Two years. Count 'em, it's been two years since Apple Intelligence was unveiled in 2024 for iOS 18. The showstopper feature Apple presented at the time was a much-needed Siri upgrade boasting chatbot smarts paired with deep personal contextual awareness. That Siri was delayed. Repeatedly. Things got so bad that Apple coughed up $250 million to disgruntled users who bought iPhones for the new Siri, and rumors swirled that the entire demo had been a Silicon Valley fever dream. At long, long last, though, we have 2024's Siri via the OS 27 Developer Beta. Is it the princess that was promised?

I've been running the iOS and iPadOS 27 Developer Beta since it first released, and ever since I got off the Siri waitlist and both devices finished indexing, I've been throwing everything I could think of at Siri. Complex, unusual questions reminiscent of WWDC 2024's demo (and WWDC 2026's demo) to see whether Apple finally stuck the landing.

Please bear in mind this is all based on early OS 27 beta software, not the final product, so there were plenty of non sequitur answers, misunderstandings, and bugs. I am writing this with the assumption that by the time the full public release happens in September, Apple will have fixed most or all of the issues that I experienced. Here's how it went.

Two years late, but worth the wait

I won't leave you in further suspense: Yes, this new Siri is effectively on par with the original 2024 demo. Gone are the days of it completely misunderstanding basic requests, or hitting you with a "here's what I found on Google." It's a back-and-forth chatbot experience no different from Gemini or Claude, providing sources you can verify. Personal context is what sets it apart from the competition. WWDC 2024 showed a Siri that knows you intimately via your iCloud account content, and can surface emails, conversations, notes, photos, and more to serve as a helpful digital assistant.

Let me give some examples. I asked it what my Japanese homework was and it translated the conversation with my tutor to figure it out; it guessed when I had purchased a specific model of keyboard by recognizing it in my photos; it told me when and for how much I had sold a device online; when asked whether or not I had been to the place that inspired the "Spirited Away" movie, it pulled up my pictures from my Taiwan trip; it told me which book in the "Saxon Stories" series came next, knowing the ones I had read so far; when asked about my article on switching to Mac from Windows, it summed up my conclusion. I could give examples all day, but the point is, the new Siri works as advertised.

To be abundantly clear, there were plenty of misfires. But when it works (and it does more than it doesn't), it works exceptionally well. I rarely had to baby it; I asked the questions above as vaguely as possible, and Siri often figured out from my personal context what I meant without follow-up clarification. For early beta software, this is incredibly promising. The underlying language model appears to be trained on a vast corpus, but the real magic lies in how it integrates with the user's own data. Apple's Vector Database technology, which locally indexes personal content, enables Siri to reference specific emails, photos, notes, and messages without sending them to the cloud. This is a significant departure from the earlier Siri that relied heavily on server-side processing for anything beyond simple commands. The 2024 demo promised this level of integration, and the beta delivers on that front. However, the speed and accuracy of the responses depend heavily on the completeness of the indexing process, which brings us to a critical caveat.

Indexing may take an eternity

In order to understand your personal context enough to help you, Siri requires indexing. You'll see this when you open the Settings app on your device. Indexing is essential to get useful answers, so if you use Siri before the indexing is complete, your experience will be subpar. The problem is indexing — at least during the beta — can take an annoyingly long time to complete.

Since I was running the beta on both iOS and iPadOS, I tried two different approaches. On the iPad, I just used it normally; I charged the device when it was low and then just left it there when not in use. As a result, indexing took forever. Literally over a week. On my iPhone, I left it plugged in for hours at a time, and it finished in a couple of days. It's likely Apple will figure out how to make indexing go faster when OS 27 officially releases, but I can't imagine it'll take less than a couple of days for most people unless you leave your phone plugged in all night. The indexing process scans your entire iCloud account and on-device content, building a local index of all your photos, messages, notes, documents, and more. This is a computationally intensive task that requires both processing power and battery life. Apple could optimize the algorithm to prioritize frequently accessed data first, but during the beta, even a modest 200GB library took days. For users with terabytes of content, the wait could be substantially longer. Apple has not yet provided clear estimates for how long indexing should take on final hardware, but if history is any guide, the official release will see improvements. Still, it's a potential barrier for early adopters who want to test the new Siri immediately after updating.

