Apple is developing an AI agent that could browse the web, use apps, write code and carry out multi-step tasks on a user’s behalf. The plans point to a serious push into computer-using AI, even as Apple makes it harder for rival agents to control Macs.
Watch Desk analysis
What happened
The Straits Times reports that Apple job listings describe a proactive assistant able to work across apps, browsers, command lines and code. The listings emphasise privacy and say the company is building agents that can plan and complete tasks, including coordinating specialised subagents. The report says Apple has eyed a launch as soon as the end of next year, though development remains at an early stage.
There is a second, more immediate move: in the beta of macOS 27.3, Apple has added warning screens for third-party computer-use tools. The report says enabling that support for ChatGPT now takes 14 steps, up from four. Apple presents the change as a safety measure; some AI companies and users see it as a possible barrier to competitors.
Read the Straits Times report.
Why it matters
An agent that can move between a browser, documents, apps and code is a bigger proposition than a chatbot that answers questions. It could take on multi-step work, but doing so means handling permissions, credentials and actions that can affect a person’s device. Apple’s emphasis on sandboxing and transparency addresses the heart of that challenge, at least in the plans described in its listings.
The Mac warnings add a competitive wrinkle. The same permission controls that may protect users could also make rival agents less convenient to use. The report does not establish Apple’s motives, but it puts a practical question on the table: who gets to decide which agents can operate across a platform, and on what terms?
Our read
This is a meaningful signal, not a product launch. Apple has described the intended capabilities and its privacy priorities, but an early-stage job listing is not a working agent, and the report gives no release date firmer than a possible target. The 14-step hurdle, meanwhile, is already a tangible change for people testing third-party tools. The next test is whether Apple can make its own agent capable without making rivals need a map, a packed lunch and the patience of a saint.
What to watch
- Whether Apple gives a firm launch date or demonstrates the agent completing real multi-step tasks.
- How its permission, sandboxing and transparency controls work in practice.
- Whether the macOS warning flow changes before release, and how Apple explains its effect on third-party tools.
Discussion spark: Should Apple’s extra steps for third-party computer-use agents be treated mainly as sensible security friction, or as a platform advantage for its own future agent?
Sources and evidence
- Source update (20 November 2026, 01:30 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.