PodSearch

Connected

604: The Floor of the Street

 

00:00:00   Hello and welcome to episode 604 of Connected, the world's greatest podcast.

00:00:15   My name is Stephen Hackett. I'm your annual chairman, and I'm here to introduce Mike. Hi, Mike.

00:00:20   Oh, hello. I'm so happy to be back.

00:00:22   Yes, welcome back. How was your time away?

00:00:25   It was good. It was good. I have a very quick, funny story to tell.

00:00:30   I will just introduce the keynote chairman, Federico Vachichi.

00:00:33   Oh, hello. Ciao, Mike. How are you?

00:00:36   Very good, very good. So I was listening to the Connected podcast.

00:00:38   I had already been told that there was a segment I wasn't supposed to listen to.

00:00:42   Actually, before I was told that, I saw the chapter names, and it said,

00:00:47   Mike, don't listen to this chapter, which I thought was very funny as a chapter name purely for one person,

00:00:53   which is pretty good. Like, everybody else got that chapter name, even though it was useless.

00:00:58   For one person, you are a co-host.

00:01:02   Co-host.

00:01:02   Okay.

00:01:03   But, so I knew this was coming, right?

00:01:06   And I was taking a walk to go pick up coffee in the morning.

00:01:10   We were in Romania.

00:01:11   It was the first trip overseas with a baby.

00:01:13   It went really well, actually.

00:01:14   Very happy about that.

00:01:16   And I was walking down the street carrying two flat whites, one in each hand.

00:01:21   And I, listening to the podcast, they're like, okay, Mike, stop listening now.

00:01:26   I'm like, oh, God.

00:01:26   I didn't have any free hands.

00:01:29   I put a coffee down on the street, on the floor, like the pavement, on the street.

00:01:34   The floor of the street.

00:01:36   To pause the floor of the street, as we call it.

00:01:40   That's how it's called in Romania.

00:01:41   They call it the floor of the street.

00:01:43   It's like a whole thing.

00:01:45   So, yeah, I had to put the coffee down in the street, on the street floor, so I could pause it, and mark it as done.

00:01:52   This is fantastic.

00:01:54   Thank you for not cheating.

00:01:55   No problem.

00:01:57   Or listening, or whatever.

00:01:58   I can't even tell you what it's about.

00:01:59   I mean, I know what it was, even though I haven't listened.

00:02:02   There is precedent for this.

00:02:04   Like, I'm aware of what probably occurred.

00:02:06   Didn't y'all make...

00:02:08   Yeah, it's not time.

00:02:10   It's not time yet?

00:02:11   Okay.

00:02:11   It's not time.

00:02:11   I think it's July.

00:02:12   So, you know when it is.

00:02:14   Okay.

00:02:15   Yeah, I last week looked because I figured you were going to do this to me.

00:02:21   Are we that predictable, Federico?

00:02:23   Are we that predictable to him?

00:02:25   You told me, don't listen to this chapter.

00:02:28   It has happened in two other instances of which I have been a part of the creation of those episodes.

00:02:33   So, like, it was pretty obvious what was happening.

00:02:36   That's fair.

00:02:36   And yeah, and I think it's sometime in July that we need to resolve the Hack Olympics.

00:02:41   Yeah.

00:02:44   So, we've got a bunch of stuff to talk about today.

00:02:47   It's Google I.O. season.

00:02:49   So, just everybody buckle in for that.

00:02:52   But first, we have to talk about some things we've already talked about in follow-up.

00:02:57   Last week, Federico and I spoke about what is happening.

00:03:02   You were just, like, typing nonsense into the outline.

00:03:05   We were talking about the pros of editable user interfaces, right?

00:03:15   Yeah.

00:03:15   Something that I think all three of us are in favor of.

00:03:18   And Edward Munn wrote in on Mastodon, and I cannot believe, like, why is no one talking about this?

00:03:24   I can't believe we didn't talk about this.

00:03:26   Edward writes, the iPod app on the very first iPhone had an edit button to customize the tab bar.

00:03:32   What a shame.

00:03:33   This wasn't preserved.

00:03:35   So, I remember this, and if you go back and watch the introduction of the iPhone or, like, the iPod Touch, it's in here.

00:03:41   I think the default was playlists, artists, songs, and albums.

00:03:44   But if you say that you didn't care about songs and you wanted podcasts in there or audiobooks, you could change what was in the tab bar.

00:03:54   And then the rest of it was just in the overflow menu.

00:03:56   And, you know, not very many iOS apps let you adjust the tab bar.

00:04:02   We mentioned a couple, like Ivory.

00:04:04   But Apple had it as a first-party thing back in the day.

00:04:08   And it was good.

00:04:10   Yeah, I mean, what it was is no matter what happened, you just had to get podcasts in there, right?

00:04:15   It's like you've got to get rid of something to get podcasts in there.

00:04:18   And that's what I would do.

00:04:19   And I got it in.

00:04:20   Yeah.

00:04:20   This is another remnant of the, like, infection of U2 at Apple, right?

00:04:27   So, we have the iPod, which was ridiculous, that they just did one special edition iPod, and it was the U2 iPod.

00:04:35   We had Songs of Innocence.

00:04:37   And then also the artist's icon in the music app was the outline of Bono.

00:04:46   Yeah.

00:04:46   It's deep.

00:04:48   It's weird.

00:04:50   It is weird.

00:04:50   It's weird.

00:04:52   Why?

00:04:52   This band, you know?

00:04:53   Yeah.

00:04:53   Why?

00:04:54   This is one band?

00:04:55   There's everywhere.

00:04:55   They're just like, no one else got this treatment.

00:04:57   It's true.

00:04:58   It's true.

00:04:59   A little bit about spotting the International Space Station.

00:05:04   Jonas wrote in to point us to a web app that they built that allows you to check when the space station was historically closest to a location, not just visible from it.

00:05:18   So, really, like, when it is directly overhead.

00:05:21   The link to this will be in the show notes.

00:05:22   It is very cleverly called Astronauts for Dinner, which I think is a very good name for this.

00:05:29   And you can plug in an address or coordinates and see the last time and all the previous times the International Space Station was overhead.

00:05:40   So, this was in another anecdote of me listening to the show last week as a listener.

00:05:46   The moment when you said about, you know, I used to, I like to know when the International Space Station is overhead.

00:05:53   I also had the same feeling as Federico of, why?

