Using LLMs to build out the nice-to-haves that I’ve always wanted but never had time for is one of their great use cases. Visualizations are a perfect use case for this because they don’t have to be perfectly architected, maintainable code. Getting to the correct visual output is good enough, and LLMs excel at iterating something until it looks right.

I agree that visualization does not need to be perfect. One issue is that "correct visual output" depends on your expertise level. A visualization that is good to teach undergrads may be frustratingly bad to experts and researchers. Standards like "looking right" depend on the audience's ability to spot nuances and how focused they are on the fine details. If you want a visualization to work for the range of people from beginners to experts, you do need to focus a bit more on what it means for something to "look right" for multiple audiences, since the errors in the fine details may hinder a visualization's usefulness for more advanced audiences.