I just spent a week living on the edge of the iOS 27 developer beta, and for the first time in a decade, Siri doesn’t feel like a glorified kitchen timer. We’ve moved past the era of “I found this on the web” and into the era of Siri actually doing things—though it still has a habit of tripping over its own shoelaces when it leaves the Apple walled garden.
The Magic of Agentic Shortcuts
The standout feature for me is the new agentic shortcuts generation. In previous versions, building a complex automation felt like writing assembly code with mittens on. Now, you just describe the intent, and Siri maps the App Intents to build the logic for you Source. It’s essentially a system-wide semantic index that understands what your apps can actually do.
I tested this with a few cross-app chains, and it’s surprisingly robust. It feels like the “democratization of automation” that reviewers have been calling for Source. Instead of dragging blocks, you’re just having a conversation with a compiler that actually understands English.
Personal Context: The ISO Audit Test
One of the big promises of Apple Intelligence is “Personal Context.” I decided to stress-test this by looking up when my next ISO audit was coming up. This requires Siri to dig through my emails, calendar, and potentially my files to find a specific date buried in a PDF or a thread from three weeks ago.
While it found the answer, this is where I noticed the first crack in the armor. Siri is incredibly smart about what’s inside Apple’s native apps, but it’s still functionally blind to third-party data unless the developer has specifically built App Intents Source. If your audit details are sitting in a Notion doc or a Slack thread, Siri is back to square one.
The Password Reset Agent: It Actually Works (Mostly)
The most “agentic” moment of the week was testing the password reset feature. I tried it with Spotify. The Passwords app identifies a weak or compromised credential, and you hit a “Fix Password” button. The agent then navigates the site in the background, handles the UI, and updates the Keychain Source.
It worked for Spotify, which was a relief, but don’t expect this to work everywhere. The agent relies on standard HTML structures. If a site uses aggressive anti-bot measures or non-standard JavaScript fields, the agent gets stuck Source. It’s a great start, but it’s not a universal “fix my security” button yet.
The Web Lookup Bottleneck
One major gripe: when Siri has to go to the web, it feels like it’s wearing blinders. It usually pulls from one or two webpages and calls it a day. Compared to “answer engines” like Perplexity that synthesize 10-15 sources, Siri’s native web skills are still very limited Source. You get a snippet, not a synthesis. I’m hoping future updates allow for a more comprehensive crawl because right now, it misses the complete context more often than I’d like.
Image Playground and the Uncanny Valley
Finally, I played around with the new Image Playground features—specifically image expansion and reframing. The tech is impressive; it uses Private Cloud Compute to fill in the gaps when you widen a photo’s aspect ratio Source.
Technically, it works well. The textures blend, and the lighting matches. But personally? We’ve officially entered the Uncanny Valley. Once you start letting AI “reframe” your memories, it’s not really a picture anymore. It’s a hallucination with a high-fidelity skin. It’s useful for fixing a crooked horizon, but as a tool for “photography,” it feels slightly dangerous to the soul of the medium.