00:05:57   Like, what are you going to do with that?

00:05:59   Like, what do you do with that information?

00:06:02   Yeah, you saw it.

00:06:04   You did.

00:06:05   You know?

00:06:05   Yeah, so I just plugged in one infinite loop, and the closest it ever got was February 8th, 1999.

00:06:15   It just zoomed right over as they were developing OS X.

00:06:19   Wait, can I plug in my, let me plug in my location.

00:06:22   Why don't you, you want to read your location out loud while you type it?

00:06:25   No.

00:06:26   Okay.

00:06:26   But, I'm typing it.

00:06:30   I'm putting in my current location at the studio.

00:06:35   It's zooming in.

00:06:36   Loading nearby passes?

00:06:39   Yep.

00:06:39   Oh, man.

00:06:40   The last one was August 14, 2006.

00:06:43   So, I guess this is the reason of, like, why it's good to know.

00:06:48   This all, oh, no.

00:06:49   I just missed it.

00:06:51   17th of November.

00:06:52   No, they're not sorted properly.

00:06:54   Actually.

00:06:55   They're sorted, I think, by the distance away.

00:07:00   The specific closeness of how.

00:07:01   Well, what does it mean four meters, then?

00:07:04   It was not four meters away from my apartment.

00:07:06   I think, like, longitude and latitude-wise.

00:07:10   Yeah, it doesn't mean like it's like.

00:07:12   Look, I didn't write this website.

00:07:14   Dustin Posh, your window.

00:07:15   Talk to Jonas.

00:07:16   If I open the window, there's the International Space Station here.

00:07:20   Hello.

00:07:20   Hey, guys.

00:07:21   You can give a national a high five.

00:07:23   All right.

00:07:24   All right.

00:07:25   It was 50 meters away in January 2024.

00:07:28   Okay.

00:07:29   That's pretty, pretty close.

00:07:30   Pretty close.

00:07:31   I missed one, like, basically exactly overhead in November.

00:07:34   It was like three meters off.

00:07:36   And if you had Home Assistant running, you could have gotten a notification.

00:07:38   Yeah, but I probably was going to be at home.

00:07:42   So, that's not so helpful.

00:07:44   Oh, yeah, that's right.

00:07:45   You put your studio address in here.

00:07:47   Very clever.

00:07:50   I like this website.

00:07:51   I think it's fun.

00:07:52   Thank you, Jonas.

00:07:54   Thank you, Jonas.

00:07:56   Okay.

00:07:57   Talking about, well, what do you do when it's going to fly overhead?

00:07:59   Jay wrote in and shared a photo that they took.

00:08:04   Actually, six photos they took of the International Space Station transiting the sky.

00:08:11   And so, they were able to stack them.

00:08:13   I think this is in front of the sun.

00:08:15   I guess it is.

00:08:17   It looks like the sun.

00:08:17   It doesn't look like the moon.

00:08:18   And you can see it just.

00:08:21   That's the orange moon.

00:08:23   That's the orange moon.

00:08:24   I think the sun is the orange one.

00:08:28   But don't quote me on that.

00:08:30   Okay.

00:08:31   Oh, I've just accidentally added this photo to my photo library.

00:08:36   Nice.

00:08:37   I don't think that's what I want to do, is it?

00:08:39   I think it's fine.

00:08:40   It's an iCloud link.

00:08:42   I think we need to put that somewhere else if we're going to put it in the show notes.

00:08:45   Okay, yeah.

00:08:45   I'll put it on someplace else.

00:08:48   Are we just like, are we kind of like Songs of Innocence, this photo into everybody's photo library?

00:08:53   Well, it probably, the iCloud link probably doesn't last forever, right?

00:08:56   I don't, I don't know.

00:08:58   Who could tell?

00:08:58   I mean, the funny thing is now I can't find this image.

00:09:01   So, that's just in there.

00:09:02   It's just buried away in there somewhere.

00:09:04   Yeah, you got to go to Recents to find it.

00:09:06   Otherwise, in the library, it's like, like it has a date metadata.

00:09:10   2024.

00:09:10   It's so hard to find Recents on the Mac.

00:09:14   Yeah.

00:09:15   Like, it's really easy on the iPhone, but I find it difficult to find, like, Recents on the Mac.

00:09:23   Yeah, well, I'll have a link to some ways you can see that.

00:09:27   Federico, you want to talk about this double-tap bottom-of-the-screen Siri business?

00:09:35   Yeah, Mike Perotin asked, interesting takes on the new incoming Siri.

00:09:39   Do you use the double-tap bottom-of-screen to type to Siri?

00:09:43   I wonder how much of the app will be similar to that.

00:09:46   Almost sounds from your takes, it will be moving to the top of the screen instead of the bottom.

00:09:51   I tried to make myself use it and started a little bit of a habit.

00:09:56   Now, your takes just reminded me of that feature.

00:09:59   So the type to Siri graduated from an accessibility feature to a proper system feature a couple of years ago, like a year and a half ago.

00:10:11   I barely use it.

00:10:13   Like, very few instances on my end of wanting to do something with Siri but not being able to speak to Siri.

00:10:22   And so in those cases, I actually did the double-tap on the home indicator to type to Siri instead.

00:10:30   I think, so there's a couple of problems here.

00:10:33   Obviously, the fact that Siri isn't great doesn't exactly incentivize the use of type to Siri.

00:10:40   And second, the activation method itself.

00:10:44   I think from the description that Gurman gave, I think it's going to be much easier to just pull down from the dynamic island, you know, kind of do it very quickly, than it is to precisely tap on the home indicator.

00:10:58   So the gesture sounds easier, in theory.

00:11:01   Man, I totally forgot that that double-tap thing even existed.

00:11:05   But if Siri will be smarter, then you will also be more incentivized to actually use it in a variety of cases.

00:11:12   I think, for me, there is an issue in having two different ways to invoke Siri, no matter what they are.

00:11:20   Like, where I feel like I would be happier to, you know, long-press the sleep-wake button and be able to then start typing or talking.

00:11:29   You know, like, maybe I tap another on-screen button to bring up the keyboard or something.

00:11:33   But there is, like, a cognitive load of, like, I want to talk to Siri.

00:11:39   Okay, do I want to type or speak?

00:11:41   And then choose the activation method that is for that method, you know, like, for that way of interacting.

00:11:49   I just feel like you just have one of them.

