Everything About AI
NextFin News - Apple's new chief executive, John Ternus, is pressing software chief Craig Federighi to pursue a more ambitious artificial-intelligence strategy, pushing the company to transform Siri into an autonomous agent that can complete tasks on its own. The push, disclosed in the company's latest software newsletter, marks one of the first clear signs that the leadership transition that put Ternus in charge on September 1 has changed how Apple thinks about AI - not as a feature to ship, but as a persistent layer that acts for users across the iPhone and Mac, and across the services business that now generates $31.6 billion a quarter. It also sets up a direct collision with OpenAI's Dots, the always-on agents launched on September 29, and puts the still-unfolding iOS 28 roadmap at the center of Apple's AI bet.
The Situation: A New Mandate Arrives Before the Product Does
The directive is deceptively simple: make Siri an agent. Not a chatbot that answers questions, and not a voice assistant that opens apps, but software that pursues a goal - book the flight, reconcile the expense report, renew the subscription - without standing by for the next prompt. That is the model OpenAI commercialized this fall with Dots, which the company describes as "remarkably capable, always-on agents built to handle everything." It is now the template Apple's new CEO wants his software organization to chase.
The timing is not incidental. Ternus took over from Tim Cook on September 1 after a unanimous board decision announced in April, ending a 15-year tenure that saw Apple's value climb to roughly $4.6 trillion. His first quarterly report, delivered on November 2, gave him a strong hand: revenue of $114.5 billion in the September quarter, up 12 percent year over year, with net income of $30.7 billion, or $2.10 per diluted share, versus $1.85 a year earlier. iPhone sales reached $57.6 billion, up 17 percent, and services hit a record $31.6 billion, up 10 percent. Full fiscal 2026 revenue reached $478.8 billion, up 15 percent, with net income of $132.2 billion.
But the earnings call also revealed the size of the AI gap Apple is trying to close. When asked about Siri, Ternus said the assistant handles more than 2.5 billion requests a day - a figure that covered Siri overall and did not break out usage of the new AI features at all. That silence is the point. Apple Intelligence and the rebuilt "Siri AI" only began rolling out in beta in English in September, with expansion to French, Japanese, Korean, Portuguese, and Spanish scheduled for the following month. Federighi, speaking at WWDC in June, framed the ambition plainly:
"A dramatically more capable and conversational assistant designed to help users find information and get things done throughout the day."
Note the verb: get things done. The ambition was always agency. The execution has lagged.
Now the company is reorganizing how it builds the software that would make agency real. iOS 28 and iPadOS 28, internally codenamed "Bell," with macOS 28 called "Poppy" - collectively "Boppy" among employees - are already in early development, and the coming release is expected to be far more significant than iOS 27. The shake-up in development strategy signals that Apple is no longer treating AI as a set of features to be dropped into an existing release train, but as the organizing principle of the next operating system.
The Mechanism: Why Agency Is Harder Inside a Walled Garden
The obvious read of this story is that Apple is late. OpenAI shipped Dots. Google has had agentic experiments in the works. Anthropic exposes computer-use capabilities through its models. Apple, which once defined the smartphone era by shipping polished products rather than prototypes, is now chasing. That framing mistakes what is actually happening.
The deeper question is not whether Apple is late to agents. It is whether the agent model can work inside Apple's privacy architecture at all - and whether the company that built its moat on keeping data on the device can build a system that needs to act across apps, across the web, and across days without ever seeing the user's data in the clear.
Here is the mechanism. An AI agent is not a single model. It is a loop: perceive the user's intent, break it into steps, take actions in external systems - browsers, apps, payment rails, calendars - verify the outcome, and recover when a step fails. OpenAI's own GPT-6 Astra system card disclosed that when Dots are given 10 related tasks in sequence - simulating seven to 12 back-to-back tasks in the same environment - the rate of moderate permission deviations reaches 19.7 percent. That means nearly one in five extended task chains produces a meaningful error, from carrying context from one task into an unrelated one to editing documents the user never targeted. OpenAI was direct about it: "dots can still make mistakes and consequential work should be reviewed."
That is the honest state of the art, and it defines Apple's problem precisely. For OpenAI, a 19.7 percent error rate on multi-step workloads is a product metric to improve. For Apple, every one of those failures happens inside a device that the user trusts not to leak their data. The company's entire AI architecture - on-device processing plus Private Cloud Compute, with Apple Foundation Models custom-built in collaboration with Google and its Gemini models - is designed so that personal data is never stored or made accessible, even to Apple. That is a competitive advantage for trust. It is also a hard constraint on the retry-and-recover loops that make agents useful.
This is why the iOS 28 reorganization matters more than the agent announcement. If agency is the goal, the operating system has to change before Siri does. Apps must expose their actions through a semantic layer - what Apple has been building toward with App Intents and the App Toolbox - so that an assistant can invoke "book a flight" or "split a bill" without screen-scraping or brittle automation. The system orchestrator that Apple built into Siri AI, which taps the Spotlight index and App Toolbox entirely on device, is the foundation of that world. But a foundation is not a product. The work of iOS 28 is to turn that foundation into a platform that third-party developers actually build against.
