Developers should do this, in this order:

• Research how LLMs work. This doesn't take long, there are many videos on youtube that describe how LLMs produce code (Here's a good one: How LLMs actually generate text). In 30 minutes you can have a pretty good understanding of how your prompt becomes code.
• Once you know how LLMs work, decide whether you want to produce code generated by the next most likely token (usually a word - a bit of language syntax), based on code that the LLM has seen before. Wait, does the LLM have to have seen the code before? Yes. It's not actually intelligent; it can't infer anything like a human would. It will respond using the next most-likely sequence of words. If your problem is novel then it won't have seen a good answer to your question before. Now, if you asked your colleagues this question and they don't know the answer then they'd hopefully respond with 'I don't know'. On the internet though this never happens. People don't (usually) post replies on Reddit saying 'I don't know' because that would be stupid. An LLM will never tell you it doesn't know the answer to your question because that is never the most likely response. What happens is you get the most likely sequence of words back. For problems that have been solved a million times this can work well; you get the sum of all human knowledge, averaged and distilled into an appropriate format. The problem is when the sum of all human knowledge is zero, or close to zero, or prior responses to the problem have been ideas, untested and unproven. An LLM will respond in the same way it does when a million programmers all agree, with absolute confidence. This is very different to a human interaction.
• If you decide to go ahead then check what the LLM tells you until you've established some knowledge around how often it can be trusted.
• Decide whether or not to continue based on how much of the generated code needs to be rejected, or fixed, or re-prompted

That's it. My trials revealed that LLMs (all of them) produce sufficiently high levels of garbage that I won't use them at all but depending on what kind of development you are doing your mileage may vary!

PYRAMID has been developed by the UK Ministry Of Defence to provide a consistent approach to modularising air system application software though the PYRAMID Reference Architecture (PRA).

The PRA encourages the well established software design principles of low coupling and high cohesion. Separation of concerns makes for reusable software components, and in an industry like aerospace being able to reuse software can have massive savings in time and money.