AI and Automation Operations: It Was Working, Nobody Could Tell

It was working. Nobody could tell.A twenty agent real estate brokerage automated part of its nurturing sequence. Every message was delivered, correctly formatted, on schedule, with every field populated from the right record.Nothing in the system reported a fault.What it had stopped carrying was what the agent knew about the family. That had never been written down as a step, because to the people doing it, it was not a step. It was what you say to someone you met on a Saturday.Automation almost never fails at the task. It fails at the check. And whether anyone can run that check depends on whether anyone left in the building could still do the work themselves.What is in it- Three conditions that decide whether a task is eligible for automation, and the fourth that decides whether it stays eligible- Why the tasks people most want to automate are the ones that qualify least- A 1983 control engineering paper arguing the supervisor of an automated process must be more skilled, not less- The randomised trial in which expert developers were nineteen percent slower and believed they were twenty percent faster- What the forty seven percent employment figure actually says, and the backtest it failedWritten against its own author where the evidence required itOne chapter carries a controlled result that contradicts the author's position on expertise, presented as a correction rather than buried. One credits its central mechanism to a paper the author had not read until this book was researched, and says so in the text.