Apple CEO John Ternus is pressing for a faster, more agentic Siri, Bloomberg reports

Apple's new chief executive, John Ternus, is privately pressing his software chief to speed up the company's AI work and turn Siri into something closer to an autonomous agent, according to Bloomberg's Mark Gurman.

Gurman reports that Ternus is publicly pleased with the revamped Siri but has been urging Craig Federighi, Apple's senior vice president of software engineering, to move past routine improvements to the assistant. Gurman describes Ternus as "frustrated by what he sees as a reluctance to embrace AI with the same urgency as competitors." He traces the friction to Federighi's reputation as an AI skeptic.

None of this has been confirmed by Apple, and it comes from a single reporter's sourcing. Gurman has a long record of accurate Apple scoops, but readers should treat a private exchange between two executives as reporting, not fact. In a column, 9to5Mac's Ben Lovejoy questioned how a closed-door conversation could have leaked at all.

What Siri is expected to gain, and when

The leadership story matters because it lines up with a fairly specific product timeline. In a separate report, Gurman laid out the next steps for Siri. With iOS 27.3, the assistant is expected to get fine-grained control over Apple's own apps. That would include rotating and cropping images in Photos and organizing tabs in Safari.

The larger change is Siri's ability to work inside third-party apps, which Gurman expects to arrive later, likely in the spring. That would let Siri do things like place a Grubhub order or book a table through OpenTable. Apple first announced these in-app actions in 2024 and then went quiet about them. Gurman noted that Apple began discussing the feature again with the first beta of iOS 27.3.

Beyond that is the agentic version Ternus reportedly wants. Gurman wrote that rival agents, including OpenAI's Dots, Meta's Muse and xAI's Grok, can track goals and decide what to do next without the user starting each step. Apple ultimately wants Siri to work that way, he wrote, and he believes it could happen as early as next year. That timing is Gurman's own estimate, not an Apple plan, and the gap between the two matters.

Gurman also argued that Siri already does a little of this. It can automatically fix weak passwords, which he described as somewhat agentic behavior.

Read together, the timeline suggests Ternus's push will first show up as Apple finishing what it already promised. That seems like the right order. The third-party app hooks are the foundation any real agent needs. A Siri that can't reliably act inside the apps people use every day has no business trying to pursue goals on its own.

Why speed cuts both ways for Apple

Ternus's impatience is easy to understand. Apple builds hardware well suited to running AI models on the device, yet it has trailed rivals on AI software for years. Federighi's team shipped a working Siri on a tight schedule. Even so, the new assistant is a much better voice helper, not a peer to the autonomous agents coming from AI-first companies.

Gurman raised a longer-term threat too. If AI companies pair their software with new devices built from scratch around assistants that see, remember and act, he wrote, Apple's long-standing advantages could start to matter less. That warning is the strongest case for urgency. Apple's power has always come from owning the device in your pocket, and an assistant that lives somewhere else could weaken that hold.

Still, Apple's slower pace has had real benefits. ZDNET argued that it has helped the company's reputation as many people push back on AI. Users have complained about bloated, half-useful AI features at a time when a memory shortage is pushing hardware prices up. Apple's privacy-first approach has also set its products apart from data-hungry rivals such as Meta's Muse.

The rivals' track record shows what speed can cost. In August, OpenAI launched a Mac-only feature for ChatGPT called Computer History, which builds a timeline of a user's app and website activity. It is opt-in and can be switched off, but OpenAI said it does not encrypt the sensitive information the feature captures.

Developer tools have run into similar trouble. Application security firm Apiiro found 104,185 public copies of a Claude Code settings file across 103,617 GitHub repositories. Cybernews reported that more than 2,000 real secrets were exposed in those files, including passwords, API tokens and cloud keys. The leaks happened after developers told Claude Code to stop asking for permission, which saved the full commands, secrets included, to a local file that then ended up public.

Lovejoy pointed to three risks in a faster agentic Siri. An agent can simply fail at a task. It can complete the task by inappropriate means. And it can mishandle the personal data it needs to do the job. He argued that Apple should stay cautious and give users a safer alternative to faster-moving companies.

That argument has weight, but it can be taken too far. The real question is whether Apple can ship agent features that keep its privacy standards, not whether it should build them at all. Apple controls the operating system, the app permission model and the chips that run models on the device. That puts it in a better position than most to build an agent that asks before acting, keeps data local and limits what each app can see. An Apple version of activity tracking, encrypted and opt-in by default, would be a more convincing product than what rivals have shipped.

The competition is not as far ahead as it looks

Rival agents are uneven. ZDNET rates tools such as OpenClaw and Claude Code as highly capable while saying Google's Gemini Spark isn't there yet. Its reviewer found OpenAI's Dot and Anthropic's Duet a bit odd and not ready for mainstream use. That should ease some of the pressure on Apple. The race for a polished consumer agent is still open.

The usefulness of a deeper Siri is easy to picture. Today, Siri's on-screen awareness covers only what is visible at that moment, such as the part of a long email the user has scrolled to. An agentic Siri could read the whole message and pull out what matters. If Apple delivers that kind of help without asking users to give up their data, many iPhone owners may see less reason to install ChatGPT or Claude at all.

The bigger risk for Ternus is not that Apple moves too slowly. It is a rushed agent that fails in public and damages the trust that is Apple's main selling point. Federighi's caution, if Gurman has described it accurately, may be the thing that keeps an agentic Siri worth waiting for.

The first test arrives with iOS 27.3. If its app controls work reliably and the third-party actions show up in the spring as Gurman expects, Apple will have the groundwork in place for the agent Ternus reportedly wants.