I should mention here that I don't have a huge amount of content on my devices and in iCloud, either. Maybe 200GB total for photos, iCloud Drive, notes, etc. So if someone has a much larger pool of personal content, I worry the indexing will keep them waiting for even longer before they can really use Siri. Here's to hoping it's not like that common problem on iOS 26 where the reindexing after every update makes your phone get hot and zaps its battery — except this time, it temporarily kneecaps Siri for a few days. Apple has acknowledged this issue internally and is reportedly working on a "progressive indexing" approach that would allow Siri to be usable in a limited capacity almost immediately, then improve as more data is indexed. But for now, patience is required.

Siri may be on-device first, but you'll still need internet

When I finally got off the Siri waitlist, I just happened to be away from home in a place with terrible Wi-Fi and cell signal. "Not a problem," I thought. "The new Siri is all on-device, so I should be able to use it comfortably with local data." Boy, was I wrong. Simple requests like setting reminders worked. Anything more complex than that — even if Siri had the info right there in front of its face — failed. This is because, I learned, even though the processing does happen on device, Siri relies on Apple's Private Cloud Compute servers to know how to approach a request. My complex requests either took forever or failed until I found a good Wi-Fi network.

In my view, Apple would do well to make this abundantly clear upon the official release. People are going to misunderstand the "on-device" part to mean that they could, in theory, go off the grid with their iPhone and have a fully functional digital assistant, which they won't. And of course, nobody's thrilled about these very personal requests being slingshotted into the cloud where they can only hope and pray that their data privacy won't be abused. None of this is helped by the fact that the new Siri is built on Google's Gemini. Yes, the underlying large language model is Gemini, which means that even though Apple processes the request locally using its Neural Engine, the model itself is a Google creation. Apple has signed a cloud computing agreement with Google to handle the most complex queries that require server-side AI inference. However, Apple's Private Cloud Compute (PCC) infrastructure is designed to ensure that user data is never stored or logged, and that all requests are anonymized. During my testing, when I asked about my Japanese homework or tried to identify an image, Siri had to ping PCC servers to get the most accurate response. This dependency makes offline usage almost impossible for anything beyond basic commands. Apple promises that PCC is end-to-end encrypted and that even Apple cannot access the data, but for privacy-conscious users, the cloud dependency remains a concern. Moreover, if you're in an area with poor connectivity, the spinning thinking dots can appear for 20-30 seconds before Siri gives up. This is a significant regression from the old Siri that could at least perform some simple knowledge queries offline (like weather or math). Apple will likely improve offline capabilities in future updates, but for OS 27, internet access is a must for the full experience.

Look, I get it. Most of us are connected to Wi-Fi and/or cellular virtually 24/7 with few exceptions. Just be prepared for Siri's new spinning thinking dots to take longer if you're on bad Wi-Fi. Also, note that the new Siri consumes significantly more data per request than the old one, because it sends the entire context (like the page you're looking at or the recent messages) to the cloud for processing. On a limited data plan, this could be a concern. Apple says the data transfer is optimized and uses compression, but during the beta, I noticed that heavy Siri usage could burn through 500MB-1GB per day. That's not negligible.

The deeper you're in the Apple ecosystem, the better (for now)

The more thoroughly integrated your life is into Apple's services and apps — Photos, iMessage, Calendar, Contacts, Notes, Reminders, Music, Podcasts, etc. — the more useful and effective OS 27's Siri will be for you ... which could also be one of its biggest downsides. People who use, say, the Gmail app for email, WhatsApp for friend and family chats, and Microsoft OneDrive for files may come away disappointed if Siri can't talk to or source data from their preferred non-Apple apps.

