Imagine you last non-technical manager trying to prompt something like "Design a full stack for an API that will unify financial APIs" and watch it probably do nothing meaningful while they get frustrated that they aren't meeting ramp with their few sentences.
This plainly isn't true ("can't make anything useful"), given the examples we've seen. It might be buggy, or not integrate so well (or at all) with other stuff you have. But it obviously can generate code that does useful stuff.
Lack of shell skills is a non-issue (anything that can be done in the shell can be done over a webform).
OP is right, nothing will really change as long as there's no tool that can turn a Jira ticket into a pull request.
Automatic programmer: GPT-4 powered Github Issue solving with Pull Requests Project
I've created a tool that reads Github Issues and creates a Pull Request to complete it. It uses GPT-4, of course. There are a bunch of reasoning steps the tool does, but it can be much, much better with better steps and prompting. It's just a Proof of Concept. I haven't even added Langchain, yet.
https://www.reddit.com/r/OpenAI/comments/11upow0/automatic_p...
And here's another: https://github.com/irgolic/AutoPR
You can keep shifting the goalposts if you want - "but the code will be buggy", "but it can only do simple stuff, doesn't understand the broader systems architecture" - but all indications are that, on a plug-and-play level, GPT is able, to at least some extent, to do what non-techies think of as "programming".
It seems you are also underestimating what non-technical managers can do. Yes, some are utter dunces and technophobes. But most are not and can run basic shell scripts and even know what "files" are.
Both of these level of engineers are affected.