That is, instead of just showing the user the code it wrote, is there a better way to show the program at a higher level that a non-programmer would understand (without just running it to show what happens)?
That would entirely depend on your audience. Some people are more capable of receiving and following instructions than others. Some people are faster to anxiety at proximity towards uncertainty than others.
You can’t, because the reason the code exists is because you need its formality in order to eliminate ambiguity. Everytime code is explained without showing the code, there’s a baseline of understanding that’s required to filter the additional ambiguity.<p>The only approach left is to refer to the domain. As the person is familiar with it, you illustrate the behavior that is currently programmed by the system. The program itself should be a blackbox. Only knobs and sliders should be left (aka any professional software).
What value is there in showing a non programmer code or an abstraction?<p>If you want to vibe code tests are likely a guardrail to ensure quality over time
I'm having the LLM work with human-readable behaviors for each component. You can have the LLM take those behaviors and create a Mermaid diagram, for instance.