AI Trends

In the First Year of AI Agents, Apple Becomes the Last Tech Giant to Enter the Race

Phoenix TechnologyUpdated Nov 20 10:51

AI summary

Apple is preparing agent capabilities for cross-application execution and multi-step task planning by recruiting talent in autonomous AI agents and modifying macOS permission mechanisms. According to Bloomberg, the relevant products remain in the early stages of development, and Apple had considered launching them as early as the end of 2027. Drawing on products and business data from Meta, OpenAI, Anthropic, Microsoft, Tencent, and ByteDance, the article analyzes the competitive pressure Apple faces and the advantages of its device ecosystem.

Why read this

AI score 58

The article places Apple’s hiring, system permission changes, and competitors’ business data within a single framework, illustrating the pressure it faces in entering the AI agent market and the ecosystem assets it can leverage.

Topics

Tags

Body · AI translation

The First Year of AI Agents: Apple Is the Last Tech Giant to Enter the Race

Download the App, Exclusive Early Access — Aifanr

November 20, 2026, 09:11:19, from Beijing Updated November 20, 2026, 09:29:43

Siri may finally be evolving from “talking” to “taking action.”

Asking Siri to check the weather, set an alarm, or open an app is nothing new. But if you asked it today to arrange a business trip, find flights and a hotel, and then organize an itinerary around your calendar, you would probably still have to do it yourself.

A series of recent moves by Apple has revealed its ambition to change that.

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

One described an agent capable of operating a computer like a human, while the other explicitly named Claude Code and Codex, requiring applicants to use such tools regularly.

According to Bloomberg, Apple is developing its own autonomous AI agent in the hope of enabling artificial intelligence to perform tasks across apps and even plan workflows on its own.

Image | Meta Muse

But across the industry, the capabilities Apple is describing are hardly new.

Meta, OpenAI, Anthropic, Tencent, and ByteDance are all bringing AI agents into real-world work and everyday-life scenarios. Some products have already amassed tens of millions of users and begun generating actual revenue.

For Apple, which has built a vast ecosystem around the iPhone, Mac, and App Store, this is probably not a competition it can afford to keep approaching slowly.

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

The job postings show that Apple is developing a general-purpose computer-use agent system. It would be able to switch among apps, browsers, command lines, coding environments, and APIs; independently plan and execute multistep tasks; and coordinate multiple specialized sub-agents working together.

Image | Codex

For a business trip, for example, the user would only need to say, “Arrange a business trip to Tokyo for me next week.” The AI could then search for flights, shortlist hotels, check the calendar, organize meeting arrangements, and ultimately produce an itinerary. Whether to look up flights or check the calendar first, and which tools to invoke, would be left for the agent itself to decide.

The other job posting focuses on the agent’s continuous improvement.

Apple is seeking a senior applied AI engineer to establish personal-intelligence evaluation benchmarks, improve evaluation infrastructure, and continually optimize the Agent’s performance through feedback mechanisms.

More interestingly, Apple specifically mentioned Claude Code and Codex in the job requirements, even requiring candidates to use these kinds of intelligent coding tools in their daily work.

If a user asked an Agent to organize old files on their computer, it might accidentally delete important data. If it could read emails and run code, how should the scope of its permissions be restricted? This is also why Apple repeatedly emphasized privacy, security, sandbox isolation, and permission management in its job postings.

Recent macOS updates happen to include changes in this area as well.

In macOS 27.3 Beta 2, Apple adjusted the authorization mechanisms for device control and Full Disk Access, separating the management of different types of system permissions and allowing users to grant access to specific apps and folders. The system also provides prominent status indicators when an AI app is controlling other apps.

Users can see more clearly which permissions an AI has obtained and what it is doing, while the way third-party Agents operate on Macs will also be affected.

By adjusting permission rules for third-party Agents while recruiting people to develop its own, Apple is clearly preparing for AI to become deeply integrated into the operating system.

It must give the AI enough capability to complete tasks without allowing it unrestricted access to users’ data and devices. Both issues have to be solved together.

As for when it will become available, there will probably still be a considerable wait.

According to Bloomberg, Apple once considered launching a related product as early as the end of 2027, but development is still at an early stage.

2026: AI Agents Truly Entered the Mass Market

In 2023, ChatGPT gave ordinary users their first large-scale exposure to generative AI. Three years later, AI’s capabilities are no longer confined to the chat window.

Image | An early ChatGPT interface

Having a model write some code or analyze a document is no longer particularly remarkable. What draws more attention is whether it can then run the code, revise it based on error messages, or open the relevant apps and keep working until it finishes the task assigned by the user.

Meta’s Muse is one of the most closely watched consumer products of the year. It can connect to users’ digital accounts and devices to complete tasks such as planning trips, booking restaurants, and organizing information.

