company

2026-11-10 17:30

Just two months after taking the helm of Apple, new CEO John Ternus can’t sit still.

Last week, in a closed conference room at Apple Park, he called across from Craig Federighi, a senior vice president of software engineering for more than a decade.

According to Bloomberg reporter Mark Gurman, Ternus issued an unprecedented stern reminder to the software team in a closed-door meeting:

Apple has fallen far behind its competitors in AI. It must embrace AI with the same urgency as its competitors, accelerate its roadmap, and quickly move beyond the stage of "just making necessary fixes to Siri."

On one side is the new leader's deep anxiety about the loss of the ecological moat, on the other side is the newly launched beta version of iOS 27.3 by the Bank of China on the other side of the ocean - Siri for Bank of China users has finally changed from the glass ball at the bottom that has not changed for thousands of years to a marquee.

But in the notes to Apple’s official press release, the next-generation Siri AI is temporarily absent from the same treatment as in markets such as the European Union.

And even in those areas where Siri AI is already available, the actual experience is quite bland, mostly staying on screen summaries and single-step instructions, and it is still far from an agent that can truly help users do things autonomously.

This is exactly the passive situation that Ternus desperately wants to reverse.

An answer sheet two years late

In order to promote Apple’s smartphone entry into China, Tim Cook, who has just been transferred to executive chairman, has been traveling frequently in the past two years:

In the autumn of 2024 in Beijing, when asked by the media, he cautiously stated that he was working hard to advance the specific regulatory process; in 2025 in Suhewan, Shanghai, the focus of his topic became "how to truly integrate AI into daily systems."

After visiting the "Blue Box" in Taikoo Li Qiantan, Shanghai last month, Cook finally confirmed in person that the first batch of functions had been approved for implementation in an exclusive interview with iFan'er.

From repeated communication and compliance to the selection of Alibaba Qianwen as a cloud model partner, Apple's "two-year period" has finally arrived.

But in Ai Faner’s actual hands-on experience, the experience of using the Bank of China AI provided by Apple this time is hard to describe - due to the lack of an active interaction framework at the bottom of the system, Siri calling Qianwen is more like a blunt system plug-in, and the experience cannot widen the gap with third-party AI Apps.

While various domestic Android camps and mobile phone AI assistants have begun to try long-chain operations such as ordering takeout, checking tickets and hailing taxis across applications, Apple's plan for the world's largest iPhone market is still two years ago, staying in the old logic of "single request, single feedback".

Two-sided Ternus

Last week, Apple released its fourth-quarter fiscal 2026 results. On a conference call with analysts, Ternus delivered his first report card as CEO, speaking in a calm and dignified tone.

When talking about Siri AI, he was not shy about his affirmation, telling analysts that Apple "finally has the Siri that I have always wanted."

When asked about the more autonomous Agent route, he responded flawlessly, emphasizing Apple's prudent philosophy: users must know what permissions they have given to AI, and Apple will never sacrifice the bottom line of privacy in the blind pursuit of autonomy.

But in the audience, in the conference room mentioned at the beginning, the new CEO, who has been in charge of hardware engineering for more than ten years, is running out of patience.

Hardware research and development is accustomed to a 3- to 5-year cycle, a highly anchored deterministic route from chip manufacturing to industrial design; while Federighi, who has been in charge of software engineering for more than ten years, has been more inclined to develop on a rolling basis in small cycles of 12 to 18 months, and has long been cautious about the safety and illusion of generative AI.

Bloomberg revealed that Federighi has an extremely strong style in the software team and is accustomed to using technical details under his control to push his position in meetings, which makes it difficult to implement many more radical explorations within the team.

When the entire external ecosystem of mobile phones and large models shifted to the Agent route of "taking over the screen and doing the work independently", the software team's step-by-step patching rhythm created a sharp temperature difference with the external competition in Ternus's eyes.

The speeding gear and the boundary of privacy

In fact, it’s not that Apple hasn’t thought about taking this step. At WWDC26, Apple demonstrated a prototype that is closest to the future Agent form: after the system password management tool identifies a weak password, it can autonomously launch Safari to access the target site, automatically complete the modification, and write the new password back to the keychain.