00:11:52   Just let me long-press it.

00:11:53   If the screen's open and I'm looking at it, maybe just bring up a keyboard as well.

00:11:57   And I can speak into it or I can type into it.

00:11:59   Like, why do we need to have two different ways to talk to Siri?

00:12:02   Like, to me, that doesn't make any sense, really.

00:12:05   Yeah, I get it.

00:12:07   I get it.

00:12:08   Didn't German say that the new Siri design in the Dynamic Island will feature both input modes, like, at the same time?

00:12:16   So, in theory, you should be able to switch.

00:12:17   Because I can imagine, like, maybe you start with a chat query and then you realize, no, actually, I want to speak.

00:12:23   And you press the microphone button.

00:12:25   Maybe they could do that.

00:12:27   I don't know.

00:12:28   Yeah.

00:12:28   Do you think if they did that, they would get rid of the...

00:12:32   the long press on the button to...

00:12:35   No, I don't think they will.

00:12:36   I think they will get rid of the double tap on the home thing.

00:12:39   Yeah.

00:12:40   Not the side button, though.

00:12:42   We don't need that bar anymore, right?

00:12:44   No, I don't think we do.

00:12:46   We can get rid of that now.

00:12:48   Isn't it...

00:12:49   It's also kind of weird now, I think, because I think as of iOS 26, the bar gets hidden by default in most apps.

00:12:56   So, like, if you want to use Type 2 Siri now, you have to double tap an invisible space.

00:13:00   Maybe that is part of them, like, starting to educate people that it doesn't exist anymore.

00:13:06   Because it's kind of wasted space on the phone.

00:13:09   Like, I don't think that a new user to an iPhone is helped by that being there anyway.

00:13:17   Like, I think it helped a lot when we were doing the transition, right?

00:13:21   From buttons to just screens.

00:13:24   But I think at this point, if you're learning the iPhone, I don't think it would be that hard to explain to someone that you just swipe.

00:13:32   It's like, how can I swipe?

00:13:33   There's no bar.

00:13:34   It's like, I don't really think that that's a thing that people are, like, really fixed on.

00:13:38   So, it'd be nice to get rid of that.

00:13:41   And I just wanted to note that the Instagram app has been fixed on the iPad.

00:13:47   They have actually done the thing they should have done in the first place.

00:13:50   They have just made an iPad version of the Instagram app as it should work.

00:13:54   Instead of that wild idea that they had where it was just reels.

00:14:00   Like, it just, how about just reels?

00:14:03   No, they've changed it now.

00:14:04   Now, just, the home tab is the first tab.

00:14:07   So, Instagram on the iPad is now just as it should be.

00:14:11   I saw a conspiracy theory about this.

00:14:14   I don't remember who it was that I saw it.

00:14:16   I'm sorry that I'm not quoting them.

00:14:18   It may have been Parker Ortolani.

00:14:20   It was like, what if this is because this will be all the way it will work on the folding iPhone?

00:14:27   And I like that.

00:14:29   I don't think it's accurate, but I like that conspiracy theory.

00:14:32   I also saw that post.

00:14:33   So, I like a good theory like that of, like, why on earth would you do it now?

00:14:37   But I actually just think it's because it was a bad experience and they fixed a bad experience.

00:14:42   So, this week was Global Accessibility Awareness Day,

00:14:48   which it has now become a tradition that Apple will show off some features coming to the next operating system before WWDC.

00:15:00   What could it be called?

00:15:01   Exactly.

00:15:02   That are the accessibility-focused features because they then get their own time in the sun.

00:15:09   And so, people that care about these features get, they see a lot of coverage about it.

00:15:14   But then also, we get, you know, like the media and us on podcasts, we spend time talking about these features where we otherwise may just gloss over them more quickly if it was part of the WWDC keynote because everything kind of gets glossed over quite quickly after the keynote.

00:15:31   So, there's a bunch of stuff going on here.

00:15:37   And I think that the overall thing is Apple intelligence coming to a lot of areas of accessibility.

00:15:43   I think that's one of the key things that is visible across a bunch of this stuff.

00:15:46   And I think that that's really great of like using models that can understand image and text and do things with those and people to be able to converse with these features a little bit more of allowing people who otherwise have issues with sight or hearing or mobility to be able to navigate the world using their devices.

00:16:09   So, I think it's really cool.

00:16:09   So, I want to run through some of the features that I found to be the most interesting.

00:16:12   And there'll be links in the show notes to some coverage and Apple's press release as well if you want to read about everything.

00:16:18   But VoiceOver is getting some big enhancements.

00:16:22   So, VoiceOver is essentially the system that you can kind of listen to and kind of converse with the phone in some ways.

00:16:32   I'll just use the phone for an example.

00:16:34   But it has always required developers to do a little bit, to do quite a bit of work.

00:16:40   And it can be quite spotty.

00:16:41   And so, both VoiceOver and Voice Control are being kind of enhanced to allow for the system to be more aware of what's going on and allowing you to kind of converse with it.

00:16:54   So, they're using, for example, Apple Intelligence to better summarize images on the screen.

00:16:58   And you can ask questions of what you're seeing as well, which is fantastic, right?

00:17:03   So, you're just like, oh, this is that.

00:17:05   And you can say, well, what does this mean?

00:17:06   And similarly, you can now use the camera with VoiceOver so you can hold your phone up and your phone can describe to you what's in front of you.

00:17:15   And you can also ask questions to your phone backwards and forwards about what it is seeing.

00:17:20   And it can use some context to understand that.

00:17:22   And like this is, you know, in the conversations about potential AirPods or glasses with cameras, these are the kinds of features that, in theory, could be enabled by devices that are permanently attached to someone invisible.

00:17:37   So, it's pretty cool.

00:17:38   Magnifier, as well, is also using Apple Intelligence to better describe what is on the screen.

00:17:42   So, this is the app that can essentially use the camera or I believe it works on the screen as well to basically make everything bigger.

00:17:49   You can ask audibly for Magnifier to zoom in on things or turn on the flashlight.

00:17:55   So, you can just have conversations with the devices that's doing this stuff.

00:17:59   And Apple kind of spent time talking about both Magnifier and VoiceOver can be used at the action button.

00:18:05   And there seems to be kind of like some additional features that the action button will be able to play in bringing up and activating these features.

00:18:12   I think the one that I looked at, you know, there's always something in the accessibility features that make people go, hang on a minute.

