With Siri AI, Apple Are Falling Further Behind

Developer at a desk watching Siri's colorful waveform show an error message, illustrating Siri AI's limits with custom App Intents

Field notes from Pilcrow: what building Siri AI support with iOS 27’s App Intents actually turned up.

I spent this past weekend building Siri AI integration for the Pilcrow iOS app, using the new App Intents and App Schemas that shipped with iOS 27. Apple’s pitch for “Siri AI” is that it’s conversational now, and that developers can extend it with their own app’s features, not just a fixed list of Apple actions.

Turns out that’s only half true, and the half that isn’t true is the interesting part.

The setup

I built a proper custom action: ask Siri what’s on a specific list in the app, and it reads the items back to you. Good title, good description, registered as an app shortcut, the works. I tested it by asking “what’s on my shopping list in Pilcrow.” Siri couldn’t find anything in Pilcrow at all, and instead offered me a completely unrelated note from Apple’s own Notes app.

What was actually going on

Digging into this was rather painful — Siri AI isn’t available in the simulator — but I found a rather telling post from a developer on Apple’s own forums who’d run into exactly the same thing testing a sample app:

“My original assumption was that indexing data from your app would make it available to Siri, but it only seems to show up in on-device search and not in conversation with Siri itself.”

“General questions about the indexed data fail to show any results at all in Siri… it seems the only way I can get Siri to show any of the referenced entities is by using the exact phrasing (or extremely close to it) in one of the donated shortcuts.”

The actual state of the Siri AI nation as of today: a custom app action only gets picked up by Siri if you say something very close to the exact phrase you registered for it, which is basically the old SiriKit model with a new name. The only way to get genuinely open-ended, ask-anything style understanding is if your feature happens to fit one of the dozen or so domains Apple itself predefined — Notes, Reminders, Calendar, that kind of thing — and that underlying framework is a few years old at this point, not new for this release.

There’s another thread on Apple’s forums where a developer explicitly asks for a general-purpose custom domain “to aid understanding from the LLM.” No sign that’s shipped. So if your app doesn’t fit neatly into one of Apple’s boxes, you’re stuck writing near-exact trigger phrases and hoping, not describing what your app does and letting Siri figure out how to use it.

Compare that to Claude

This is also why Siri AI isn’t taking Claude’s spot as the main AI tool on my phone anytime soon. Pilcrow already has an MCP server, and when I ask Claude to do something through it, Claude just reads what the tool does and works out how to use it on its own. No exact phrasing needed, no waiting on the tool’s author to shoehorn it into a category someone else defined. That’s the actual bar for an assistant you can hand real tasks to across a pile of different apps.

Siri AI is better than it used to be, but right now it’s still choosing between two narrow paths: match the phrase a developer registered, or hope your use case fits a box Apple built a few years back. Until that changes, it’s a decent voice assistant for a fixed set of things, not a general-purpose one.

What this means for Pilcrow

I’m removing Siri AI support from the Pilcrow iOS app until it can actually be useful for our users. At present, it would just frustrate people.

Leave a Reply

Your email address will not be published. Required fields are marked *