Siri was the first widely available AI bot but they forgot to open it to developers. The ecosystem would have grown a million fold in all these years
Never underestimate the power of a bored programmer
Siri turn on the lights
Siri code a Doom clone in Swift
Siri design a landing page for a Consulting Agency
Siri summarize the latest news about AI
Siri find a solution to the Riemann hypothesis
Now that's an agent I'd use. Can we start with Siri for the terminal?
Amazon did exactly this with Alexa skills and it hasn’t exactly been a booming success. Turns out that all of the things you listed as examples are stuff that most consumers don’t actually want or will use.
“AI as a chat bot” has always been cool and fun in sci-fi movies, but it’s the least imaginative real world use for it. We need to move past it.
> $ Siri 'List the files in this folder'
Why in the world would I ever use this instead of just ‘ls’?
The problem with Alexa is that by the time it appeared anywhere other than Echo devices, it was already deeply enahittified. Siri is in your pocket, on your wrist, and in your ears. Alexa was stuck on your kitchen counter asking you if you wanted to try Amazon Music.
I'd settle for a siri that can at least turn the lights off properly.
Google assistant on the nvidia shield left Siri in the dust years ago. Siri can't even chain basic commands in your smart home, it can't retain any kind of context (e.g. dim the lights...a bit more).
Siri is Gemini now, so expect similar stories from both major phone OS providers going forward, since we have lost more diversity in the mobile market (seems anti-trustworthy to me)
I tried "turn off the lights in [room a] and [room b]" in the new Siri app and rather than Siri saying they can't do compound commands they just turned off the lights in one of the two rooms listed.
So, progress?
Any agent with the Home Assistant MCP hooked up can do it fine.
$ Siri 'rename this thousand of photos in these folders describing their content and make sure to use the same name for duplicates regardless of their size'
The thing is, Apple will have never lost this opportunity. The distribution advantage of the user base is so deep they can basically always implement this at will.
And you can also sideload apps (only in the EU), change you default navigation app from apple maps (only in the EU), and use more features of the NFC reader (only in the EU).
The code for all of those things is there, but not enabled if for most iOS users. There's a non-zero chance this code is there for Apple to try and convince the EU to let them release a Siri AI, and then just like sideloading have some impossible-to-actually-use alternative path to make the DMA happy.
It’d be short sighted to not have a wrapper around which AI to use, even without implementation for them. Like, having an abstract parent class with a Siri implementation of it would be baseline good software engineering.
Yeah; plus we know they're designing with potential future local-only Vs. hybrid Vs. cloud-only in mind depending on hardware/market changes. So having it be this flexible buys them a ton of future potential.
People have speculated that with FPGAs or even an ASIC you could take some smaller models we have today, turn the chip into essentially that model-as-hardware, and get some really compelling performance/power-usage characteristics.
At that point running a GOOD quantized Q8 model on a phone is not beyond the realm.
This is such a funny comment. I’d been hearing about the un-repairability of Apple products for years and I believed it. Then I went to Shenzhen and walked past an entire floor of individuals doing all sorts of “impossible” repairs right in front of the customer.
I’m not even sure how they would implement these features without them being interoperable. All the AI systems use pretty similar if not identical API standards.
My favorite Siri behavior is telling it, on my phone, to turn on a focus state only for it to transfer the request to a nearby HomePod that tells me it can't help with that. I hope the new changes to it are an improvement, but the bar is currently buried in the dirt.
Apple Foundations Models Framework already aimed towards some generalized API for models. Personally I like it, just as I would be a little happier if instead of codex/claude/antigravity (rip gemini cli) I would have single standardized interface.
With mobile devices especially where ideally models could be shared across several apps.
But this specific code I'd assume is targeted to comply with EU regulations more than anything.
OT:
Since I've moved to Spain my principal Apple account is still in my home country but I've made another one for local app-store (since some local apps aren't available if you're not in the Spanish store).
iPhone is acting really strange, I'm logged to my principal account for iCloud+ and anything and local App Store.
