company

2026-11-20 09:17

Siri may finally be evolving from "mouth" to "hands".

Asking Siri to check the weather, set alarms, and open apps is nothing new. But if you now ask it to arrange a business trip, check the air tickets and hotels, and then organize the itinerary according to the calendar, I'm afraid you still have to do it yourself.

Apple's recent series of actions reveal its ambition to change this situation.

In mid-November, two job postings related to autonomous AI Agents appeared on Apple’s official website.

Swipe left to see more

One of them describes an intelligent agent that can operate computers like a human, and the other even directly names Claude Code and Codex, requiring candidates to use these tools on a daily basis.

According to Bloomberg, Apple is developing its own autonomous AI Agent, hoping to allow artificial intelligence to perform tasks across applications and even plan workflows autonomously.

Picture | Meta Muse

But looking at the entire industry, these capabilities described by Apple are nothing new.

Meta, OpenAI, Anthropic, Tencent, and ByteDance are all pushing AI Agents into real work and life scenarios. Some products have accumulated tens of millions of users and begun to generate actual revenue.

For Apple, which has built a huge ecosystem relying on iPhone, Mac and App Store, it may be difficult to continue to take its time in this competition.

Picture | Apple

Apple is finally letting AI do its own work

How far does Apple want AI to go? There is a description in the recruitment information that can almost be used as a product introduction.

Apple is looking for a senior engineering manager to develop an AI assistant with proactive service capabilities. In the job description, Apple describes a vision for AI to operate computers like humans:

Imagine an AI agent that can use a computer like a skilled human. It can browse the web, work across apps, run commands, write and debug code, call services, and choose the right tools for each step.

Recruitment information shows that Apple is developing a general computer operating intelligent system. It can switch between applications, browsers, command lines, code environments, and APIs, plan and execute multi-step tasks independently, and coordinate the work of multiple specialized sub-agents.

Figure | Codex

Taking business as an example, the user only needs to say "Help me arrange a business trip to Tokyo next week", and the AI ​​can check flights, filter hotels, check calendars, organize meeting arrangements, and finally give an itinerary plan. It is up to it to decide whether to check the flight tickets first or the calendar first, and which tools need to be called.

Another job posting focused on the continuous improvement of the agent.

Apple is looking for a senior application AI engineer to be responsible for establishing personal intelligence evaluation benchmarks, improving evaluation infrastructure, and using feedback mechanisms to continuously optimize Agent performance.

What’s even more interesting is that Apple mentioned Claude Code and Codex by name in its job requirements, and even required candidates to use such smart programming tools on a daily basis.

Judging from these requirements, Apple is no longer satisfied with letting AI answer questions and execute simple instructions. It hopes that the agent will find its own way to complete the task and become more reliable with continuous use.

However, once the files, accounts and applications in the computer are handed over to AI, trouble may ensue.

If the user asks the Agent to organize old files on the computer, it may delete important data by mistake; if it can read emails and run code, how should the scope of permissions be limited? This is why Apple repeatedly emphasizes privacy, security, sandbox isolation and permission management in its recruitment information.

The recent macOS update happens to involve adjustments in this area.

In macOS 27.3 Beta 2, Apple has adjusted the authorization mechanism for device control and full disk access permissions, splitting management of different types of system permissions, and allowing users to authorize specific applications and folders. The system also provides visible status cues when AI apps control other apps.

🔗 Related reading: Mac tightens permissions, Apple begins to guard against AI

Users will have a clearer idea of ​​what permissions the AI ​​has obtained and what it is operating on, and the way third-party agents run on Mac will also be affected.

Figure | macOS 27.3 menu bar shows tools that are controlling other applications / 9to5Mac

While adjusting the permission rules of third-party Agents and recruiting people to develop its own agents, Apple is clearly preparing for AI to deeply intervene in the operating system.

The two things must be solved together so that it has enough ability to complete the task, but also cannot allow it to access the user's data and device at will.

