I see several key groupings for ways tech companies operate with LLMs:

  1. human-written code that was reviewed and audited by humans • slower & potentially insecure
  2. human-written code that was reviewed and audited by LLMs • slower & more secure
  3. LLM-written code that was reviewed by no one • faster & potentially insecure
  4. LLM-written code that was reviewed by humans* & LLMs • faster & more secure

I think the gap will widen between the organizations that don’t use LLMs and those that do both with regard to productivity as well as security.

And any company ignoring the skills and importance of the humans involved I suspect will also end up with problems. If not immediately then eventually.

* companies are currently mostly working with people who already have coding skills running LLMs – this would shift drastically if for some strange reason you assume the average person is going to get the same results out of an LLM. People who don’t have any coding experience or institutional knowledge are not the same as someone who does. I think I’m already seeing this being ignored by people in tech in select cases.