Siri Had a Head Start. Can Apple Finally Make It Count?

Dag Kittlaus thought he was about to take a call from Apple's software chief. Instead, the voice on the line said: “This is Steve Jobs.”

In Kittlaus's account of the encounter, told on the Builders podcast in 2022, Siri had been an App Store app for only about two and a half weeks when Jobs called in 2010. He invited the co-founder to his house the next day to discuss working together.

There is a pleasing irony in that origin story. The assistant now associated with moving beyond apps began as one. Sixteen years later, Apple is still trying to turn the excitement of talking to a phone into the habit of trusting it to do things.

Siri AI began rolling out in beta in English on 14 September. Apple's example has it retrieve a family recipe from messages and email, then add ingredients to Reminders. Fewer searches, fewer switches between apps, less clerical work on a small screen.

The ambition has consequences beyond convenience. If an assistant chooses and coordinates services, whoever controls it gains influence over which businesses get used. But Siri's own history supplies a warning: being everywhere is different from becoming indispensable.

App Store

@AppStore

·Follow

Hello, iOS 27. 👋 Take a tour of the latest updates, including the introduction of Siri AI, an entirely new version of Siri that draws on your personal context and helps you search for content or take action across a variety of apps. See what's new: apple.co/iOS-27

Watch on X

7:02 PM · Sep 14, 2026

11.8K

Reply

Copy link

Read 197 replies

Billions of Requests, an Unfinished Promise

By June 2016, Apple said Siri was handling two billion requests a week across 36 countries. This was already a product operating at enormous scale, more than six years before ChatGPT's launch.

That number measures requests, not people, successful answers or complicated jobs completed. A timer and a demanding research question can each count as a request. It tells us Siri had traffic; it cannot tell us how much responsibility users were willing to hand over.

Nor had the hardware stood still. Apple's advertised Neural Engine capacity rose from 600 billion operations a second in the 2017 iPhone X to 15.8 trillion in the 2021 iPhone 13. Those are chip specifications, not Siri intelligence scores. A faster engine does not tell an assistant what a person means.

Siri's software evolved too. Apple's researchers described deep-learning-based voices in 2017, and iOS 15 brought on-device speech recognition to supported hardware in 2021. Recognising speech, producing a natural voice and sustaining a useful conversation are different accomplishments.

Then ChatGPT arrived on 30 November 2022, offering follow-up exchanges, explanations and help with tasks such as debugging code. Its launch announcement also acknowledged a fundamental weakness: plausible answers could be wrong. The contrast exposed two separate challenges for assistants: holding a conversation and reliably carrying out what that conversation asks for.

Siri's new test is to bring those abilities together. Talking convincingly about dinner is one thing. Finding the right recipe and arranging the shopping without making a mess is another.

Developers Have Something to Lose

In a September 8 report, The New York Times described uneven support in an early developer version. Amazon's shopping app could open and search for paper towels through Siri, but stopped short of placing an order. Gmail had not added support.

Those are pre-release observations, not a definitive inventory of launch-day capabilities. Apple now lists WhatsApp texting and Audible playback as available, with Outlook drafting and other integrations to follow. Support for a particular action also does not establish access to everything inside an app.

The distinction matters to the person holding the phone. An assistant that can complete one request but only begin a similar one asks users to learn an invisible map of its limitations.

“You don't know what you can and can't do, so you end up not doing it,” Creative Strategies analyst Carolina Milanesi told the Times.

RevenueCat's David Barnard expected developers to integrate to different depths. Mobile advertising analyst Eric Seufert raised the business risk: handing the assistant more responsibility could weaken a developer's relationship with customers. Both spoke to the Times.

Who Gets the Customer?

Asking a phone to open a shopping app leaves the destination in the user's hands. Asking it to buy paper towels could hand the assistant a more valuable decision: which retailer gets the order.

That is where convenience becomes a business-model question. An app could remain installed and useful while losing the chance to introduce products, explain a subscription or build loyalty. Developers are being invited to make their services easier to reach through an interface somebody else controls.

For users, the test is practical: can they see why a service was chosen, pick another and undo a mistake? An assistant that saves three taps but obscures those choices has introduced a different kind of work.

Better Siri, Familiar Habits

Apple has attempted to make Siri more central before. The Times recounts its 2011 debut, its later opening to third-party apps and the postponed overhaul promised in 2024. That history makes this launch a test of regained trust as well as new capability.

Early experience offers both encouragement and caution. In WIRED's September 6 account, Reece Rogers found the beta substantially better at retrieving personal information, yet gradually returned to manual browsing. Analyst Avi Greengart reported the opposite trajectory: “I'm finding myself not only using it more, but for more things.”

Those experiences are anecdotes, not an adoption study. Together they expose a gap the grander predictions tend to skip: a tool can improve dramatically without becoming someone's default way of doing things.

Habit also changes the standard by which an assistant is judged. A familiar sequence of taps may be tedious, but the user knows where it leads. Delegating a task introduces questions: did it find the right message, choose the right person and complete the intended action? If checking takes as much effort as doing, convenience evaporates.

What Would Count as Success?

Apple says the new models were built with Google and run on device and through Private Cloud Compute. That architecture supports its privacy pitch; it does not establish blanket access to third-party data or prove that rivals cannot protect privacy.

The launch announcement also confirms daily limits on some server-based features and future paid expansion. It says limits can depend on demand, request complexity and other factors. Lifting them would therefore not, by itself, prove that reliability had improved.

The useful measures are more concrete: how often Siri completes an ordinary task without intervention, how clearly it handles unsupported requests, and whether people keep using it after the novelty fades. Independent testing across different apps would tell us more than another carefully chosen demonstration.

At the iPhone X launch in 2017, Jony Ive described a future in which “the device disappears into the experience.” He was talking about the phone's design, not promising an AI assistant. But it is a useful standard for this launch: the less a user has to think about operating the machinery, the more convincing the experience becomes.

Jobs's phone call helped bring Siri into Apple. Billions of requests established its reach. The next milestone will be harder to count: the moment people stop wondering whether Siri can handle something and simply ask.