00:18:21   And it can be either design or functionality.

00:18:25   And I think voice control this year is the one that is maybe raising the most eyebrows.

00:18:32   So, I'm going to read from Apple's press release.

00:18:35   The option to say what you see is a feature they have is great for navigating any app, including those with visual layouts, such as Apple Maps or files.

00:18:45   Using intuitive language like tap the guide about best restaurants or tap the purple folder, this can help users overcome barriers when elements aren't properly labelled for accessibility.

00:18:55   So, essentially, a user who maybe has some mobility issues that allow them to control their phone correctly can just speak to the phone and ask it to select stuff.

00:19:07   Now, that is very interesting because that is agentic operation of iOS, essentially.

00:19:16   Isn't this the paper they published like a year ago that we talked about, Ferret 2?

00:19:23   Yes.

00:19:24   Wow.

00:19:25   I would not have remembered that as a good poll.

00:19:27   But, yeah, I think this is very interesting, right?

00:19:30   And, like, they show a video in the press release of somebody, like, asking, like, select the orange folder.

00:19:39   There's, like, a green, orange, and yellow folder.

00:19:41   And then you see the phone select the orange folder.

00:19:44   And so, like, this kind of stuff is really interesting because it is showing the iPhone and the user being able to have a dialogue and do things on the device.

00:19:57   Which is kind of, at the moment, I mean, we're going to talk about Google I.O. in a little bit, seems to be one of the big new areas of attack, right?

00:20:06   Of, like, controlling devices and how that's done.

00:20:10   So, this is very interesting.

00:20:11   And then a couple other things that were, I thought, we'll call accessibility reader can now handle more complex page layouts that may include tables, columns, and images.

00:20:21   So, it can, this is essentially giving people more ways and tools to be able to change fonts, change sizes, you know, like, maybe, like, for some, like, dyslexic fonts as well, like, fonts for dyslexia.

00:20:33   They have those in there, so it can kind of adjust text.

00:20:36   It can also do translation.

00:20:37   You can generate subtitles for personal videos.

00:20:42   So, it's pretty cool.

00:20:43   And then one that was just, like, wild, and they have a really good video about this in the first release, is Apple Vision Pro wheelchair control.

00:20:53   Where, essentially, for wheelchairs that can connect, and Apple's working with a couple of partners about building a connection between the two things, you can use the eye tracking functionality of the Vision Pro to control and move a connected wheelchair.

00:21:11   So, it shows the UI that the user is seeing, where they're essentially got, like, like a D-pad, I'll say, you know, like video game controls.

00:21:20   And using the eye tracking, you can, like, look at the go forward, and it will go forward, look at left, return left.

00:21:26   And the video is astounding, like, to see this in action.

00:21:30   And it, you know, really made me think, like, the Vision Pro is not the device for this, right?

00:21:34   Like, because it's huge.

00:21:36   And you don't want to be wearing it all day, every day.

00:21:39   But you can imagine a world in the next few years where this functionality is built into something more wearable, like, all day.

00:21:47   And it's going to absolutely change the lives of people.

00:21:50   Because they'll be able to have autonomy in their movement where they otherwise may not have had.

00:21:59   Like, so yeah, every year this is interesting stuff.

00:22:03   But I think the combination of what they're doing with Apple Intelligence and then this Vision Pro thing, I think is, like, I think it's super cool.

00:22:12   Yeah, I mean, Apple does such a thorough job with all this technology.

00:22:16   And one thing that, you know, we've known from hearing interviews from people who work on this at Apple and our friends who cover it more closely, like, Apple doesn't develop these features in a vacuum.

00:22:27   Like, they're working with people whose lives will be changed because of this technology.

00:22:34   And it's very grounded work in a way that I really appreciate.

00:22:38   And yeah, the Vision Pro thing definitely jumped out at me.

00:22:41   It's like, that seems a little strange.

00:22:43   But yes, so much of this is, like, building towards the future.

00:22:47   Right?

00:22:48   And I think it was on, I think it was on Upgrade.

00:22:52   You were talking about the AirPods and mentioned the door detection technology that Apple announced maybe two cycles ago with this stuff.

00:23:00   Mm-hmm.

00:23:01   Like, all of this stuff builds together.

00:23:04   These aren't necessarily discrete features or technologies, but, like, they're building these things together in a way that they can be used together in meaningful ways.

00:23:14   And, like, I always find these posts so exciting and encouraging because Apple really does care about this stuff.

00:23:20   And I think it shows in the way they go about developing them.

00:23:24   This episode of Connected is brought to you by Squarespace, the all-in-one website platform designed to help you stand out and succeed online.

00:23:34   Whether you're starting something new or you're scaling something up, Squarespace gives you everything you need to claim a domain name, showcase your offerings with a professional website, grow your brand, and get paid all in one place.

00:23:50   One of the things I love about Squarespace is that it can grow with you over time.

00:23:55   So that you've got a site that you built on Squarespace maybe a couple years ago, and now you've got new branding for your company or your project.

00:24:03   Well, you don't have to go around page by page and change each color and each font and make sure it all lines up because Squarespace design tools are really easy to use.

00:24:13   You can make site-wide changes really easily and really quickly.

00:24:17   It's easy to understand what's going to happen with their design tools.

00:24:20   But maybe you're not a designer.

00:24:24   Well, Squarespace's AI-enhanced website builder lets you quickly and easily build a site bespoke to your business.

00:24:31   Just input some basic info, and it's going to combine that information with Squarespace's professionally designed templates.

00:24:39   So however you start, you have beautiful design options.

00:24:42   No experience required.

00:24:44   And you can manage your videos easily.

00:24:47   Maybe you've got how-to videos.

00:24:50   Maybe you have a video course and you want it to be available for sale, not to the general public.

00:24:55   Squarespace can handle all that as well.

00:24:56   Video libraries are really easy to build out.

00:25:00   Paywalled content is easy to manage.

00:25:02   It's perfect for online courses, tutorials, or premium workshops.

00:25:06   So head on over to squarespace.com slash connected for a free trial.

00:25:13   And when you're ready to launch, use the offer code CONNECTED for 10% off your first purchase of a website or domain name.

00:25:20   That's squarespace.com slash connected and the code CONNECTED for 10% off your first purchase.

00:25:28   Our thanks to Squarespace for the support of the show and all of Relay.

00:25:32   It was Google I.O.