If I log back to my principal account, I loose HCE mode and my building NFC app stops working until I re-log in the Spanish Apple Store.
If I'm logged at the Spanish account, I loose Siri AI which is blocked in the EU...
That’s actually a good solution. Create a platform where app actions can be invoked with a permission model by any AI model.
Maybe apple can even have something similar to MCPs for tool invocations of apps and publish it so all models can actually be trained to use that framework
I actually think this is what they ended up doing - just regular conversations with tool calling into your various apps, behind a Siri AI API. not HTTP based mcps (for now) but rather local ones.
Once Siri gets MCP support, it’ll be my main orchestrator.
I built an app that lets me run terminals on my MacBook. It’s a Next app so I can go to that website on my phone using Tailscale. It has a built-in MCP server, so I can connect it to ChatGPT and have it create terminal sessions, orchestrate terminal sessions, all that good stuff.
Down the line, I’d like to be able to connect it to Siri and just talk to my phone and it can control terminal sessions and projects that I’m working on on my laptop.
I'd love it if Siri does, but I don't expect big corps to keep supporting MCP since MCP gives so much control to the end user, can't charge for eg. Google integration when someone can just add an MCP to handle it.
This is a very clever platform play. If you make Siri / your mac the central access point for all AI related workflows and make models interchangeable, then Apple has a huge weapon in the AI race.
I don't disagree, but I do feel the need to point out this was not Apple's plan A. Luckily for them they have so many tools and options at their disposal, that they are able to massively fail at developing their own models and still end up in a good position.
Apple “genius” has been to accept their plan A failed (the original Apple Intelligence), and instead of doubling down, let the others explore the space, then offer a good interface (I say that, I haven’t seen if it’s actually good).
So many other actors in the industry would have just doubled down and follow the race to the bottom
I imagine that Google assistant only makes money for Google because they’re an advertising company. Alexa loses money for Amazon and it’s becoming more gated to Prime subscription features as a result. Amazon thought everyone would be buying stuff with their voice and that future never panned out.
It turns out that voice assistants aren’t all that important. Customers aren’t willing to pay for them and they’re more of a nice to have convenience feature.
They didn’t invest in Siri and became the most valuable company in history. I don’t think Siri being better would have mattered much to be honest. Not joining the financial black hole that is AI training seems to have benefited them well so far. But yes, it’s definitely a case of failing upwards.
Adding quotes around “genius” as it seems you misunderstood what I meant
Meh, AI is deep in it's diversification stage -- there are soooooooo many players in the game right now and basically one maybe two will survive once the dust settles. All these frontier labs are fighting tooth and nail to be the one left standing. Apple is making a measured approach saying "hey, we're probably not going to be the Google of AI, but the smartest thing for us to do is to put infrastructure in place so that once there is a contraction (which there will be, think the search engine landscape in the 2000s) we're relatively ahead for relatively little effort/cost."
I don't know why you're getting downvoted. You're spot on.
Almost everyone developing their own AI is doomed to fail. I would not be surprised that when the dust settles, Microsoft CoPilot and Google Gemini will be left standing while OpenAI and Anthropic fade away. They will get what they need and then throw them away.
"Embrace, extend, extinguish" is the same playbook for decades, and you can only do that when you make money. Yes, making money still matters and post-capitalism is a big fat lie.
Handle the query on device. If the device cannot handle the query, push the query to Private Cloud, If Private Cloud cannot handle the query, push the query to a third party of the users choosing.
Which is pretty much the same as what they announced a few years ago.
Apple underestimated the amount of time and work needed to get to work and really the only bit that I think changed is that Private Cloud (and possibly the on device model) is a white label Google model rather than being developed in-house by Apple.
Mark Gurman seems to get extraordinarily accurate and detailed leaks. I wonder how Apple got this leaky. Pretty much everything about the Duo was known months in advance.
Well, either corporate security got incompetent or they intentionally let select leaks exist to "test the market response". Established companies tend to get more risk averse overtime.
Third-party calls won't eat into Apple's compute, so it's not entirely against the interests of Apple as they put caps on people's AI use.
