Main idea

In a highly specialized organization, the person who could act on a domain was also the person whose domain it was: the developer wrote the code and answered for it, the designer drew the interface and answered for it. Capacity to act and responsibility overlapped, so it was rarely necessary to say who answers for what.

When an AI lets a salesperson build an internal tool, a Product Manager change an interface or a developer settle a product behavior, the capacity to act moves between professions. Responsibility does not move with it: the salesperson who builds the tool does not become responsible for platform security, the PM who touches the interface does not become the guarantor of the design system. Being able to act and answering for the domain, which used to go together, come apart.

Why it matters

It explains why the blurring of professions does not remove roles but forces responsibilities to be made explicit: the overlap no longer protected them, they now have to be named. And it gives a simple test when a contribution comes from elsewhere: the question is no longer "who is allowed to touch what?" but "at what level of risk or novelty must the domain's guarantor take back control?"

Nuances and limits

The overlap was never perfect: even in a highly specialized setup, some responsibilities floated between professions. AI does not create the separation, it makes it visible and frequent. And making responsibilities explicit is not enough: if every contribution has to be validated by the domain concerned, the old mandatory checkpoints are recreated under a new name.

Open questions

• What level of risk or novelty should be set for a domain to take back control, without that threshold turning into systematic validation?