00:25:35   Hey-o!

00:25:37   I'm going to be honest with you guys.

00:25:39   I did not watch the entire Google I.O. keynote.

00:25:42   Me neither.

00:25:43   No, you'd be dead.

00:25:44   Because I find the whole Google I.O. keynote to be rough.

00:25:47   What I love and wait for is The Verge.

00:25:50   The Verge posted a 35-minute video, which was all of the big announcements.

00:25:56   I watched that, and I think I got everything that I needed from that and then reading some articles.

00:26:02   But the whole Google I.O. keynote, I mean, it suffers from the problem, and I actually think is evident in what we're going to talk about, of too many teams working on similar things who all get their time on stage.

00:26:17   Which is just, like, a fundamental issue with Google and how it operates as an organization.

00:26:22   I think Ben Thompson referenced it as, like, the keynote is the structure.

00:26:28   I forget.

00:26:29   He has a very funny turn of phrase, basically.

00:26:31   Yeah.

00:26:32   You are showing the internal structure of the company through who's on stage.

00:26:37   Yep.

00:26:38   I.O. has always been that way, or at least in the last, I don't know, five or ten years.

00:26:43   And increasingly feels like we are not showing products.

00:26:51   We are showing, like, projects within the company.

00:26:54   Yeah.

00:26:54   And I think Google has struggled in the past.

00:26:57   I think definitely this year, too, of, like, we have this technology.

00:27:01   We have this idea of a thing.

00:27:03   But we don't know how to, like, get it into the world, so we're just going to talk about it.

00:27:06   And, like, maybe it'll show up, or maybe it won't.

00:27:08   I mean, sort of famously, in the past, a lot of I.O. announcements are just vaporware.

00:27:15   They just never ship.

00:27:16   Or they ship to a very small number of users as an experimental beta, and then it goes away.

00:27:24   Because it doesn't click, it doesn't land.

00:27:26   Like, one of the classics is, like, Google Photos removing a chain link fence.

00:27:30   Right.

00:27:30   Where is it?

00:27:31   Which is a demo they showed multiple years, and I'm not sure it ever shipped.

00:27:35   Yeah.

00:27:35   They just keep showing, like, oh, look how great this is.

00:27:38   Okay.

00:27:38   Or, like, the call holding thing that they did, you know, like, all that kind of stuff.

00:27:44   Now, I will say, like, there are some stuff, there's some stuff in Google I.O. that I found really interesting.

00:27:50   And, like, obviously a lot of it is Gemini stuff, and we'll talk about those.

00:27:55   But my overall feeling watching this 35-minute video was a feeling just of great despair at some of the stuff that Google is doing.

00:28:03   And the majority of my despair is in their video and image creation tools.

00:28:10   And the applauses that these features are getting from the crowd.

00:28:15   And I don't know who it is that is applauding this.

00:28:20   Because I guess it's people that work at Google.

00:28:22   Because I just, I really struggle to be excited about image and video generation.

00:28:32   I do.

00:28:33   And they have, like, multiple systems, right, that are doing this.

00:28:40   So we have Gemini Omni, which is a new model that you can use in Gemini.

00:28:47   Or PIX in Google Workspace.

00:28:51   Or Google Flow.

00:28:53   Three distinct tools that do the same thing.

00:28:58   Welcome to Google.

00:28:59   And it's like, what, and PIX is new.

00:29:02   It's like, why are we doing a new one?

00:29:03   Why do we need a new one?

00:29:05   But, like, the thing for me, like, they showed some demos of what Gemini Omni could do.

00:29:10   And technically seems impressive.

00:29:14   Some of the demos looked better than others.

00:29:17   They were all presented as looking amazing.

00:29:19   But they didn't all look even, you know, some of them were very convincing.

00:29:24   Some of them really weren't.

00:29:26   But the question I have to ask is, is why?

00:29:29   Like, why do we need video and image creation?

00:29:35   Like, I don't, I feel like I don't know why it should be so easy to create fake videos.

00:29:45   And even now, like, they're really like, ah, look at this video I made.

00:29:48   And I'll just change it.

00:29:49   And look how great this is.

00:29:50   It's like, but I don't know.

00:29:52   It's not great.

00:29:53   It's extremely problematic.

00:29:55   Yeah, like, I don't know what this does.

00:29:59   Like, a lot of the other stuff that they're showing, you can see how, and I understand not everybody likes it, and I get it, but you can see how people's lives could be made more productive, potentially, by some of these tools.

00:30:15   I don't know what they're doing.

00:30:19   And any worthwhile application outweighing the vast problems that will be created by this technology existing.

00:30:36   And, like, they follow up, you know, they show, like, oh, look at Gemini Omni, it's amazing.

00:30:40   And, like, oh, and by the way, like, we're also now building tools that will help people understand if the content that they're seeing is generated by AI.

00:30:50   It's like, is that supposed to, like, give you an out, right?

00:30:55   That, like, oh, people could just ask Google if this video or photo is real or not, and either we'll, you know, we'll have the credential check stuff that people have to opt into.

00:31:08   But I'm still not sure how that's actually supposed to work, but whatever.

00:31:11   But then they're like, oh, you know, we're going to use a bunch of functionality to be able to detect it.

00:31:15   But, like, that's all great, but it still requires people to think, to ask.

00:31:21   And the issue is the people that think to ask are not the people that you need to worry about, right?

00:31:29   Like, if somebody sees a video or an image and they're like, I'm not sure if that's real, awesome.

00:31:34   You're good.

00:31:36   But it's the rest of the people who are like, well, that's real.

00:31:39   And they're never going to ask Google if it was real or not.

00:31:42   Yeah.

00:31:43   Media literacy continues to circle the drain.

00:31:45   Yeah.

00:31:46   Because you literally cannot believe what you're seeing anymore, which is very difficult.

00:31:53   Because the scale that it is created at, people could always make fake things, but it was hard to do.

00:32:01   Yeah.

00:32:02   Like, I learned Photoshop in college.

00:32:04   I can make some fake stuff in college and Photoshop, but it was not at the scale that it is now.

00:32:09   And to go back to something we talked about months ago when Grok was just undressing people, it wasn't built into the social networks.

00:32:17   Because there weren't, you know, like, it's so prevalent and so easy to create and to see and to find.

00:32:25   It is a different thing than, oh my God, you know, Photoshop.

00:32:30   Like, they're different.

