Meta, OpenAI, and Uber launch proactive AI agents that speak first

What's Happening

• Meta released Meta Muse on September 8, a personal agent that operates with the application closed to book appointments, draft emails, and request user approval via WhatsApp, according to MarkTechPost.
• OpenAI introduced OpenAI Dots on September 29, deploying always-on agents that conduct read-only monitoring of applications like Slack and Microsoft Teams to flag forgotten invoices or software bugs.
• Uber announced a hands-free voice version of its driver assistant on September 24, which turns live marketplace signals into advice to direct idle drivers to higher-demand zones.

Why It Matters

The shift from reactive chatbots to proactive agents requires developers to solve interruption management. Excessive notifications prompt users to mute applications, while delayed messages render information useless. AI builders must calculate the exact moment a message's value outweighs the cost of distracting the user.

The Context

• Traditional chatbots rely on a pull interface where users choose the moment, the channel, and the question for each interaction.
• Large language models generate the text for proactive messages but serve as the wrong tool to decide whether to send them.
• Notification and growth teams historically manage user interruptions using uplift models, contextual bandits, and receptivity models to predict engagement.

What They're Saying

• Uber's product team stated their driver assistant principles prioritize staying "always on the driver’s side" while surfacing opportunities proactively.
• MarkTechPost notes that every proactive message represents a bet where the "expected value to the user exceeds the cost of interrupting them."

What to Watch

• Developers plan to integrate specialized decision models like TypeSafe's Jev and Supersonic Labs' Julia 1 to handle notification timing.
• These decision models return typed judgments rather than text, allowing them to run on standard CPUs in under 35 milliseconds per decision.

The Bottom Line

As AI agents gain the ability to initiate tasks, successful platforms must master the timing of user interruptions.