As for when it can be used, I'm afraid we have to wait a while.

According to Bloomberg, Apple has considered launching related products as early as the end of 2027, and development is still in the early stages.

In 2026, AI Agent will truly enter the mass market

In 2023, ChatGPT will give ordinary users access to generative AI on a large scale for the first time. Three years later, AI’s capabilities are no longer limited to chat windows.

Figure | Early ChatGPT interface

It is no longer unusual for a model to write a piece of code and analyze a document. What attracts more attention is whether it can continue to run the code, modify it according to the error report, or open the relevant application and continue to do the work assigned by the user.

Meta’s Muse is one of the most talked about products in the consumer market this year. It can connect users’ digital accounts and devices to complete tasks such as travel planning, restaurant reservations, and information organization.

Picture|Muse releases promotional picture / Meta Newsroom

According to a report by "The Information" at the end of September, citing Meta internal data, a few weeks after Muse went online, the number of users actively sending tasks to it every week has exceeded 3 million, and the number of users actively sending instructions to it every day has exceeded 1 million.

Meta later revealed during its third-quarter earnings call that Muse already has more than 10 million users in the U.S. and Canada, about 25% of whom use Muse on a daily basis.

At that time, Muse had not yet been launched globally, but it had already accumulated such a group of users that it was enough for other technology companies to take notice.

Picture | The promotional image used by Meta when announcing the launch of Muse in Australia / Meta Newsroom

At the end of September, OpenAI launched Dots. This is a personal AI Agent that can work continuously. Each Dot has its own cloud computer and can connect to more than 4,000 applications to follow up on the goals assigned by the user for a long time.

Users can ask it to research a project, organize information, create documents, and even continue to advance a software development work. It can continue to perform tasks in the background even if the user temporarily leaves.

Figure | Dots user interface displayed by OpenAI at DevDay / OpenAI

Anthropic enters the software development and general agent markets through Claude Code and Claude Duet respectively.

Among them, Claude Duet can actively organize the information that users need to pay attention to, read authorized accounts, draft emails, arrange meetings, and continue processing tasks after the user leaves. It is currently open for research preview.

Figure | Some of the applications and services that Claude Duet can connect to / Anthropic

Muse, Dots, and Claude Duet all do things differently, but have one thing in common: they start to have their own working environment, the ability to connect to external services, and the ability to continue working on tasks after the user leaves.

Picture | Anthropic

Compared with personal Agents, which are still competing for ordinary consumers, the software development and enterprise office markets have emerged with clearer business returns.

Microsoft disclosed during its fiscal 2026 fourth quarter earnings call that the number of GitHub Copilot users has reached 50 million, and about one-third of Pull Requests on the GitHub platform involve AI Agents.

At the same time, Microsoft 365 Copilot has more than 30 million paid seats, and Copilot business revenue has increased by more than 60% month-on-month.

Figure | GitHub Copilot desktop agent task interface / GitHub

Competition in the domestic market is also quite fierce.

On November 18, Tencent disclosed in its third quarter financial report that the number of daily active users of WorkBuddy has exceeded 10 million. Based on the last seven days of revenue, its annualized revenue run rate exceeds $500 million.

Doubao Work, a subsidiary of ByteDance, has more than 8 million daily active users, and annual recurring revenue from corporate customers exceeds 2 billion yuan.

Both Tencent and ByteDance are relying on their respective office software, cloud services and user ecology to promote Agent into daily work scenarios. Compared with cultivating user habits from scratch, existing enterprise customers and software portals also make it easier for them to find actual needs.

Looking at these products together, several different approaches have emerged in the market. Meta, OpenAI, and Anthropic hope that AI can handle more personal matters; Microsoft puts agents into software development and corporate office processes; Tencent and Bytedance use existing businesses to find room for growth.

Picture |Lemonade

Whoever can receive users' needs most conveniently and then mobilize various tools to get things done will have a better chance of becoming the AI ​​product that users open first.

