How AI assistance impacts the formation of coding skills

A few days ago, Anthropic published a study on how new developers progress using AI assistance vs not using AI.

Now you might think that the guys selling the shovels for the vibe coding AI gold rush might want to skew the data in their favor, but here are the results in their own words:

The AI group averaged 50% on the quiz, compared to 67% in the hand-coding group.

So those of us who choose to do it the old-fashioned way still have an advantage over delegating our thinking to AI.

“Old-fashioned” is relative, as software and software engineering are newer professions in the grand scheme of things.

It doesn’t end there though; they found that the developers who were allowed to use AI fell into 2 distinct categories that had very different scores.

The developers who heavily relied on AI and delegated debugging to AI scored 40%.

Contrast that with developers who used AI more like Stack Overflow (That is my interpretation) and copied and pasted what AI had given them, while still doing their own debugging, score 65%.

They go into more detail in the paper, but that is how I read it.

What does this mean for SWEs?

Nothing is completely concrete, and I am glad to see that they are going to continue to study this.

With that said, it sounds like using AI like a tool, or how Stack Overflow used to be used 5-10 years ago, while still taking the time to understand at minimum and do your own debugging, gives you an advantage as far as comprehending what you are building.

I would be curious to hear what my peers think about this study. Does this change your opinion on AI-assisted programming for newcomers? If so, how?