My take is that this is largely right. To be specific though, I feel the future is natural language commands that will be translated into a proper language through llm systems.

In this way, the generated output, whether used as a proper program or just some scripts to do small activities, they can be audited later, reused, and tested even if a client created the request in plain language.

We would then mostly be translating bugs from english to the underlying language and bugs would mostly be people speaking ambiguously or just misspeaking