Apple's New CEO Is Pushing Federighi: Siri Must Become an AI Agent

Apple usually takes its time, and in the world of AI that is both its strength and its problem. According to Bloomberg's Mark Gurman, the company's new CEO John Ternus believes the pace is already too slow. Publicly, he is pleased with the launch of the updated Siri AI. But internally, Gurman claims, he is nudging software chief Craig Federighi not to stop at incremental improvements and to move toward Siri as a full-fledged agent. An important caveat: this is a retelling of insider sources, not Apple's official position.

How an Agent Differs from an Assistant

An assistant answers a request. An agent takes a task and carries it through to a result on its own: it opens apps, gathers data, and makes intermediate decisions. Competitors are already going down this path. Claude Code and OpenClaw are described as quite workable, while Gemini Spark, according to ZDNET, is still raw. Newcomers OpenAI Dot and Anthropic Duet didn't impress either: after several weeks of testing, the reviewer got just one useful notification from them.

A simple example with the current Siri. The "screen understanding" feature sees only what you are showing on the display right now. If a long email is open, the assistant knows only about the part you have scrolled through. An agentic version would scroll through the whole thing itself and deliver the gist. The difference looks small, but it is fundamental: the assistant moves from an "ask me about what's visible" mode to a "do it for me" mode.

What Is Missing

Apple also hasn't paid off its old debts. It announced expanded in-app actions back in 2024. The idea is that third-party developers can give Siri access to their apps' features through plugins, so the assistant works more accurately with both first-party and third-party services. Gurman notes that this week, in the first iOS 27.3 beta, Apple finally mentioned this capability again. So in the near term, Ternus's pressure will most likely lead not to a revolution, but to the fulfillment of old promises.

The paradox is that Apple has strong hardware suited to local models, but its software lags behind. At the same time, the slowness partly plays into its reputation. Amid growing discontent with AI "bloat" in products and rising device prices due to memory shortages, many people like a restrained and private Siri. For comparison: ZDNET called Meta Muse the worst agent in terms of privacy. And the ChatGPT Computer History feature, which builds a timeline of activity on a Mac, raised questions because OpenAI does not encrypt the sensitive data it sees.

Limitations and Practical Takeaways

Apple's main risk is not that Siri will lose to ChatGPT or Claude right now. The gap is still large, and users won't rush to delete those apps. The risk is that AI companies are building their own devices, designed around assistants that see, remember, and act. Gurman writes that in that case, the familiar advantages of Apple's ecosystem could matter less.

What to do about it:

• iOS developers. Follow the iOS 27.3 beta and the documentation on in-app actions. If Siri gets proper access to third-party app features, products that opened up their capabilities first will have an advantage in voice and agentic scenarios.
• Product managers. Don't build strategy on rumors about timing: for now this is insider info, not an announcement. But do check which actions in your app an assistant could perform on the user's behalf.
• Founders. Privacy remains a real differentiator. If Apple builds agentic features with local processing and opt-in, the niche of the "AI agent that collects everything about the user" will become noticeably narrower.

Bottom line: Apple isn't promising an agent tomorrow, but its internal course, judging by the leak, is changing. What to expect is not a splashy presentation, but the gradual appearance of those very plugins and in-app actions.