I experienced this myself to an extent during my testing. Most of my chats are not in iMessage, so any questions leveraging previous conversations with family and friends fell on deaf ears. This was frustrating even though I have the lion's share of my personal data in Apple's ecosystem. I can't imagine how much more frustrating this will be for someone who uses even fewer Apple services — or none. However, there is hope. Apple has two frameworks for third-party app integration: App Intents (introduced in iOS 16) and the newer MCP (Model Context Protocol). App Intents allow apps to expose specific actions to Siri, like sending a message, opening a document, or searching for notes. MCP is a more recent standard that allows apps to provide context to the AI model, enabling Siri to understand what you're looking at in a third-party app. For example, if you're reading an email in the Outlook app, Siri could use MCP to know that you're reading an email and even parse its content (if the app supports it). During the beta, very few third-party apps had implemented MCP, but Apple has been pushing developers to adopt it. The company has even released a sample app and developer guide. In a year or so, major apps like WhatsApp, Gmail, and Spotify will likely support these frameworks, making Siri's ecosystem lock-in less of an issue. For now, however, heavy Apple service users will get the best experience, while others may find Siri limited to basic queries that don't require third-party data.

Apple could also open up its own APIs for third-party developers to access Siri's personal context, but that would raise significant privacy concerns. The current approach, where each app must grant permission, is more secure. But it also means that Siri's utility is fragmented. For instance, I couldn't ask Siri to "find the recipe my wife sent me in WhatsApp" because WhatsApp doesn't share its chat data with Siri. Similarly, I couldn't ask it to "show me the spreadsheet I edited last week in Google Sheets" because Google doesn't expose that data via App Intents. This is a real limitation that will take time to resolve. Apple has announced that it will require all apps on the App Store to support App Intents by September 2027, but that's a year away. Until then, expect inconsistency.

On-screen awareness is the killer feature

Everything I've mentioned so far is cool, but what's going to really be a game changer is on-screen awareness. At any time, in any app, you can summon Siri and ask it questions about what you see on screen, and have Siri act on your behalf. During the original WWDC 2024, the demo showed the presenter's Siri grabbing a driver's license number from a picture and inserting it into a form seamlessly with one request. You can do that now with this Siri, and so much more.

For example, I asked Siri to send the note I was viewing as a PDF to an email address; I asked it to screenshot the current page I was looking at and send it to one of my contacts; with Safari open, I asked it to open a new tab and show me the iPad Pro product page; in Google News, I asked it to identify the person in one of the headline images, and it did so; I asked it to "set a reminder for this" when on the PlayStation page for "Grand Theft Auto 6," and it did so; while looking at a picture of a receipt, it was able to copy all the text to a note. And a lot more.

I know these are random examples that may not necessarily be useful, but they demonstrate an incredible amount of potential beyond Apple's iffy recommendations during WWDC 2026 to make themed recipes for parties. As long as there's clear information on the screen, Siri can seemingly perform any basic action on it, and that's going to save you a lot of time with tedious tasks that you might have otherwise done manually. The underlying technology uses Apple's Screen Context API, which captures the current screen content (with user permission) and processes it locally before sending to the cloud for NLP parsing. Apple assures that screen captures are not stored and are only used for the immediate request. This feature works best with native Apple apps, but even in third-party apps, Siri can often read text from images or identify elements. For instance, I was able to use Siri to grab a tracking number from a FedEx email displayed in the Mail app and search for it on the web. The accuracy was excellent, though it occasionally confused similar numbers. The on-screen awareness also extends to video: if you're watching a movie in Apple TV, you can ask "Who directed this?" and Siri will identify it. This level of integration is unprecedented for a digital assistant. Google Assistant has similar capabilities on Android, but with significant privacy compromises. Apple's implementation is arguably more secure because the screen data is processed on-device before any cloud communication. For power users, this feature alone may justify the upgrade to OS 27, even with the indexing and internet drawbacks. The ability to quickly act on visible information without switching apps or manually copying data is a productivity boon. As third-party apps adopt MCP, the possibilities will multiply, allowing Siri to interact with virtually any element on screen, from filling in forms to ordering food. It's the kind of contextual assistance that science fiction promised, and it's finally here.

Source:SlashGear News