Apple’s next AI product may not be a better chatbot. It may be an agent that operates a computer the way a person does.

That is the implication of a job listing Apple posted on 17 November for a “Senior Engineering Manager, Agentic AI – Proactive” in Cupertino. Bloomberg’s Mark Gurman spotted it first; Engadget, 9to5Mac and Macworld followed. By 20 November, Apple had pulled the posting. The text is still the most detailed public description of what the company is building.

What the listing actually said

The summary opened with a concrete pitch: “Imagine an AI agent that can use a computer the way a capable person does.” It continued: the agent “browses the web, works across apps, runs commands, writes and debugs code, and calls services, and it chooses the right tool for each step.” Apple then added the qualifier it puts on almost everything now: “It does all this privately and with the polish people expect from Apple.”

The team description is more revealing than the summary. Apple described “general-purpose computer-use agents” that “plan and carry out multi-step tasks across apps, browsers, command lines, code, and APIs” and “can coordinate specialized sub-agents along the way.” The manager would own “the technical direction and delivery of our agent platform,” lead managers and engineers, and work with teams across “Siri, UI frameworks, and AFM”, Apple Foundation Models.

That is not a feature brief. It is a platform brief. Apple is not describing one assistant that answers questions. It is describing a runtime that can take actions, call tools and hand pieces of a job to other agents.

A cautious company moves into an aggressive category

Apple has spent the past two years watching its rivals ship agents. OpenAI has Dots, Meta has Muse, and Google has Gemini Spark, as Engadget noted. Each is an attempt to turn a model from something you talk to into something that does work for you: opening apps, filling forms, moving files, writing and running code.

Apple’s public position has been more conservative. The company has emphasized on-device processing, private cloud compute and narrow, permissioned features. That caution is not only branding. An agent that can control a computer needs access to the filesystem, the clipboard, the browser and the apps where personal data lives. The attack surface is the product.

The listing suggests Apple has decided the capability is worth that risk, provided the permissions model is built first. The phrase “capable, reliable and safe” appears in the role’s description of what the team must deliver. In Apple’s vocabulary, “safe” is not a modifier. It is a shipping requirement.

The permission problem Apple has already flagged

Days before the listing surfaced, Apple warned developers that agentic AI “could put users at risk” by unwittingly granting Full Disk Access. It then rolled out new controls to give users more oversight over what AI agents can reach. The sequence matters: Apple is shipping the guardrails before the agent.

That is the opposite of how the agent race has gone elsewhere. Most agent launches have started with capability and added controls after researchers found ways around them. Apple appears to be trying to reverse the order. Whether that produces a better product or simply a slower one is the open question.

There is also a competitive cost. Every month Apple spends hardening permissions is a month OpenAI, Google and Meta spend teaching users what an agent feels like. Habit is hard to displace, even with a better permission dialog.

What to watch, and what not to conclude

A job posting is evidence of investment, not a launch date. It does not tell us whether the agent will ship as part of Siri, as a developer framework, or as a separate app. It does not say which models run on-device and which run in the cloud. And it does not promise that the agent will control third-party apps rather than a curated set of Apple services.

Three details are worth tracking. First, the reference to “AFM” implies Apple’s own foundation models are part of the stack, not just a partner model. Second, the mention of “UI frameworks” suggests the agent will interact with interfaces rather than private APIs, a more general approach, and a harder one to secure. Third, the word “Proactive” in the job title hints at an agent that initiates work, not one that waits for a prompt.

Apple has pulled the listing, which is what companies do when a posting says too much. The description is gone from Apple’s careers site, but it is preserved in copies at Macworld and iHireEngineering. That is enough to establish the direction of travel.

Bottom line

Apple is not building a chatbot with more personality. It is building an agent platform, staffed by a team that spans Siri, interface frameworks and Apple’s own models, with safety and privacy treated as launch blockers. The listing is the clearest signal yet that Apple sees computer-use agents as the next layer of the operating system, and that it intends to arrive late, with fewer sharp edges, rather than first.

Have a tip? Email [email protected] or use the contact form and choose Editorial. We do not promise anonymity.

Sources: Engadget; Macworld, including the full listing text and the note that Apple pulled it; 9to5Mac; iHireEngineering, which archived the role description.