Meta, OpenAI, and Uber are developing proactive AI agents that initiate communication with users. These systems must determine the optimal timing and channel for messages to avoid user fatigue or missed opportunities. The companies are focusing on balancing the value of a notification against the cost of interrupting the user.
New decision models are emerging to manage these interactions. For instance, the Julia 1 model from Supersonic Labs operates on CPUs with a decision latency of approximately 33 ms. These models provide structured judgments, such as scores or binary choices, rather than generating text, allowing agents to filter triggers before engaging an LLM.
The selection of communication channels is tiered by urgency and intrusiveness. In-app cards are used for low-priority updates, while SMS and voice calls are reserved for time-sensitive alerts. Uber utilizes its outcome-tracking data to refine these decision-making processes.
Meta is integrating commerce features into its Muse agent, while OpenAI has introduced the Dots interface alongside a $500 monthly subscription tier. Success in this field depends on the ability to prioritize user value over promotional frequency to maintain credibility and prevent user opt-outs.
