Meta, OpenAI and Uber Just Taught AI Agents to Talk First. What About When to Stay Quiet?

AI agents that initiate contact with users represent a shift from traditional chatbots, which users controlled by choosing when and how to interact. Companies are now deploying always-on agents that proactively reach out across multiple channels with suggestions, alerts, and opportunities. The core challenge is determining when to interrupt users, as sending too many messages causes users to ignore them while sending too few means missing time-sensitive opportunities. Rather than using language models to decide whether to send messages, companies should use decision models that evaluate whether the expected value to the user exceeds the cost of the interruption, considering factors like urgency, likelihood of user action, and which communication channel is appropriate.