But this feature was briefly revealed in the beta version and then quietly removed. Its suspension is actually a typical exposure of Apple's hesitation in making intelligent agents. In the past decade or so, iOS has relied on a very conservative order—each app is kept in its own security grid, and users must click and confirm every step personally.

And for intelligent tasks like "automatic password change", the system needs to directly take over the browser for you, shuttle through unknown web pages to identify forms, click submit, and even handle abnormal pop-ups. Once the target website is slightly revised or the loading is stuck, it will be difficult for the process to close the loop independently.

For Apple's software architecture, which is accustomed to "when a person sends an instruction and the system takes a step," it is far more complicated than expected to cross the threshold of letting AI continuously perform tasks for users.

To break the deadlock, Apple's internal engineering rhythm is being fine-tuned.

Bloomberg revealed that in the development of the next-generation system code-named iOS 28, Apple will officially adjust the "4 weeks of coding + 4 weeks of system integration" model that has been used for many years in the first phase (M1) to "6 weeks of coding exploration + 2 weeks of system integration."

The obvious shift in focus is to free up time for bottom-level engineers to experiment freely, forcing the team to explore radical Agent architecture.

Actions to open up the ecology to the outside world are also following up. The day after the financial report, Apple's official developer website officially launched new tools for iOS 27.3. In addition to deepening App Intents, it also launched a new "Skills" framework, which attempts to package complex workflows into modular building blocks and use the power of developers to fill in the shortcomings of system-level operating capabilities.

In its own iWork suite and Creator Studio, Apple is also paving the way for Agent in advance and introducing support for the MCP interface.

However, what is more difficult to overcome than reconstructing the development process is the foundation on which Apple is based - privacy.

For a long time, Apple has had a kind of restraint in AI that makes it difficult for competitors to imitate: For example, in the audio summary (Siri Recap) function of the overseas version of Siri, the system promises not to retain any original recording files, all translated texts have the speaker’s name removed, and only a structured summary without identifying information is generated.

This kind of "anonymous, no original" end-side processing forms the basis of trust for users to dare to keep their mobile phones aware.

However, the underlying logic of autonomous agents needs to be based on highly coherent and detailed personal "long-term memory."

If even the user's time, what they said to whom, and which restaurants they used to go to must be physically erased, then a fully autonomous manager who can order meals for you in advance and plan vacations will be logistically impossible.

How far must the system operate for you before you are willing to truly hand over control of your phone? This is not only a choice in engineering implementation, but also a trust test that Apple must answer clearly to users.

Two years ago, when Cook first pronounced the name "Apple Intelligence" at WWDC24, Apple described a decent balance to the outside world: having the wisdom of a large model without sacrificing personal privacy.

Now, it is stuck on the threshold from "single interaction assistant" to "real agent", testing back and forth between the boundaries of privacy and convenience.

Ternus stood on the edge of this threshold. When he closed the door to the conference room and asked Federighi to step on the accelerator and turn the steering wheel toward the more aggressive Agent, Apple had actually made a choice: The era of prudent defense and small jogging to win the house was over.

To create a next-generation smart device that truly understands you, Apple has to take a big step out of the safe zone it has been stuck in for more than ten years.

Zheng Tingxu

Hardware editor

Could be better.

A total of 333 articles have been published

This article comes from the column

Previous article2026-11-10 17:30

With both "eyesight" and "cleanliness", what has the Dyson smart sweeping and mopping robot changed?

Aifaner App

Love style, make the future within reach

Digital trendy brands focusing on tomorrow’s products

Advanced tools, prophets first, AIGC’s inspiration guide

A new era of car manufacturing, tomorrow’s travelers

Explore new lifestyles in the city and be your guide to tomorrow’s life

Copyright © Guangzhou Liwo Zhiyuan Investment Management Partnership (Limited Partnership) 2008 - 2026. Use of Aifaner website content for commercial purposes requires permission. Non-commercial use is licensed under CC BY-NC 4.0.

All content is made available under the CC BY-NC 4.0 for non-commercial use. Commercial use of this content is prohibited without explicit permission.