“The experience of building a customer education program with AI agents at ThoughtSpot showed that the agents could not be fully autonomous and required constant human-in-the-loop oversight to correct errors and prevent model drift.”