Image | Muse launch promotional image / Meta Newsroom

According to a report by The Information in late September citing internal Meta data, several weeks after Muse launched, more than 3 million users were actively sending it tasks each week, and more than 1 million were actively giving it instructions each day.

Meta subsequently disclosed during its third-quarter earnings call that Muse already had more than 10 million users who had set it up in the United States and Canada, about 25% of whom used Muse every day.

Image | Promotional image used by Meta to announce Muse’s launch in Australia / Meta Newsroom

In late September, OpenAI launched Dots, a type of personal AI agent capable of working continuously. Each Dot has its own computer in the cloud, can connect to more than 4,000 apps, and can pursue goals assigned by the user over the long term.

Users can ask it to research a project, organize materials, create documents, or even continuously advance a software development project. It can keep working in the background even when the user steps away.

Image | The Dots interface demonstrated by OpenAI at DevDay / OpenAI

Anthropic, meanwhile, has entered the software development and general-purpose agent markets with Claude Code and Claude Duet, respectively.

Claude Duet can proactively organize information that requires the user’s attention, access authorized accounts, draft emails, schedule meetings, and continue handling tasks after the user leaves. It is currently available as a research preview.

Image | Some of the apps and services that Claude Duet can connect to / Anthropic

Muse, Dots, and Claude Duet take different approaches, but they have one thing in common: they are beginning to have their own work environments, can connect to external services, and can continue handling tasks after users leave.

Image | Anthropic

Compared with personal agents that are still competing for ordinary consumers, the software development and enterprise productivity markets have already produced clearer commercial returns.

Microsoft disclosed during its fiscal 2026 fourth-quarter earnings call that GitHub Copilot had reached 50 million users and that about one-third of Pull Requests on the GitHub platform involved AI agents.

Meanwhile, Microsoft 365 Copilot had surpassed 30 million paid seats, while revenue from the Copilot business grew by more than 60% quarter over quarter.

Image | The agent task interface in the GitHub Copilot desktop app / GitHub

Competition in China is also intense.

On November 18, Tencent disclosed in its third-quarter earnings report that WorkBuddy had surpassed 10 million daily active users. Based on revenue over the previous seven days, its annualized revenue run rate exceeded $500 million.

Doubao Work, owned by ByteDance, had also surpassed 8 million daily active users, while annual recurring revenue from enterprise customers exceeded RMB 2 billion.

Tencent and ByteDance are both drawing on their respective productivity software, cloud services, and user ecosystems to bring agents into everyday work scenarios. Compared with cultivating user habits from scratch, their existing enterprise customers and software entry points also make it easier for them to identify real-world needs.

Viewed together, these products show that several distinct strategies have emerged in the market. Meta, OpenAI, and Anthropic want AI to handle more personal tasks; Microsoft is embedding agents into software development and enterprise workflows; and Tencent and ByteDance are using their existing businesses to seek room for growth.

Image | Lemonade

The capital investment behind this competition is equally astonishing.

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

Anthropic, meanwhile, has just completed its initial public offering (IPO), listing on Nasdaq at $83 per share, with the offering price implying a valuation of approximately $1.5 trillion.

Image | Anthropic co-founders attend the Nasdaq opening-bell ceremony for the company’s listing / Associated Press

Investment in AI infrastructure continues to rise, and capital markets remain willing to assign lofty valuations to leading AI companies. More importantly, agent products have begun to demonstrate their user scale, frequency of use, and commercial revenue, showing that they have the potential to become sustainable long-term businesses.

Apple is now trying to catch up in a market that already has tens of millions of users, real commercial revenue, and massive capital investment.

Apple cannot afford to lose the battle for the next generation of computing interfaces either

Apple certainly does not lack users or money.

According to its latest fiscal 2026 fourth-quarter earnings report, Apple’s quarterly revenue reached $114.5 billion, up 12% year over year, while full-year revenue hit a record high of $478.8 billion.

Image | Apple’s fiscal 2026 fourth-quarter financial statements / Apple Newsroom

Of this, quarterly revenue from Services reached $31.573 billion, also setting a record. Apple’s installed base of active devices likewise reached all-time highs across every major product category and geographic region.

Apple spent more than a decade building an ecosystem centered on the iPhone, Mac, iPad, its operating systems, and the App Store. Users turn on a device, find the apps they need, and then use those apps for shopping, work, socializing, and entertainment.

The operating system manages the apps, the apps provide services, and users decide which ones to open and how to use them. Through its integration of hardware and software, as well as the App Store’s distribution system, Apple controls the key parts of this ecosystem.

AI Agents could make some of those actions unnecessary.

Image | Siri AI Q&A / Apple Newsroom

Airlines, booking platforms, and payment services would still exist, but users might no longer need to open their apps themselves. Actions previously scattered across multiple apps could be handled by a single agent.

