do everything for you.

80% of your code is just tell me what to do, but you're actively losing the critical thinking skills.

So then there's another from Modern Software Engineering, a YouTube video, where they did a study about the maintenance of code.

And the idea is, look, writing code is, getting the first draft done is a small portion of the overall work it takes to do software.

The maintenance of it after you've made it is, in his estimate, three to five times more than the lifetime, over the lifetime of the software.

Some quotes I liked.

It's largely naive nonsense to imagine you can develop software once and then never revisit it.

Yet most AI studies stop at, did the AI software developer finish faster?

And they just kind of brag about the speed and quality of what's written and don't think at all about, is this even useful in the longterm?

And again, with all of this stuff, you can start to imagine how this applies to any industry, right?

Not just programming.

So they did a cool study.

Phase one, they had a whole bunch of programmers add features to an app, some with AI, some without.

Phase two, a second group of programmers came in and tried to improve the code from phase one, but they didn't know whether it was AI written or not.

So the question is, is it harder to maintain AI software?