If we use GPT-X in our company, it will help us with 1% of our workload

I think there are many such cases. Another one that comes to mind is adding features to a large/legacy code-base. Writing the new code/function is a small part of the work. The main part of the work is first understanding and agreeing on how/where to implement the changes, sometimes across multiple teams, and the implications/knock-on effects to other software components, potential API changes, updating test suites, etc...