A workflow that requires people to work somewhere new is a change-management project wearing a technology costume. A workflow that reaches into the tools they already open is just a better version of their week.
Connectors are also where the security surface is. Each one is a set of permissions granted to something that acts on its own, so scope them narrowly, prefer read where write is not needed, and log what they do.
The unglamorous truth is that most of the build time on a creative automation engagement goes here, not into anything anybody would call intelligence.
Why do internal AI tools fail to get adopted?
Most often because they live somewhere new. A system that reaches into the tools a team already opens is adopted by default; one that asks them to move is a project.
What is the security posture for a connector?
Narrow scope, read-only where writing is not needed, and a log of what it did. It is a permission granted to something that acts without being asked each time.