Apple Is Building an AI Agent That Can Work Across Your Apps, Job Listing Suggests

Apple is working on an AI agent that can browse the web, work across apps, write code and call services, according to The Verge, which reported the project on November 19, 2026.
The clearest signal is an Apple job listing for a Senior Engineering Manager tasked with leading the creation of an agentic AI assistant. An agentic assistant plans and carries out a sequence of steps on its own, rather than answering a single question. The posting describes agents that plan and carry out multi-step tasks across apps, browsers, command lines, code and APIs (the interfaces that let software programs talk to one another). It also says the agents can coordinate specialized sub-agents, a design in which a supervising agent hands subtasks to narrower workers.
The role is expected to work closely with teams across Siri, UI Frameworks and Apple Foundation Models. Those groups cover, respectively, the assistant itself, the interface layer that apps are built on, and the foundation models, the large general-purpose AI models that power Apple's AI features. A hire at this level would sit where those three areas meet.
Apple's AI-upgraded Siri began rolling out in September. According to The Verge, the current version can answer questions about what is on screen and perform actions across apps. The company's upcoming iOS 27.3 update will let developers integrate Siri AI into “all apps” through a Developer Preview, an early-access period in which developers can test the feature before wider release.
The job listing describes capabilities Apple intends to build and a role it is filling. It is not a product announcement.
In my view, the more consequential question concerns the boundary between the assistant and third-party software. An agent that moves across apps depends on what those apps expose, whether through intents (declared actions that other apps or the system can call), APIs or scripting hooks, and on what the user permits. The Developer Preview suggests Apple is treating that integration as something developers will shape, not something the operating system can do alone.
The broader context here is security. An agent that browses the web and calls services reads content it does not control. That content can contain hidden instructions, a tactic known as prompt injection, that attempt to redirect the agent's actions while it holds the user's permissions. Confirmation prompts, scoped credentials and clear limits on what an agent can do without approval will likely matter as much as the planning model itself. The listing's reference to command lines and APIs suggests the design surface is wide, which raises the stakes for those controls.
The optimistic reading is also clear. Much of everyday computing consists of moving information between applications: copying a flight number into a calendar, pulling a document from one service into another, repeating the same sequence across tabs. If an agent can reliably complete those multi-step chores on the device, the friction between apps shrinks. Whether Apple's version delivers that depends on how the planning, permissions and developer hooks come together, and the company has not yet shown the complete system.