00:32:31   The levels are so different.

00:32:33   And if you can't see that, like, I don't know what to tell you.

00:32:37   I would expand your angsty question, though, to more than just image and video generation.

00:32:44   Although I wholly have it about those things.

00:32:47   At some point, we're going to talk about some of the, frankly, bonkers things that Google DeepMind CEO said.

00:32:53   But if you really, like, all of these companies, well, not all of them, many of these companies will say, like, this is very powerful technology.

00:33:05   And it comes, basically, it's the Spider-Man quote of, like, with great power comes great responsibility.

00:33:12   But I don't believe any of these companies will be able to exercise the responsibility that they say is needed.

00:33:22   Right?

00:33:23   Yeah.

00:33:23   And, like, if you really think, I'm just going to read it because it's so stupid.

00:33:27   The, um, if you really believe that we are standing in the foothills of the singularity.

00:33:36   Yes.

00:33:38   If you really believe that, Jim, you're going to have to bleep this.

00:33:42   Like, just f***ing stop.

00:33:44   Like, just stop.

00:33:46   But also, what does that mean, Demis Sasabes?

00:33:50   What are you at?

00:33:51   Like, did anybody check with him what he was going to say?

00:33:56   Sounds like a pretty dope name for an album.

00:33:59   It does.

00:33:59   It sounds like an album Muse would make.

00:34:01   No, no, it reminds me of the Oasis standing on the, Oasis did standing on the shoulders of giants.

00:34:09   And this is like, we have Oasis at home, and it's Demis Sasabes saying, we're standing on the foothills of singularity.

00:34:17   Because it's like, what does that mean?

00:34:19   Like, what does that actually mean?

00:34:21   What is the singularity in this scenario?

00:34:24   Because he says that after saying a bunch of really interesting stuff.

00:34:28   You know, this is the kind of quote that the previous Siri guys, women to Hawaii, would say.

00:34:34   Yes, because it's like, he says that after talking about like, oh, we're going to cure every disease.

00:34:40   And it's like, you know what?

00:34:42   If you can do that, like, awesome.

00:34:45   You know, like, I'm on board.

00:34:47   If you can cure every disease, like, I'm in this.

00:34:50   And to Google's, well, to Alphabet's credit, like, and this, Demis, how do you, I don't know how to pronounce his name.

00:35:00   Demis Sasabes, like, is involved with Isomophoric Labs, which is like, trying to use this technology to quote, I wrote it down.

00:35:09   Our mission is to reimagine the drug discovery process and the goal of one day solving all diseases.

00:35:17   That's fantastic.

00:35:19   And like, I'm on board.

00:35:20   Me too.

00:35:21   Right?

00:35:21   Demis Sasabes is not an idiot.

00:35:24   No, no.

00:35:25   He has a Nobel Prize in chemistry.

00:35:27   I'm just saying, like, he's smarter than the three of us put together.

00:35:32   Yes.

00:35:33   Oh, for sure.

00:35:33   But that doesn't mean you don't say unhinged things on stage.

00:35:36   Yes.

00:35:37   But my point is, I'm trying to rein this back in, if you really think, if you really think AGI is on the horizon, if you really think, if you're anthropic and you really think that Mythos is too powerful to put into the world because it's going to break all software humanity has ever written.

00:35:57   Yet you keep going.

00:36:00   They all keep going.

00:36:01   Right?

00:36:01   It's like someone, like, you go to somebody and you're like, hey, man, like, you have this really destructive habit in your life.

00:36:09   Like, it would be, we would love you to stop.

00:36:12   You know, whatever it is.

00:36:13   Smoking, drugs, using windows, whatever it is.

00:36:17   And so many times those conversations heartbreakingly end in somebody saying, I know it's bad for me, but I'm not going to stop.

00:36:25   And I just, the whole time I watched it, I watched the whole Verge thing and a bunch of other stuff.

00:36:30   And that was the feeling I had the whole time.

00:36:33   It's like, if you really think, if you really think we're this close or that it's on the horizon or we're on the foothills or whatever analogy you want to use.

00:36:42   If you really think, oh, we need to build a whole system so people understand that the video they're watching is real or fake, stop making the video creation tools.

00:36:52   Yes.

00:36:54   Now, what their answer is to this is, well, no one else is going to stop, but we are the ones in our eyes who are going to be responsible with it.

00:37:02   So we have, this was how, why OpenA was founded and why, you know, they were initially a nonprofit.

00:37:07   It was, we're going to build up this technology so we know how to control it and we know how to be put responsible safeguards around it.

00:37:14   So when other people inevitably get here and do similar things, then we have an understanding of how to manage it.

00:37:21   But that's not how this goes.

00:37:22   It's not how it's panning out at all.

00:37:24   No, no.

00:37:25   Yeah, they're like, they're like the meme of, we're all looking for the guy who did this and they're the guy.

00:37:31   Like, yeah, they created a solution to a problem that they created in the first place.

00:37:37   So I just looked up, by the way, the definition of the singularity.

00:37:42   Please.

00:37:42   And this is being provided to me by the AI overview, which is just like, what are we doing?

00:37:47   A hypothetical future tipping point where artificial intelligence and technological growth accelerate so rapidly that humanity is irreversibly transformed.

00:37:58   Now, to try and throw Hassabis a rope here.

00:38:03   Sure.

00:38:03   I think what, maybe what he was trying to say is, if we can get to the point where we cure all disease, that is essentially singularity moment.

00:38:17   Because humanity is irreversibly transformed for the positive at that moment, right?

00:38:23   Yeah.

00:38:23   That like, we don't get sick anymore, right?

00:38:28   Which is like, again, there are many other issues that I think would be caused in that scenario.

00:38:34   But it's good, right?

00:38:36   But in theory, it's a good thing.

00:38:38   Right.

00:38:39   We need to manage that as a human race.

00:38:41   But like, that is a good thing.

00:38:42   And if you just follow that thought through, it's like, yes, okay, if you believe that's true, we are standing in the foothills of the singularity.

00:38:52   However, I recommend you do not use that language because it sounds unhinged.

00:38:59   And also, the singularity sounds really scary as a phrase.

00:39:05   And it's like, are we in a simulation kind of stuff?

00:39:08   Yeah.

00:39:08   I would love to know if anyone knew he was going to say that at Google before he said it.

00:39:15   Like, was anyone checking his slides?