The financial investment behind this competition is equally staggering.

Meta's revenue in the third quarter of 2026 reached $64.6 billion, a year-on-year increase of 26%, and capital expenditures during the same period reached $34.9 billion. Large-scale investment in AI infrastructure has become one of Meta’s most important strategic expenditures.

Anthropic has just completed its initial public offering (IPO), listing on Nasdaq at a price of $83 per share. The issue price corresponds to a valuation of approximately $1.5 trillion.

Picture | Anthropic co-founders attended the Nasdaq listing opening bell ceremony / Associated Press

Investment in AI infrastructure is still increasing, and the capital market is also willing to give high valuations to leading AI companies. More importantly, Agent products have begun to show user scale, usage frequency and commercial revenue, proving that they have the opportunity to become a long-term business.

What Apple wants to catch up with is already a market with tens of millions of users, real business revenue and huge capital investment.

Apple can’t afford to lose in the battle for next-generation computing entrance

Apple certainly doesn’t lack users, nor does it lack money.

According to the latest financial report for the fourth quarter of fiscal year 2026, Apple’s single-quarter revenue reached US$114.5 billion, a year-on-year increase of 12%; full-year revenue reached US$478.8 billion, another record high.

Figure | Apple’s fourth quarter financial statements for fiscal year 2026 / Apple Newsroom

Among them, the services business’ quarterly revenue reached US$31.573 billion, also a record. Apple's active device installed base has also reached all-time highs across all major product categories and geographies.

But these results cannot guarantee that Apple will continue to take the initiative in the next round of AI competition.

Apple has spent more than a decade building an ecosystem that revolves around iPhone, Mac, iPad, operating systems and the App Store. Users turn on the device, find the applications they need, and then use the applications to complete shopping, office work, social networking, and entertainment.

The operating system is responsible for managing applications, and applications provide services. Users decide which ones to open and how to operate them. Apple controls key links in this ecosystem through software and hardware integration and the App Store distribution system.

AI Agents may make some of these operations redundant.

For example, if a user wants to book a flight to Beijing tomorrow, he directly tells the AI ​​the departure time and budget, and it can search for flights, compare prices, filter options, and complete the booking.

Picture | Siri AI Q&A / Apple Newsroom

Airlines, booking platforms, and payment services still exist, but users don’t necessarily need to open their apps themselves. Operations that were originally scattered in multiple applications can be completed uniformly by one intelligent agent.

If users become accustomed to handling daily affairs through ChatGPT, Claude or other third-party agents, Apple may lose some of its dominance over user interaction portals.

Whoever masters the user's needs will have the opportunity to decide how users use the entire digital ecosystem.

Figure | Apple demonstrates the integration of Siri and ChatGPT / Apple Newsroom

However, Apple also holds many good cards.

Its huge device base, unified operating system architecture, and personal data that can be accessed after user authorization have the opportunity to become its advantages in developing AI Agents.

If third-party agents want to operate the device, they often need to apply for system permissions and then connect to different applications and services. If Apple deeply integrates Agent into the operating system, it will have the opportunity to more conveniently call device functions and complete tasks based on schedule, location and device status.

This integration capability may be more valuable on new hardware in the future.

Take smart glasses as an example. They do not have a large screen suitable for complex operations, and they are not convenient for users to touch for a long time. If a user is walking on the street and wants to find a nearby quiet cafe with sockets, the ideal experience is to make a request directly to the glasses, let AI filter the location based on location and preferences, and then start navigation.

The whole process does not require taking out your mobile phone, nor does it require opening maps, reviews and navigation applications one by one.

Picture|BBC

Back at home, AI also has the opportunity to take over more trivial operations.

Most current smart home assistants rely on preset instructions to operate. You can ask it to turn on the lights, play music, and adjust the air conditioner, but asking it to coordinate different devices and outside services around family members' schedules isn't that easy.

