Apple eyes a Siri that acts on its own as soon as next year, Gurman reports

Apple wants Siri to stop waiting for instructions. The company could turn its assistant into something closer to an autonomous agent, one that tracks a goal and works through the steps on its own, as soon as next year, according to Bloomberg's Mark Gurman.

Gurman made the report in Sunday's edition of his Power On newsletter. He was careful about the timing and presented it as his own read rather than a fixed plan. He compared the idea with agents such as OpenAI's Dots, Meta's Muse and Grok, which can hold onto an objective, work out what comes next and keep going without a person prompting each move. "Apple ultimately wants to go in this direction, I'm told," he wrote. He added that he believes the company could make Siri "more like an agent" as early as next year.

Apple has not announced any such feature, and the report says nothing about how an agentic Siri would work or which tasks it would handle. Readers should treat it as a direction, not a product.

The more revealing detail concerns Apple's leadership. In a closed-door meeting within the past several weeks, Gurman reported, CEO John Ternus told software engineering chief Craig Federighi that Apple must "accelerate" its software roadmap. According to the report, Ternus also said Apple needs to "move quickly beyond merely making necessary improvements to Siri."

That wording matters. It suggests the head of the company sees the Siri overhaul that just shipped as overdue catch-up work, not a finish line. It's an honest assessment. Apple first showed its personalized Siri at its developer conference in 2024, then delayed key pieces of it the following year. The features finally arrived with iOS 27. By most accounts Apple delivered what it promised, but it took two years to get there while rivals moved on to assistants that take action without being asked.

Today's Siri is smarter and more aware of personal context than the old version. It still mostly reacts: you ask, it answers or carries out a single command. The agents Gurman cites do something different. Some run in their own virtual machines and handle multistep jobs in the background, then return with results. That is the gap Apple now appears to want to close.

Apple has a real structural advantage. For years it has pushed developers to adopt App Intents, a framework that lets apps describe their actions in a form the system can call directly. An agent that can trigger those actions does not have to operate an app the way a person would, by reading the screen and tapping buttons. That kind of computer use is slow, costly to run and easy to break when an interface changes. A Siri agent built on structured app actions could be cheaper and more dependable, provided enough developers have done the integration work.

The risks are just as concrete. An assistant that answers a question badly is an annoyance. An assistant that books, sends, buys or deletes something on its own initiative can do actual damage. Apple's brand rests on predictability and privacy, and a proactive agent strains both. It would need broad access to messages, calendars and accounts, and it would need to know when to stop and ask. Apple's slow, cautious rollout of the personalized Siri suggests it understands this. The open question is whether Ternus' push for speed fits with that caution.

There is also a credibility problem. Apple promised too much with Siri once and paid for it in a long public delay. Committing to an agent on a one-year horizon invites the same trap. The safer bet would be a narrow first version that does a few proactive tasks reliably, rather than a broad launch that matches rivals on paper and disappoints in daily use.

Gurman also reported a change in how Apple builds its software, and it may matter as much as any single Siri feature. Development of a new iOS version normally moves through milestones, each split into two four-week phases: one for writing code and building features, and one for convergence, when that work is merged into the operating system and stabilized.

For the first milestone of iOS 28, Apple changed the split. Engineers got six weeks for coding, exploration and feature development, and only two weeks for convergence. Employees like the change, according to the report, because it gives them more room to try ideas before they have to lock anything down.

9to5Mac suggested the shift was probably driven by rapid gains in AI coding tools, which can shorten the time needed to write and test code. That reasoning is plausible, though the report does not say so. Either way, the change fits the Ternus directive. More exploration time early in the cycle is how a company tries to produce bigger features, not just polish existing ones. The cost could come at the other end: two weeks of integration is a tight window for an operating system that runs on hundreds of millions of devices, and rushed integration is where bugs tend to slip through.

Apple usually previews its next iOS at its developer conference in June and ships it in the fall. If Gurman's timeline holds, that June presentation will be the first real test of whether the new process and the CEO's urgency have produced something more ambitious than a better chatbot. For now, the clearest signal is internal. Apple's leadership appears to have decided that fixing Siri was the minimum, and that the company is behind on what comes next.