00:39:17   Like, man, what a thing to say.

00:39:19   So let me play devil's advocate for a minute.

00:39:21   Please.

00:39:22   So they pitched Gemini Omni as a world model, which I think is a really fascinating technology.

00:39:35   So a world model is not just a large language model.

00:39:38   That's obviously where it all starts.

00:39:39   It's not an image model.

00:39:41   It's not a video model per se, although it can generate videos.

00:39:45   And I'll get to this in a minute.

00:39:46   It can build a real-time simulation based on real-world physics.

00:39:52   That's like the short description of it.

00:39:53   So if you're thinking, we want to achieve the singularity and we want to be able to cure every disease,

00:40:00   I think it's obvious that you start from a place of, well, let's teach the AI how physics work.

00:40:08   So that in a not-so-distant future, when we want to put in a nanocamera inside of your body and the nanocamera needs to analyze blood flow and muscle structure and bone density,

00:40:19   it starts from a place of understanding how real-world physics work.

00:40:25   My problem with all of this is that I think if they're not lying about this—so big emphasis on this sentence—if they're not lying about this, that is a noble goal.

00:40:39   And it makes sense to start from a place of giving the AI knowledge about the world, about physics, about how walking on solid ground works, how gravity works, how the sky looks.

00:40:55   Like, it all makes sense that you start there.

00:40:57   My take is that—two takes.

00:41:02   First, not every bit of research needs to become a product.

00:41:07   Yes.

00:41:08   And two, all these AI companies have a serious PR problem.

00:41:13   Oh, yeah.

00:41:13   They are pitching products that journalists and employees in the audience clap for because they are technically impressive.

00:41:23   Normal people in the real world, they look at that stuff and they're like, look at this.

00:41:28   Jim, I'm sorry.

00:41:29   Can you also blip this?

00:41:31   Let's go.

00:41:31   Real people—people in the real world are looking at these announcements and they're thinking, look at this.

00:41:37   Like, look at—what are they even announcing?

00:41:39   Man, where am I going to get mine in today?

00:41:40   I'm excited.

00:41:41   And you have Dario Amodei saying, ah, pay attention.

00:41:45   The AI is going to take all of your jobs in the next 12 months.

00:41:49   First of all, somebody needs to make Dario stop saying those things.

00:41:54   Yes, because then it's also like, and we're going to give $15 billion to SpaceX.

00:41:58   It's like, okay, are you taking the jobs?

00:42:02   Like, who's taking the jobs?

00:42:03   So, not every bit of research needs to become a product.

00:42:08   Like, you can't build a world model without also saying, ah, and we're making Gemini Omni based on that research.

00:42:16   And now you can make four realistic videos that look just like real life.

00:42:20   No, just don't release it.

00:42:21   Just don't make a video product.

00:42:24   So, I think there is an inkling of nobility to all of this.

00:42:30   If you can't make the AI autonomously research and create cures for diseases, that's awesome.

00:42:39   Yep.

00:42:39   But to go from that to, ah, just type in a prompt, hey, make me a fake video of Osama Bin Laden actually being alive and chilling in Paris.

00:42:50   Like, no, no.

00:42:51   Because it's like even the demo they show, right?

00:42:55   Like, it's interesting.

00:42:55   Here is a marble moving.

00:42:57   And it understands gravity.

00:43:00   It understands friction.

00:43:02   It understands momentum.

00:43:03   So, we can make this video now.

00:43:05   Great.

00:43:06   Nobody needs it.

00:43:07   Right?

00:43:08   But the things that Gemini Omni can do in other areas, oh, that's really interesting.

00:43:16   But, like, why I agree with you is, like, you don't need to apply every possible outcome to this technology.

00:43:25   Yeah.

00:43:26   Like, I could somewhat understand why these companies were pushing so hard on photos and videos when they didn't have an interesting product.

00:43:35   Right?

00:43:36   Like, a couple of years ago or whatever, you know, this all came into our lives because of Dali, right?

00:43:43   Like, that was the first thing that.

00:43:46   And what was the other one?

00:43:47   Mid-journey?

00:43:48   Yeah.

00:43:49   Yeah.

00:43:49   But Dali was first, right?

00:43:51   Like, it broke onto the scene before ChatGPT did.

00:43:54   Like, it was a thing that, like, people could see and use.

00:43:57   And we were understanding it was being generated, right?

00:44:00   And, like, that technology was like, okay, like, here's the thing you can do.

00:44:06   But at this point, there are other things that this technology can do that is more interesting.

00:44:12   Why do we continue to advance the video stuff?

00:44:16   And, like, there was just, like, a phrase that one guy said is, like, this is the nano banana moment for video.

00:44:22   And it's like, shut up.

00:44:24   Like, just shut up.

00:44:26   You're talking ridiculous nonsense.

00:44:29   The nano banana moment is like, okay, but it's not made our lives better.

00:44:34   You know?

00:44:35   The fact that Google made a model that can make images look really good.

00:44:38   Also, how have we reached the point where that's an actual English sentence that's allowed?

00:44:46   And also, like, I am British, so I can say this.

00:44:49   There was something, like, this guy was British, too.

00:44:51   And there was something about his British accent saying the nano banana moment for video.

00:44:57   I was like, no, that's terrible.

00:44:58   No, no, no, no, no, no, no.

00:44:59   You can't say this.

00:45:00   Americans can say this.

00:45:02   You can't say this.

00:45:03   You have been living in America too much, my friend.

00:45:05   Yeah.

00:45:06   No, no, no, no, no, no.

00:45:07   Nano, nano, taco, banana.

00:45:11   Say that.

00:45:11   Say that.

00:45:12   No, no, taco, blah, blah, blah.

00:45:15   All right, let's push this aside.

00:45:19   Yeah, existential angst.

00:45:21   I'm going to take a huge broom.

00:45:22   Off the table.

00:45:23   Push this aside, because there were some other things I wanted to talk about.

00:45:26   One of the things they're showing in a bunch of their demos is Gemini and voice.

00:45:32   Lots of people were dictating to their phones, and Gemini was doing things.

00:45:37   Like, long rambling lists of tasks, corrections, and, like, they were purposely flubbing stuff,

00:45:44   and, you know, like, just...

00:45:45   It was really interesting where sometimes, like, a person in one thing was, like, they're

00:45:50   dictating an email, like, and then they start giving instructions to Gemini, go back to

00:45:56   dictating, and give more instructions.