If the future HomePod or HomeView can understand the schedule, device status and living habits of family members, and then proactively coordinate home appliances and external services, home AI will have the opportunity to change from passively responding to voice commands to an intelligent assistant that can independently judge needs, plan tasks and take action.

Picture | HomeView

It is difficult for smart glasses, home devices, and new forms of wearable products that may appear in the future to copy the touch screen operation methods on mobile phones. AI that can understand user needs and call different services may become one of the most important ways of interacting with these devices.

Apple is already trying to make Siri take on more tasks on existing devices.

The iOS 27.3 developer beta opens up Siri AI’s third-party application access capabilities, allowing developers to integrate their own application functions into Siri. Apple is also continuing to enhance Siri’s ability to operate system applications.

Figure | Siri AI screen sensing function demonstration / Apple Newsroom

However, being able to execute more instructions is still a long way from truly completing complex tasks independently. Getting Siri to keep tracking a goal, decide on its own what to do next, and even move forward with a task without repeated reminders from the user is still a problem Apple needs to overcome.

Apple management seems unwilling to wait any longer.

According to Bloomberg, new CEO John Ternus pressured software engineering chief Craig Federighi in a closed-door meeting to speed up software development and move beyond Siri's regular feature updates as soon as possible to adopt a more aggressive AI strategy.

Picture | Apple CEO John Ternus/Apple Newsroom

From the release of Apple Intelligence in 2024 to the launch of the new Siri AI in 2026, Apple took more than two years to gradually integrate generative AI into the operating system. Competitors have begun to let Agent take over real work, and Apple's development pace obviously needs to be accelerated.

There is another noteworthy detail in this recruitment information: Apple’s AI Agent may not appear as an independent application.

This set of capabilities will be developed in conjunction with Siri, UI frameworks, and the Apple Foundation Models team, according to the job posting. Another job posting more clearly stated that related technologies will become core platform capabilities of the operating system and Apple Intelligence.

Figure | Typing Siri demonstration when Apple Intelligence is released in 2024 / Apple Newsroom

If it develops in this direction, Siri in the future may be just an interface for users to contact AI Agents.

On Mac, users can let AI handle complex tasks through text; on iPhone, users can complete cross-application tasks through voice; on smart glasses and home devices, AI may become the main way of interaction.

The same set of agent capabilities can be extended to different devices to complete tasks based on their respective hardware characteristics and personal context.

Ecological integration between multiple devices has always been Apple's moat. The addition of AI Agent is expected to further leverage the advantages of this ecosystem, allowing iPhone, Mac and home devices to no longer be limited to transmitting information to each other, but can also collaboratively plan and execute complex tasks.

Of course, the extent to which this set of capabilities can ultimately be achieved depends on whether Apple can enable the model to stably complete complex tasks, handle cross-application permissions, and provide a consistent experience across different devices.

Apple once redefined mobile phones through multi-touch, established a software ecosystem for the mobile Internet through the App Store, and regained control of the core computing platform of the Mac through Apple Silicon.

Figure | Siri AI visual intelligence demonstration on Apple Vision Pro / Apple Newsroom

The emergence of AI Agent has brought a new problem to Apple: when users no longer need to open applications and operate software in person, how can iPhone, Mac and future new devices retain users?

Apple can launch products later, but it cannot miss the formation of the next generation of computing portals.

Zheng Tingxu

Hardware editor

One more revision.

累计已发布 352 篇文章

本篇来自栏目

上一篇2026-11-20 08:29

早报|苹果招人打造「接管电脑」的AI智能体/史上最大IPO,Anthropic上市/华为官宣Mate 90 Air主打轻薄

爱范儿 App

爱范儿,让未来触手可及

关注明日产品的数字潮牌

先进工具,先知先行,AIGC 的灵感指南

造车新时代,明日出行家

探索城市新生活方式,做你的明日生活指南

版权所有 © 广州利沃致远投资管理合伙企业(有限合伙) 2008 - 2026。以商业目的使用爱范儿网站内容需获许可。非商业目的使用授权遵循 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.