Siri’s Next Phase: Apple Explores Agentic AI That Acts Without Prompts

Apple is quietly laying the groundwork for an agentic AI system that could let your iPhone, iPad, or Mac act on your behalf without waiting for a prompt at every step. The signal comes from Bloomberg’s Mark Gurman, who predicts the company is actively exploring this direction after comments made by Siri engineering chief Mike Rockwell following the recent developer conference.

Rockwell’s framing was careful but pointed. He acknowledged that Siri today is fundamentally request-based: you ask, it responds. What changed, he argued, is that the assistant now runs on a completely modern architecture, and that foundation makes expanding into agentic AI highly possible rather than speculative.

What agentic actually means here

An agentic system operates on a continuous loop: it takes in information, makes decisions, and executes actions without requiring a button press for each step. That is a meaningful departure from the current setup, where Siri handles one request at a time. The gap between those two models is not cosmetic.

Craig Federighi, Apple’s senior vice president of software engineering, offered a measured counterweight. He described this category of AI as still experimental and said finding the right user experience is the current priority. He stopped short of ruling out a future release, which is about as close to a non-denial as Apple typically offers. For context, how that experience eventually lands on devices matters: on-device processing constraints could shape what any agentic Siri is actually capable of at launch.

The privacy dimension is harder to ignore. A system that independently operates your device, reads context, and takes autonomous action sits in very different territory than a voice assistant that sets timers. Apple has spent years positioning privacy as a differentiator, and how that architecture handles autonomous decisions across apps and data will matter as much as the feature itself.

The current Siri runs on a large language model but stays within the boundaries of direct prompts. Closing the distance between that and fully automated task execution across an entire device is the actual engineering challenge Apple has not yet publicly committed to solving.