The second-order consequence is where this gets interesting for investors. If Apple succeeds, it does not need to win the model race at all. The company has spent years being second-guessed for not owning frontier models the way Alphabet, Microsoft, and OpenAI do. An agent layer flips that logic: the value migrates from the model to the distribution point that holds the user relationship. Apple's installed base passed 2.5 billion active devices in January 2026. If Siri becomes the front door through which users access third-party AI services - and Apple takes a revenue share on those subscriptions, as the company has been positioning - then the company wins by owning the gateway, not the engine.
That is the structural read. The cyclical read is simpler and less flattering: Apple's AI efforts have been delayed repeatedly. Earlier in 2026, the Siri overhaul was internally targeted for iOS 26.4 in March, then pushed into evaluation for iOS 26.5 in May or iOS 27 in September, with testers reporting that the assistant did not always process queries correctly and responded more slowly than expected. The rebuilt Siri AI only reached beta in September 2026, more than two years after it was first introduced with the promise of personal context and on-screen awareness. If the agent push is just the latest item in a string of slipped timelines, then the news is a management mandate, not a product inflection.
The Counter-Thesis: The Privacy Moat Could Become a Cage
The strongest argument against this strategy is that Apple's core advantage is also its core constraint. OpenAI's Dots run on cloud computers with browsers and plugin ecosystems numbering in the thousands. They can reach anywhere on the web, install anything, and retry failures with resources Apple's on-device architecture deliberately does not have. A privacy-first agent that refuses to let data leave the device - or even Apple's Private Cloud Compute - is competing with one hand tied behind its back.
There is evidence for this concern in the adoption numbers, or rather the absence of them. Ternus declined to separate AI-feature usage from the 2.5 billion daily Siri requests. Apple's privacy posture means the company does not instrument its AI features the way a cloud-native competitor would. That is consistent with its values, but it also means Apple is flying partially blind on the one metric that matters most for an agent platform: how often users actually let it act on their behalf, and how often those actions succeed. A company that cannot measure failure transparently will struggle to fix it quickly.
The analyst case for the upside is nonetheless concrete. Gene Munster of Deepwater Asset Management has estimated that if Apple gets Siri right, a subscription of $20 a month from 100 million users could generate roughly $24 billion a year - about 6 percent of Apple's fiscal 2026 revenue of $478.8 billion. Munster called that a "low-end example just for scale." Services already grew 10 percent in the September quarter to a record $31.6 billion, and gross margin on services runs far above hardware. The math is not fanciful; it is a plausible second engine for a company whose iPhone growth is increasingly dependent on upgrade cycles and emerging-market expansion.
But the falsifying signal is equally concrete. If, by the time iOS 28 enters developer beta in mid-2027, Apple still cannot disclose meaningful AI-agent usage separate from baseline Siri volume - or if third-party developers have not adopted the App Intent / App Toolbox layer at scale - then the agent strategy is a positioning exercise, not a platform shift. Watch for two numbers on the next two earnings calls: AI-attributed Siri requests as a share of the 2.5 billion daily total, and the count of apps exposing actionable intents. Flat or declining on either measure, and the agency narrative collapses back into a feature list.
Scenarios Across Three Time Horizons
Short term - through the next two quarters - expect Apple to frame the story around adoption and developer momentum rather than revenue. The Siri AI beta expansion to five additional languages is the first visible step. The new MacBook Pro, which went on sale alongside this report, is the hardware anchor: a touch-screen machine that only makes sense if the software is moving toward direct, agentic interaction rather than menu navigation. The stock, which closed at $342.37 on November 8, down $1.80, or 0.52 percent, on volume of 43.75 million shares, is not pricing a pivot. At a market capitalization of about $5.0 trillion, it sits below its 52-week high of $355.38 and well above its low of $243.42.
Medium term - into 2027 - the test is iOS 28 itself. If the "Boppy" release delivers a genuinely open agent framework that third parties build on, Apple will have converted its installed base into a distribution channel for agentic services. The base case is that Apple ships a narrower version of the vision: agents that work reliably within Apple's own apps and a curated set of partner actions, with full autonomy arriving later. That is the company's historical pattern - ship the polished subset, expand over two or three cycles.
Long term - the structural question is whether the agent becomes the interface. If it does, the search and app-discovery economics of the iPhone change materially. Users would no longer open apps; they would state outcomes. That threatens the App Store's link-driven discovery model and the advertising business built on top of it, even as it opens the services subscription opportunity. Apple would be cannibalizing its own funnel to own the layer above it. That is a bet Cook-era Apple was reluctant to make. Ternus-era Apple, apparently, is willing to try.
The bottom line: Apple is not late to AI agents because it misunderstood the technology; it is late because agency and privacy are genuinely in tension, and the company chose privacy first. The question for Ternus is not whether he can build an agent as capable as OpenAI's. It is whether he can build one users trust enough to hand their lives to - and whether iOS 28 can turn that trust into a platform before the market decides Apple's moment has passed.
Explore more exclusive insights at nextfin.ai.
