What are your thoughts on checking the AI's process and output? I can see junior professionals doing some of this manual work, as a way to improve their skill.
Explainable AI can be another concern. When you come across something that seems bizarre, you need to be able to go back to figure out how it got to it's conclusions, and tell it to do it over, but better this time.
When AI can speed up tasks, the Human Overseer checking the output is the bottleneck. This seems to be the very definition of slowing down to speed up.
We can't completely trust AI to do everything without oversight.
I think this is part of the "AI learning + trust curve".
At the beginning, you automate and have a human in the loop to see the results. You correct when the result is off. Eventually, with multiple cases and corrections, you remove the human in the loop (and can change the approach to "alert me if there is a deviation" mode).
For this initial part, you will likely still need an experienced PM.
As an example from the development world, there are companies now using code review agents that check the automatically generated code from other agents, and only involve the human when they find something odd.
What are your thoughts on checking the AI's process and output? I can see junior professionals doing some of this manual work, as a way to improve their skill.
Explainable AI can be another concern. When you come across something that seems bizarre, you need to be able to go back to figure out how it got to it's conclusions, and tell it to do it over, but better this time.
When AI can speed up tasks, the Human Overseer checking the output is the bottleneck. This seems to be the very definition of slowing down to speed up.
We can't completely trust AI to do everything without oversight.
Good point!
I think this is part of the "AI learning + trust curve".
At the beginning, you automate and have a human in the loop to see the results. You correct when the result is off. Eventually, with multiple cases and corrections, you remove the human in the loop (and can change the approach to "alert me if there is a deviation" mode).
For this initial part, you will likely still need an experienced PM.
As an example from the development world, there are companies now using code review agents that check the automatically generated code from other agents, and only involve the human when they find something odd.
Well done!