Apple speeds up agentic Siri for iOS 28 and widens China rollout

Siri heads toward autonomous tasks

Bloomberg's Mark Gurman reports that Apple wants Siri to complete tasks without a user starting each step, with the first such abilities possibly arriving alongside iOS 28 next year. He points to rivals such as OpenAI's Dots, Meta's Muse and Grok, which can track goals and keep working toward them. According to the reporting, CEO John Ternus has pressed software chief Craig Federighi to speed the schedule and move past the improvements Siri already needs. Apple has started a structural review of its development process for iOS 28, giving the first phase six weeks for design and coding instead of four, and shrinking technical convergence to two weeks from four. An opinion piece in 9to5Mac warns that agents can fail at a task or finish it in inappropriate ways, which makes faster delivery riskier.

Design and coding, traditional model4

Design and coding, iOS 28 first phase6

Integration, traditional model4

Integration, iOS 28 first phase2

Design and coding, traditional model

4 weeks

Design and coding, iOS 28 first phase

6 weeks

Integration, traditional model

4 weeks

Integration, iOS 28 first phase

2 weeks

Data from the article text

Pressure inside the software team

Gurman says the push followed a closed-door meeting in the weeks before his report, where Ternus was frustrated by what he sees as reluctance to embrace AI with the urgency of competitors. Federighi is known for planning only 12 to 18 months ahead, which makes long-term goals hard for Apple to set, and some engineers reportedly feel they have little say in product direction. On the product side, Gurman expects fine-grained control over Apple's own apps, such as rotating images in Photos or organising Safari tabs, with iOS 27.3. Integration with third-party apps, such as ordering through Grubhub or booking with OpenTable, is expected later, in the spring.

Leadership questions around Siri

Mike Rockwell, the vice president of Siri engineering, is still weighing whether to step down and might leave Apple next year, Gurman reports. Rockwell built the Vision Pro before taking over Siri's turnaround after a two-year delay that Gurman called embarrassing, and reports of his exit first surfaced in April. If he goes or changes roles, Siri would most likely pass to Amar Subramanya, a Google veteran hired last year to strengthen Apple's AI work, or to Sebastien Marineau, a deputy to Federighi who leads the teams bringing AI into iOS. Some inside Apple see Marineau as a possible eventual successor to Federighi. Gurman calls a succession to Federighi a distant prospect, noting that he has led software engineering for more than a decade and has given no sign of weighing a departure.

China rollout and open questions

On 9 November Shanghai's cyberspace regulator added iPads and Mac laptops as two new uses in Apple Intelligence's filing, a step China requires before AI services reach the public. The iPhone approval came in mid-July, when Alibaba said its Qwen model would power the service on Apple devices in China. Alibaba's statement to CNBC in July read in part:

“Qwen will be integrated into Apple Intelligence experiences within iOS, iPadOS, macOS, and visionOS for users in China.”

At launch Qwen runs Siri, Writing Tools and visual intelligence, while Siri AI and other tools shown at the June developer event are excluded. Apple said on 31 October that the first features would arrive with iOS 27.3, a test version for app makers went live on 4 November, and the full release is due in December. Apple's October list covered only iPhones, including the iPhone 15 Pro, 15 Pro Max, iPhone Air, iPhone Duo and models from the iPhone 16 onward, with Siri and the phone set to English or Simplified Chinese. Apple has not said when iPads and Macs will get the features.

An unfinished memory system

Code explorer pdfu found a personal memory system in the iOS 27.3 beta that could let Siri retain facts about users' lives. The memory graph would organise relationships, places, routines, health and motivations. Only one AI-driven process, which uses 60 built-in questions to find motivations, barriers and resources, appears present so far. No process creates memory atoms yet, and nothing writes to the separate store for explicit memories. Nothing uncovered so far gives a release date or explains how users would control their data.