I wonder what the margin on this stuff is for them:
Some Apple Intelligence features in the Home app, such as
video summaries, are available for a limited number of
cameras depending on your iCloud+ plan — up to 1 camera
in the 2 TB plan, up to 2 cameras in the 6 TB plan, and
up to 5 cameras in the 12 TB plan. Supported camera
counts and other iCloud+ plan features may change over time.
I don't want to tell Siri to ask Claude or GPT - I want Siri to automatically delegate to another model of my choice when it can't answer something, instead of showing irrelevant web searches.
In my opinion the OS should broker the communication between harnesses (which is part of Siri), LLMs and app functionality. It can take into account the permissions, the user could configure which LLM to use at an OS system settings level and complex workflows can integrate permissions like FaceID only once as the harness then interacts with APIs exposed by applications (arguably already there with app intents).
Feels like this is based on the idea that any model is interchangeable with any other, while all of them are better at some stuff, worse at some stuff.
Besides, the modern concept of "agent" is what, some year old? Bit fast to ossify stuff in the OSes already.
I think that's the end goal, at least looking at where their hardware is going. Or flexible routing with easy tasks running locally and only the more complex ones hitting the cloud
This appears to be the current feature prototype intended for the EU and potentially China. Apple has chosen not to release the new Siri AI in the EU because our Digital Markets Act requires them to allow interoperability. I.e. that users be allowed to choose another AI provider.
Ostensibly they'll activate this in the EU when it's ready. Given the current functionality, it looks ready (or close to being ready). My guess is it is ready and Apple is doing their usual negotiation tactics with the EU to see who blinks first.
Apple proposed a solution to the EU after reportedly extensive talks and asked the EU if it would satisfy them.
The EU said “build it then we’ll tell you”, which is crazy.
The problem is not having Siri use different models beneath. That’s not what the EU wants. They want Siri to be replaced by Gemini/etc completely. Which fits with the DMA.
Apple’s proposal was for something akin to an agent that could gather the necessary data for the 3rd party AI to use. Because what they refused to do is let any other AI have unfettered access to everything whenever it wants.
So I suspect this is really for Apple’s benefit. I don’t think the EU would be happy if you could still use Siri, but it was grok underneath. Because that would still be Siri.
My secret hope is that Tim Cook was the main weight behind cutting off half the new features from Apple software in the EU an attempt to weaponize their users against legislators. Even features that have had analogues where nobody would've ever asked for interoperability (i.e. Handoff/Continuity exists in the EU, but iPhone mirroring doesn't)
This is not what EU regulators asked for. They want first class access to user data and devices for would-be alternative providers… which Apple simply won’t allow outside PCC.
This is not that. This is expanding the Extensions that you already have. You can look at it in your Settings already.
All of which is why EU users are being left behind on the new Apple Intelligence (for now).
There’s a reason everyone knows the various wake words.
All the big companies know how suicidal that would be. They already get accused of it by people who don’t understand how targeted web advertising works.
Some of the new AI start-ups don’t seem to understand that, but Apple absolutely will not make a mistake that stupid.
They could instead make a good assumption that you'll search the web, open the Yelp or OpenTable app. Take your phone to your favorite restaurant and/or pay with a credit card at the grocery store.
If you walk to a taco truck without your phone and pay with cash, you probably don’t do that every time and you generate enough data through other means to be accurately advertised to.
Apple is highly sensitive to dangers to its reputation. They would nuke any service that seemed to be spying on users, just like they revoked Facebook's cert back in 2019 over the same issue.
Just playing the devil advocate: what happens when the answers from a not-Apple LLM are garbage? Who the users are going to blame, Apple or OpenAI, Anthropic, etc..?
I'm not 100% on Apple side on this, I would love to chose whatever I want, even local models, but I get why trey trying to control 100% of the user experience. If they could trust any provider with customers data, they would let them in, since they're not competitors. They would even let Apple not spend a ton of money on developing LLMs....
