An inherent flaw in transformer architecture (what all LLMs use under the hood) is the quadratic memory cost to context. The model needs 4 times as much memory to remember its last 1000 output tokens as it needed to remember the last 500. When coding anything complex, the amount of code one has to consider quickly grows beyond these limits. At least, if you want it to work.
This is a fundamental flaw with transformer - based LLMs, an inherent limit on the complexity of task they can ‘understand’. It isn’t feasible to just keep throwing memory at the problem, a fundamental change in the underlying model structure is required. This is a subject of intense research, but nothing has emerged yet.
Transformers themselves were old hat and well studied long before these models broke into the mainstream with DallE and ChatGPT.
You haven’t really described what you are imagining.
Telling a computer specifically what to do and how to do it without making mistakes is coding. Programming is a level above that, in designing the architecture of how to approach the business problem.
What the other commentator is saying, is that simple being able to tell some model ‘build an app that does XYZ’ requires AGI because that set of instructions is not complete - the machine requires outside knowledge and the ability to make judgement calls in order to complete it.
If that isn’t what you meant, it is at least what you said. The breakdown in communication here, between humans, should also serve as another reminder how difficult it is to convey an idea to another entity and how that problem will remain difficult for a very long time.