If users grow accustomed to using ChatGPT, Claude, or other third-party Agents for everyday tasks, Apple could lose some of its dominance over the gateway through which users interact with technology.

Whoever controls the gateway to users’ needs has an opportunity to determine how they use the entire digital ecosystem.

Image | Apple’s demonstration of Siri integration with ChatGPT / Apple Newsroom

However, Apple also holds quite a few strong cards.

Its enormous device base, unified operating-system architecture, and the personal data it can access with user permission could all give it an advantage in developing AI Agents.

Third-party agents seeking to operate a device generally need to request system permissions and then connect to various apps and services. If Apple deeply integrates an Agent into its operating systems, it could more easily access device functions and complete tasks using information such as schedules, locations, and device status.

This integration could prove even more valuable on new hardware in the future.

Take smart glasses, for example. They lack a large screen suitable for complex operations, and prolonged touch input is impractical. If a user walking down the street wants to find a quiet nearby café with power outlets, the ideal experience would be to simply tell the glasses what they want, have AI filter locations based on their position and preferences, and then start navigation.

Image | BBC

Most current smart-home assistants rely on preset commands. Users can ask them to turn on the lights, play music, or adjust the air conditioning, but asking them to coordinate different devices and external services based on family members’ schedules is not so easy.

If a future HomePod or HomeView can understand family members’ schedules, device status, and daily habits, and then proactively coordinate appliances with external services, household AI could evolve from passively responding to voice commands into an intelligent assistant capable of independently assessing needs, planning tasks, and taking action.

Image | HomeView

Smart glasses, home devices, and potential new forms of wearable products cannot easily replicate the touchscreen interactions used on smartphones. AI that can understand user needs and access different services may become one of the most important ways to interact with these devices.

Apple has already been trying to make Siri handle more tasks on its existing devices.

The iOS 27.3 developer beta opened Siri AI to third-party app integration, allowing developers to incorporate their apps’ functions into Siri. Apple is also continuing to strengthen Siri’s ability to operate system apps.

Image | Demonstration of Siri AI’s on-screen awareness feature / Apple Newsroom

Still, being able to execute more commands is a long way from completing complex tasks truly independently. Enabling Siri to continuously pursue a goal, decide on its own what to do next, and even keep advancing a task without repeated reminders from the user remains a challenge Apple must overcome.

Apple’s management also appears unwilling to keep waiting patiently.

According to Bloomberg, newly appointed CEO John Ternus pressured software engineering chief Craig Federighi at a closed-door meeting to accelerate software development, quickly move beyond routine Siri feature updates, and adopt a more aggressive AI strategy.

Image | Apple CEO John Ternus / Apple Newsroom

From the release of Apple Intelligence in 2024 to the launch of the all-new Siri AI in 2026, Apple took more than two years to gradually integrate generative AI into its operating systems. With competitors already putting Agents to work on real-world tasks, Apple clearly needs to pick up the pace of development.

There is another noteworthy detail in the latest job postings: Apple’s AI Agent may not take the form of a standalone app.

The job postings indicate that the capability will be developed in coordination with the Siri, UI frameworks, and Apple Foundation Models teams. Another posting states even more explicitly that the technology will become a core platform capability of Apple’s operating systems and Apple Intelligence.

Image | Demonstration of Type to Siri at the 2024 launch of Apple Intelligence / Apple Newsroom

If development continues in this direction, Siri may eventually be just one interface through which users access the AI Agent.

On the Mac, users could instruct AI by text to handle complex work; on the iPhone, they could complete cross-app tasks by voice; and on smart glasses and home devices, AI could become the primary mode of interaction.

The same agent capabilities could extend across different devices, completing tasks based on each device’s hardware characteristics and the user’s personal context.

Integration across multiple devices has always been Apple’s moat. The addition of an AI Agent could further capitalize on the strengths of this ecosystem, allowing the iPhone, Mac, and home devices to move beyond simply exchanging information and instead work together to plan and execute complex tasks.

Of course, how capable the system ultimately becomes will depend on whether Apple can enable its models to reliably complete complex tasks, properly manage cross-app permissions, and deliver a consistent experience across different devices.

Apple once redefined the phone through multi-touch, built the mobile internet’s software ecosystem through the App Store, and regained control of the Mac’s core computing platform through Apple Silicon.

The emergence of AI Agents presents Apple with another new question: When users no longer need to open apps and operate software themselves, what will enable the iPhone, Mac, and future new devices to retain them?

Apple can launch products later than others, but it cannot afford to miss the emergence of the next generation of computing interfaces.

“Special notice: The above content (including any videos, images, or audio) was uploaded and published by users of Dafeng Hao, a self-media platform operated by Phoenix New Media. This platform merely provides information storage services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services.”

Source

Phoenix Technology