Are people blaming Apple when Google search in Safari gives garbage results? Some probably are, but that really shouldn’t be a relevant discussion, in particular when the user explicitly chose which AI backend to use.
Well, we can generalize this a bit to model routing in general. Actually, just AI in general - even when directly interacting with GPT we haven't figured out good protocols yet.
Agreed. I'm not looking forward to iOS 27 where basically the only option is this new Siri. I was perfectly content with using the old "dumb" Siri to set timers, turn on a light and control my Apple TV and nothing else.
If this consistently has happened at every place you've worked in your career, I think you're luckier than you realize. Independent of the implementation, it's also still interesting as a sign of what might be supported for users down the line; I don't agree that this is how they'd write the code if they never expected to allow another provider, and other comments in the thread have already provided some insight into what the context might be (e.g. EU market).
Apple had enough experience relying on and being screwed by third parties that they have a built-in allergy to it. They do not like being boxed in and always want to be in control of their destiny.
This would’ve been useful during their bake off where they were trying to choose if they wanted to go with Google, Anthropic, or OpenAI. But even without that, they would absolutely not want to be in a situation where they had to redo everything to get off a bad partner.
You can download an app that does that. You don’t have to use Siri, and it can be turned off entirely, or decoupled from AI entirely, and the wake word can be disabled entirely. A lot of options are available.
Make a Shortcut that will open your local voice assistant app of choice when you press the action button.
Never underestimate the power of a bored programmer
Now that's an agent I'd use. Can we start with Siri for the terminal?$ Siri 'List the files in this folder'
“AI as a chat bot” has always been cool and fun in sci-fi movies, but it’s the least imaginative real world use for it. We need to move past it.
> $ Siri 'List the files in this folder'
Why in the world would I ever use this instead of just ‘ls’?
Google assistant on the nvidia shield left Siri in the dust years ago. Siri can't even chain basic commands in your smart home, it can't retain any kind of context (e.g. dim the lights...a bit more).
So, progress?
Any agent with the Home Assistant MCP hooked up can do it fine.
I am waiting for this function to sort my folders
https://apfel.franzai.com/
The code for all of those things is there, but not enabled if for most iOS users. There's a non-zero chance this code is there for Apple to try and convince the EU to let them release a Siri AI, and then just like sideloading have some impossible-to-actually-use alternative path to make the DMA happy.
People have speculated that with FPGAs or even an ASIC you could take some smaller models we have today, turn the chip into essentially that model-as-hardware, and get some really compelling performance/power-usage characteristics.
At that point running a GOOD quantized Q8 model on a phone is not beyond the realm.
That arrangement will change, somewhat, later today.
With mobile devices especially where ideally models could be shared across several apps.
But this specific code I'd assume is targeted to comply with EU regulations more than anything.
OT:
Since I've moved to Spain my principal Apple account is still in my home country but I've made another one for local app-store (since some local apps aren't available if you're not in the Spanish store).
iPhone is acting really strange, I'm logged to my principal account for iCloud+ and anything and local App Store.
If I log back to my principal account, I loose HCE mode and my building NFC app stops working until I re-log in the Spanish Apple Store.
If I'm logged at the Spanish account, I loose Siri AI which is blocked in the EU...
Maybe apple can even have something similar to MCPs for tool invocations of apps and publish it so all models can actually be trained to use that framework
I built an app that lets me run terminals on my MacBook. It’s a Next app so I can go to that website on my phone using Tailscale. It has a built-in MCP server, so I can connect it to ChatGPT and have it create terminal sessions, orchestrate terminal sessions, all that good stuff.
Down the line, I’d like to be able to connect it to Siri and just talk to my phone and it can control terminal sessions and projects that I’m working on on my laptop.
So many other actors in the industry would have just doubled down and follow the race to the bottom
This looks more like a participation medal, if at all.
It turns out that voice assistants aren’t all that important. Customers aren’t willing to pay for them and they’re more of a nice to have convenience feature.
Adding quotes around “genius” as it seems you misunderstood what I meant
What makes you think that? They insisted on Siri for so long and even made a big deal of Apple Intelligence in WWDC.
