[tomaytotomato]

As a senior dev I would say there is a risk of muscle memory atrophy, but the learning opportunities outweigh the risk.

Today I learned about a more elegant helper method in Apache Commons' StringUtils library for Java.

The function was `trimToNull()`

Normally I would have just done

    if (StringUtils.isBlank(foo)) {
        responseDTO.setFoo(null);
    }

Now I can just do responseDTO.setFoo(trimToNull(foo));

I had written the original code, Claude suggested the improvement.

I enjoy shipping code and reviewing what Claude writes.

To add to that, what I find most helpful is the boring stuff, the JIRA cleanup, trawling Wikis and other sources to find out what the historical context of something was.

Normally that would take me all day to do, with a 30 minute code change.

Now I can do that in about 15 minutes and think about building or shipping some tool which I never had time to do.

[beej71]

One thing I tell my students is that after they've coded something up and it's working, they can give it to an LLM and get a code review. Code reviews have always been beneficial, and AI gives us a chance to get them a lot more frequently. I also tell students to be critical of the response that the LLM gives them; be opinionated and have a rationale.

But in almost all classes, I tell them to not use LLMs to generate the solution up front. This ties into what you were saying about the boring stuff. If you know something like the back of your hand, the knowledge gain by rewriting it for the zillionth time is minimal. And so the loss from using an LLM to generate it is also minimal. But if it's something new and you're punting to the AI, you are sacrificing skills gain.