Ask HN: What's missing for a true deterministic abstraction layer?

If I wanted to develop a system but in such a way that implementation is abstracted away but the same spec always yields the same system, so we as humans still understand the system and not it's implementation: what is missing? What is stopping us from such language, visualization or construction method?

Llms are not deterministic and can implement the same spec in diferent ways or slight variations, how can we get deterministic approach so we can gain development speed and accountability over our systems? Can something like that exists ?