Then Apple threw the towel and waved the white flag with regards to Siri and replaced their efforts with third party AI.
Almost everyone developing their own AI is doomed to fail. I would not be surprised that when the dust settles, Microsoft CoPilot and Google Gemini will be left standing while OpenAI and Anthropic fade away. They will get what they need and then throw them away.
"Embrace, extend, extinguish" is the same playbook for decades, and you can only do that when you make money. Yes, making money still matters and post-capitalism is a big fat lie.
Handle the query on device. If the device cannot handle the query, push the query to Private Cloud, If Private Cloud cannot handle the query, push the query to a third party of the users choosing.
Which is pretty much the same as what they announced a few years ago.
Apple underestimated the amount of time and work needed to get to work and really the only bit that I think changed is that Private Cloud (and possibly the on device model) is a white label Google model rather than being developed in-house by Apple.
https://www.macrumors.com/guide/apple-command-center/
Mark Gurman seems to get extraordinarily accurate and detailed leaks. I wonder how Apple got this leaky. Pretty much everything about the Duo was known months in advance.
I wonder what the margin on this stuff is for them:
https://support.apple.com/en-us/127901Self promotion blog article:
https://braw.dev/blog/2026-04-06-mobile-os-as-agentic-broker...
Besides, the modern concept of "agent" is what, some year old? Bit fast to ossify stuff in the OSes already.
Ostensibly they'll activate this in the EU when it's ready. Given the current functionality, it looks ready (or close to being ready). My guess is it is ready and Apple is doing their usual negotiation tactics with the EU to see who blinks first.
The EU said “build it then we’ll tell you”, which is crazy.
The problem is not having Siri use different models beneath. That’s not what the EU wants. They want Siri to be replaced by Gemini/etc completely. Which fits with the DMA.
Apple’s proposal was for something akin to an agent that could gather the necessary data for the 3rd party AI to use. Because what they refused to do is let any other AI have unfettered access to everything whenever it wants.
So I suspect this is really for Apple’s benefit. I don’t think the EU would be happy if you could still use Siri, but it was grok underneath. Because that would still be Siri.
This is not how politics works here.
EDIT: by broader I mean all apps the user installs on their phone.
I don’t think that would meet the requirements of the DMA. Would it?
This is not that. This is expanding the Extensions that you already have. You can look at it in your Settings already.
All of which is why EU users are being left behind on the new Apple Intelligence (for now).
All the big companies know how suicidal that would be. They already get accused of it by people who don’t understand how targeted web advertising works.
Some of the new AI start-ups don’t seem to understand that, but Apple absolutely will not make a mistake that stupid.
High rep accounts like yours always come to defend, and you will probably dead this comment.
You claim that the AOP (always on processor) - which must always run in order to listen for “Hey Siri” - doesn’t actually listen for “Hey Siri”.
Or you claim that it does but they would never ever use the PhraseSpotter for that purpose because they’re too benevolent.
They could instead make a good assumption that you'll search the web, open the Yelp or OpenTable app. Take your phone to your favorite restaurant and/or pay with a credit card at the grocery store.
If you walk to a taco truck without your phone and pay with cash, you probably don’t do that every time and you generate enough data through other means to be accurately advertised to.
You can’t even search this because Google thinks you mean “app” even when you use quotes around “aop”.
It’s the thing that runs PhraseSpotter 24/7
Apple is in the game of providing an application platform. Maybe in the future they will provide their own LLM.
But it will never prompt you anymore, because Siri AI can handle that stuff now when it couldn’t before.
Whether you think it’s better or not is up to you. I do. I also found the version of ChatGPT it used to be worse than the one in the app.
This would’ve been useful during their bake off where they were trying to choose if they wanted to go with Google, Anthropic, or OpenAI. But even without that, they would absolutely not want to be in a situation where they had to redo everything to get off a bad partner.
I do not want my personal data or the queries against it to leave my network boundaries if at all possible. This should be a default option.
Make a Shortcut that will open your local voice assistant app of choice when you press the action button.