Is someone building something like this for AI?

One flaw that AI has right now is the lack of full context. Not being aware of the whole context makes the AI err because it simply doesn't know better. The more context you give the more correct the output is. Easy right? That's why stuff like debugging is hard right now for AI because so many things can be going on.

When I worked at Rolls-Royce, I collaborated in a project that ended up in an article titled "Functional requirements-based automated testing for avionics" (link down below) where I played around with SPARK, which is Ada language but on top of that it allowed you to specify functions input/output contracts. Essentially you could specify your function inputs and outputs very well.

Then other people in the project created a mathematical model and software which ingested such contracts and then with all that data it could create tests cases with high accuracy.

Today I was wondering, could this sort of annotation/contract help AI become better? Or do you think it isn't needed? Kind of giving some sort of annotations to the code or whatever task is at hand to give better hints to the AI.

I know AI agents are meant to do specific tasks but I haven't played with them enough to see if they also have this aforementioned problem.

Does anyone know if someone is doing something like this? Or want to share any insight about this?

Article link: https://www.researchgate.net/publication/318223993_Functional_Requirements-Based_Automated_Testing_for_Avionics