Don' think SoC compliance is as automatable as much as investors hoped to. This mistrust and over trust in AI is based on a technology that Google invented and didn't pay much attention to themselves because they knew it isn't as reliable or that useful to the point where its output is so definitely reliable that it requires zero human input.

The coding agents succeeds because apart from wanna be SaaS indie vibe coders, other serious users of AI agents for coding are themselves pretty strong and competent software engineers that won't let slip things easily and have years of experience and a taste in what is architecturally correct and what is nonsense and when and how to steer in what direction.

Other fields - if they have to review every output of the LLM such as in finance running totals and such to verify the results of an LLM makes their usage not as much useful.

It's fully automatable. The secret ingredient is fraud.