Apple wants an AI that uses your Mac the way you do, and it is tightening the doors for everyone else's agents at the same time.
Apple is developing its own autonomous AI agent that can browse the web, control apps, run commands and write code on a user's behalf, according to Bloomberg's Mark Gurman, who spotted new job listings Apple added to its careers site this week. The report landed on Thursday, November 19 (early Friday India time), the same day Google gave its Gemini app the ability to operate a Mac. Apple has not commented, and Bloomberg says development is still in its early stages, with a launch eyed for as soon as the end of 2027.
What Apple's job listing actually says
The clearest evidence comes from a posting for a senior engineering manager to lead work on a "proactive, agentic AI assistant." The listing paints an ambitious picture. "Imagine an AI agent that can use a computer the way a capable person does," it reads, describing a system that "browses the web, works across apps, runs commands, writes and debugs code, and calls services," choosing the right tool for each step, and doing it "privately and with the polish people expect from Apple."
The posting also says Apple already has a team working on "general-purpose computer-use agents" that can plan and carry out multistep tasks across apps, browsers, command lines, code and APIs, and that can "coordinate specialized subagents along the way," Bloomberg reported. In other words, Apple is describing the same class of technology that already lets ChatGPT and Meta's Muse take control of a Mac.
Privacy and safety feature heavily. The listing states that "agents that can run code, use credentials, and act on a user's behalf need strong permission models, sandboxing and transparency." TechCrunch said it asked Apple for more information but did not get a response.
The numbers and the timeline
A few concrete details stand out in the reporting:
- Launch window: Apple has eyed introducing the service as soon as the end of next year, per Bloomberg, but the project "remains in the early stages."
- Market reaction: Shares of Meta fell nearly 2% and Anthropic, which listed on Nasdaq the same day, dropped more than 3% in late trading after the report, according to Bloomberg.
- App Store pressure: Meta's Muse sat at No. 2 among free iPhone apps on Thursday, behind only TikTok, with ChatGPT and Claude also in the top five, Bloomberg noted.
- Friction for rivals: In the macOS 27.3 beta, enabling ChatGPT's computer-use tools now takes 14 distinct steps, up from four.
Bloomberg also reported that new CEO John Ternus has been pushing software chief Craig Federighi to move faster and take a bolder approach to AI, with an agent seen as central to that effort.
Background: Apple tightens control over Mac agents
The job listing arrives just days after Apple changed how third-party apps get deep access to the Mac. In a developer post dated November 16, Apple said macOS 27.3 beta 2 splits the old "Device Control and Data Access" permission into two settings, System & App Protection Override and Accessibility. It adds guidance nudging users to grant access only to specific apps or folders, an option to make apps ask every time, and an on-screen indicator when one app is controlling another. "Before granting an app such extraordinary access, users should understand that it puts all of the data on their Mac at risk," Apple wrote. Previously granted permissions stay in effect.
Apple frames these changes as user protection. But as Bloomberg points out, some AI companies and users see the extra steps as a way to slow down competing agents while Apple builds its own. That tension matters because Apple has already been sued by the US Justice Department over practices that include blocking third-party access to features and hardware.
Why the privacy pitch matters right now
Apple's emphasis on permissions and sandboxing lands at a sensitive moment for the category. Earlier this week, Reuters reported that internal Meta posts showed employees testing Muse flagging the agent surfacing messages that appeared to belong to other people, including a bank balance one staffer said was not his, and reading Signal message text. Meta called those "isolated incidents" and said many were hallucinations that had already been addressed; Reuters said it could not establish whether the data was real or made up.
That is the kind of episode Apple's listing seems written to avoid. Whether Apple can deliver an agent that is both capable and tightly locked down is an open question, and nothing about its product has been shown publicly.
What it means for users and the AI race
Our read: for Mac and iPhone owners, the near-term change is not an Apple agent but more consent screens when setting up tools like ChatGPT, Gemini or Muse. For the industry, the report confirms that the last major platform holdout intends to compete directly in computer-use agents, alongside OpenAI's Dots, Anthropic's Duet, Google's Tasks and Meta's Muse. Because Apple controls the operating system those agents run on, how it sets permission rules could shape how well rival agents work on its devices.
Apple's nearer-term step is Siri: its upcoming iOS 27.3 release lets the assistant work more deeply with outside apps, for example to book a car or a restaurant table, according to Bloomberg. That is still short of a fully autonomous agent that can carry out complex jobs in the background.
What to watch next
Key things to track: whether Apple confirms or previews agent features publicly, how the final macOS 27.3 permission flow looks when it reaches all users, whether AI developers push back on the extra setup steps, and whether Apple fills the role the listing describes. For now, the job listings are the best window into Apple's plans, and they point to an agent that is still a work in progress rather than a product on the way.