Is Siri Actually Good Now? Apple's iOS 27 Overhaul Explained

Siri has been a punchline for so long that skepticism feels like the responsible position. So here's the direct answer to the question everyone's asking after WWDC26 this week: is Siri actually good now? Not yet. But for the first time, there's a credible structural argument that it could be and that argument is worth taking seriously, with clear eyes about where it breaks down.

The case rests on something Apple hasn't had before: a Siri that can read your screen, search inside your apps, and take action on your behalf without you explaining the context first. Apple Newsroom called it "an entirely new version of Siri" this week. Essentially the same promise was made at WWDC24 two years ago, then delayed from 2025 into 2026, as Indian Express reported five months ago. The developer beta is open now across iOS 27, iPadOS 27, macOS 27, and visionOS 27; ordinary users get an English beta later this year.

Whether this version of Siri actually delivers depends on three things lining up simultaneously. Understanding what those three things are and the honest odds on each is more useful than either Apple's keynote optimism or reflexive skepticism.

What Apple claims Siri can do now

Apple organized the new Siri around three concrete capabilities, each corresponding to something users have wanted phone assistants to do for years.

Siri can now reach inside your apps and pull out meaningful content. Ask "When and where is my next meeting?" and it pulls the answer from your calendar directly, understanding which meeting is relevant and what details to return, without redirecting you to a search screen. Then there's action across apps: say "Send my latest report to Mary" and Siri identifies the document, resolves the contact, and sends it. The assistant handles the language; the app handles the execution. Both described by Apple Developer, WWDC26 this week.

The third capability is on-screen context. Commands like "Get me reviews for this product" or "Text my wife this conversation" work because Siri can read what's currently on your screen and pass that content to another app, no copying and pasting required. A dedicated Siri app, with iCloud-synced conversation history across devices, would also address the assistant's long-standing statelessness problem Siri has never remembered anything between sessions, which has been a fundamental usability issue independent of any model quality concerns. Siri can now also search the web and synthesize answers for queries that need current information, per Apple Newsroom this week.

These are reasonable things to want. The question is what's actually powering them.

Why the architecture appears different this time

The reason the new Siri may be more capable than its predecessors isn't better AI research. Apple appears to have addressed a structural problem in how Siri accesses app data. Siri could always launch apps, set timers, play music. What it couldn't do was tell you which emails were about a specific project, or find the document you'd been editing on Tuesday, because apps didn't expose that content in any form Siri could work with.

iOS 27 changes this through App Intents, which Apple Developer, WWDC26 describes as the foundation for everything in the new Siri experience. Developers use it to define what their app contains and what it can do. Those definitions get indexed into a system-level semantic store, which lets Siri match by meaning rather than exact keywords so "show messages with Flare about movies" works even if those words never appeared together in a single thread.

The developer incentive is real: actions defined through App Intents automatically surface across Shortcuts, Spotlight, Widgets, and Siri, so integration work pays off across the whole OS rather than just for voice commands. For on-screen awareness, apps annotate visible content so Siri knows what the user is looking at, and a separate content-transfer mechanism lets that content move to another app for action, while keeping the originating app in control of how it's handled.

This is at least a more tractable problem than asking Siri to infer everything from scratch. Siri isn't trying to guess at app content anymore; it's asking apps to teach it, using structured definitions. That's a solvable engineering problem. Whether it gets solved in practice depends on how many developers actually do the work.

Siri improvements in iOS 27 depend on more than Apple's demo

Siri will feel meaningfully better only when three things are true at once: Apple's underlying model handles the request correctly, the app in question has adopted the relevant schemas and intents, and the request falls within a domain the system explicitly supports. When all three align, the experience should be substantially different from the Siri of two years ago. When one is missing, you get the old Siri or worse, a confident wrong answer.

Apple's model is the hardest variable to assess from the outside, and the history is the most concerning. TechSpot, citing Bloomberg-sourced reporting four months ago, found that internal testing ahead of the now-missed iOS 26.4 launch uncovered stalling on complex queries, slow responses, misinterpreted instructions, and inconsistent contextual memory. In some builds, Siri reportedly fell back to OpenAI's ChatGPT integration even when Apple's own models were available to handle the request. 9to5Mac noted eight months ago that Apple had characterized three specific new features as taking longer than expected, with no new delivery date offered. The delays since then have bought time for fixes. Whether those fixes landed is unknown; no independent hands-on testing with the new iOS 27 build has surfaced publicly as of this week.

App adoption is structural and will shape the launch-day experience directly. First-party Apple apps Mail, Calendar, Messages, Notes will almost certainly implement the new frameworks fully; they have the most incentive and no adoption barrier. Third-party apps are another matter. Many won't have done the work by the time the consumer beta ships, which means Siri's quality ceiling on day one of iOS 27 will be defined by which apps showed up, not by what the framework is theoretically capable of.

Supported domains are the third constraint. App Intents schemas are organized by category mail, messages, photos, and similar groupings, per Apple Developer, WWDC26. Requests that fit a well-defined domain and land in a supporting app should work gracefully. Novel requests, cross-domain tasks, or apps that haven't adopted the relevant schema may still behave unpredictably.

A word on what's powering the model underneath all this: TechSpot reported four months ago that Siri's rewritten architecture is increasingly intertwined with Google Gemini technology, with computationally intensive tasks routed to Google's cloud infrastructure. Apple and Google confirmed a formal multi-year AI collaboration five months ago via a joint statement cited by Indian Express, noting that next-generation Apple Foundation Models will be based on Gemini models and cloud technology. Apple maintains that user data stays on-device or passes through privacy-protected servers. What that means, in plain terms, is that "Apple Intelligence" is a product brand resting on infrastructure that is partly Google's which is neither a disqualifier nor a reassurance about quality, but it's worth understanding before taking the branding at face value.

What to actually test when the beta ships

The developer beta is open now; the public English beta arrives next month, per Apple Newsroom this week. Here's what that beta will actually tell you about whether Apple delivered.

First-party app integration is where the experience is likely to be strongest. Ask Siri about calendar events, Messages threads, Notes. These are the apps Apple rehearsed and the ones with the most complete framework adoption. Web-based answers pulling current information from the web and synthesizing a response are also worth testing early, since that feature doesn't depend on any third-party developer doing anything.

Third-party app actions and cross-app workflows will probably be patchy. If you're trying to use Siri inside a productivity or communication app that isn't Apple's, the experience will depend entirely on whether that specific developer adopted App Intents. Assume many haven't, at beta launch.

The feature most worth watching carefully is persistent conversation history synced across devices. It sounds mundane next to on-screen context and cross-app actions, but statelessness has been Siri's most persistent practical failure. An assistant that forgets everything between sessions creates friction regardless of how capable the underlying model is. If that works reliably, it's the most meaningful quality-of-life improvement in this release.

Two constraints worth noting: Siri AI won't be available in China at launch due to regulatory requirements, and language support beyond English will expand on an unspecified timeline, per Apple Newsroom this week. Also, as 9to5Mac observed eight months ago, Siri won't be judged against the chatbots of 2024 it'll be judged against whatever ChatGPT, Gemini, and their successors look like when iOS 27 actually ships to users.

The architecture Apple showed this week is more coherent than anything it has shipped under the Siri name. Structural problems that made previous versions feel like dead ends appear to have been addressed at the design level. But architecture is a prerequisite, not a product. The real test starts when people who didn't build this thing use it on real devices, for things they actually need done.

Interested is the right posture. Convinced would be premature.