Article · AI translation
Thanks to IT Home reader 乌蝇哥的左手 for the tip!
IT Home reported on November 20 that Bloomberg’s Mark Gurman said today Apple is developing its own AI agents, with plans for AI to perform tasks on users’ behalf such as controlling devices, operating apps, browsing the web and writing software, further entering the AI agent market where companies including OpenAI, Anthropic and Meta compete.
Gurman found that Apple has posted job openings on its website, including one for a senior engineering manager titled “Agentic AI - Proactive.” The posting says Apple is assembling a team to develop AI assistants that can carry out tasks on users’ behalf. The systems will be able to write code, browse the web, control applications and operate devices.
In the job posting, Apple described how its future AI agents would work: users could have AI use a computer like a skilled human operator, moving among browsers, documents, terminals, code editors and various services, and independently choosing the right tools to complete tasks.
The system would be able not only to execute a single instruction but also to plan and complete multistep tasks across apps, browsers, command lines and APIs, while coordinating multiple subagents specializing in different work. Apple also placed particular emphasis on privacy, saying agents that can run code, use user credentials and act on users’ behalf need strict permission management, sandbox isolation and transparency about their actions.
The project is still in an early stage of development, however. Gurman said Apple had considered launching the service as early as the end of 2027, but the timing has not been decided. An Apple spokesperson did not respond to Bloomberg’s request for comment.
Gurman said Apple’s new CEO, John Ternus, has been pushing software engineering chief Craig Federighi to accelerate AI development and adopt a more aggressive strategy. Autonomous AI agents are seen as an important part of Apple’s efforts to advance that strategy.
Several competing products have already emerged in the AI agent market. Meta’s Muse has recently attracted many users with capabilities such as booking restaurants, arranging travel and generating news briefings. Anthropic is expanding user testing of its own agent, Duet, while OpenAI continues to expand the capabilities of Dots. Google adjusted its agent this week, renaming it Tasks and adding capabilities such as making phone calls on users’ behalf.
Meanwhile, OpenAI, Anthropic and Elon Musk’s SpaceXSI have also developed developer coding tools including Codex, Claude Code and Cursor, respectively. Gurman noted that AI agents already occupy a prominent place in Apple’s App Store. As of Thursday local time, Meta’s Muse had long held second place in the free apps rankings, behind only TikTok, while ChatGPT and Claude were also among the top 5 free apps.
As Apple develops its own agent, it is also adjusting how macOS manages permissions for third-party apps. The upcoming macOS 27.3 beta adds permission warnings for third-party AI agents, requiring users to confirm again when granting those apps permission to control a Mac. The changes increase the number of steps ChatGPT users must take to enable third-party computer-control tools from 4 to 14. Some AI companies and users believe that, beyond protecting system security, this could also make it harder for competitors to gain access to macOS.
Apple, for its part, describes the changes as strengthening the protection of user data. On November 16, Apple had already announced planned permission changes for macOS 27.3 Beta 2 covering device control, data access and full disk access, with a focus on making permission management more transparent and offering more granular access controls.
In addition to a standalone AI agent, Apple is working to integrate Siri with third-party apps. The upcoming iOS 27.3 will give developers deeper Siri AI integration, allowing the voice assistant to call on external apps to complete tasks, such as booking a car or a restaurant.
At this stage, however, Siri’s integration with third-party apps still differs from a fully autonomous AI agent. The latter can independently plan and carry out complex tasks in the background across multiple apps and services, reducing the need for users to repeatedly give instructions during execution.
Related reading:
“Apple’s New Rules Curb AI Apps From ‘Taking Full Control’ of Macs; Gurman Says They Lay the Groundwork for Its Own AI Agent 2 Years in Advance”
“Apple Warns: AI Agent Behavior May Be Hard to Predict, Risking Lawsuits and Damage to the Company’s Reputation”
“Apple Intelligence Officially Announces Its Entry Into China: Alibaba’s Qwen Joins the Apple Ecosystem and Can Help Users Write Copy, Read Documents and Recognize Images”