00:45:58   And the model could understand what all of...

00:46:01   Got all of that, right?

00:46:02   It knew the parts that needed to be presented as text and an email.

00:46:06   It knew the parts that was go and grab these images from Google Drive, like, or whatever.

00:46:11   Like, and it did all of that.

00:46:12   And this kind of thing is the most exciting to me right now, of a future where we can have

00:46:19   a dialogue with our computers, and they understand what we want them to do.

00:46:25   And I find that to be quite exciting, that, like, in a few years' time, I could communicate

00:46:32   with my Mac, like, ask it to do things, and it would do them.

00:46:36   Like, I find that kind of technology to be really interesting.

00:46:40   And then this kind of goes into a product that they're...

00:46:46   Because they've got to keep...

00:46:47   Everything needs a name.

00:46:48   Gemini Spark.

00:46:49   This is their open floor.

00:46:51   Yeah.

00:46:52   And so, the way they set it up, it runs on dedicated virtual machines, it's running in

00:46:57   the background, you can use it in the Gemini app, currently integrates with their tools, it

00:47:02   will have MCP integrations later on, so it can integrate third parties.

00:47:05   The fact that now everyone's kind of doing this...

00:47:10   Yeah.

00:47:11   This is more similar to...

00:47:13   This is actually more their take on perplexity, personal computer, more than anything.

00:47:18   But that was a take on...

00:47:20   They're all trying to do an open-claw thing, but in a way that's not as risky, right?

00:47:26   Yeah.

00:47:26   Yeah.

00:47:27   The idea that you have this agent, you connect the agent to a bunch of tools, and the agent

00:47:31   runs in a sandbox, and it can do things for you, and sort of the industry is kind of settling

00:47:38   on this terminology where you assign tasks to an agent, whereas you have chats with the

00:47:44   chatbot.

00:47:45   So, in Spark and in perplexity, you create a task, whereas if you go to Gemini or ChatGPT

00:47:51   regular, you start chats.

00:47:54   I'm very skeptical of Spark, mostly because the Gemini models historically have sucked at tool

00:48:04   calling and integration with external tools.

00:48:06   Right.

00:48:07   Yeah, we spoke about that in terms of Apple intelligence, right?

00:48:11   Exactly.

00:48:12   That could be a weak point.

00:48:14   Yeah.

00:48:14   And this Gemini 3.5 flash doesn't look that exciting in benchmarks.

00:48:20   I mean, it's fast, for sure.

00:48:24   It's more expensive than before.

00:48:27   And it continues not to be state-of-the-art for calling external tools.

00:48:32   So, I don't think it's going to be that great.

00:48:37   They're launching with a subset of integrations, and later on, they will let you plug in any MCP.

00:48:46   So, I don't know.

00:48:47   All of these companies, I mean, Google, SpaceX, slash XAI, whatever it's called now, they're

00:48:55   all kind of late to the agent party, so to speak.

00:48:59   And they're trying to catch up now.

00:49:00   Meanwhile, OpenAI and Anthropic are so far ahead, it's kind of ridiculous.

00:49:05   And that makes me, you know, it's like, you know, we're on the precipice of WWDC again, right?

00:49:10   And it's like, oh, are they, do they have anything like this?

00:49:17   Now, there was, I believe, a rumor a while ago about the idea of talking to your Mac,

00:49:22   one of the Mark Gurman rumors, and it being able to do stuff for you.

00:49:26   But I just, I feel like, you know, we're going to do WWDC, and they're going to have a bunch

00:49:30   of stuff, but then the thing everybody else is doing, they don't have an answer for it.

00:49:34   And I figure we're just, this is just what it's going to be for years, just because of

00:49:39   the way that they develop Apple.

00:49:40   Like, I just, if this is the thing, do they have an answer to this thing?

00:49:46   And if we go through WWDC, and then all the questions are, where is Apple's agentic Siri,

00:49:53   right?

00:49:54   Are they going to have to change how they work if they want to be in this world, right?

00:50:03   Because clearly the way Apple works, if that, if this scenario plays out, the way Apple works

00:50:09   just does not fit with the way the AI stuff is developed.

00:50:12   And I just wonder if they're going to have to adapt to that, basically release stuff more

00:50:18   frequently than once a year.

00:50:20   Google search looks like it's going to become rough.

00:50:27   So there's a bunch of things going on with Google search.

00:50:33   And I think the best way that I can describe it is they are just putting more and more AI

00:50:37   into search.

00:50:38   So like, the AI overviews are getting bigger, and they're going to have more stuff in them.

00:50:44   Like, they're going to generate UI for you.

00:50:49   They're going to generate code for you and put it in the Google search.

00:50:51   So you could have these visualizations.

00:50:54   They're going to have agents that can keep searching for something in the background for you.

00:50:59   It feels like it is going to become harder and harder to just find the links.

00:51:06   As a website owner, that's my reaction.

00:51:09   Yeah.

00:51:09   Not good.

00:51:10   It's not great.

00:51:11   It's wild to me.

00:51:13   I mean, so much of this stuff is bananas, which we've spoken about.

00:51:16   But changing search is wild.

00:51:20   That's what pays for all this stuff.

00:51:22   And, you know, they've been, especially on mobile, the Google AI overview, like, it will

00:51:30   take over, like, modally when you search, just in Safari.

00:51:32   Like, I'm not using the Google app.

00:51:34   Maybe it's different there.

00:51:35   But making this the primary interaction for search, while I understand it from the business

00:51:42   perspective of, like, wow, ChatGPT and others have, like, made a pretty good case that LLMs

00:51:49   are useful in search and research and those sorts of things.

00:51:52   I mean, that's the thing Google does.

00:51:55   It's those links and the ads that come along with them.

00:51:58   And it's fascinating that they're willing to change it.

00:52:03   I think it's the, I mean, it's, like, fundamental to who they are as a business.

00:52:08   It's wild.

00:52:09   I mean, you've got to assume that the plan is they'll just start putting ads in the AI overviews

00:52:14   and that's how they pay for it, right?

00:52:16   Yeah.

00:52:17   Like, the AI overviews are making the ads less visible, which means they will just have

00:52:24   to find a way to put the ads in the AI or just put the ads above the AI or, like, whatever.

00:52:30   Because I agree with you.

00:52:31   Like, they're doing this now.

00:52:33   At some point, the rub is going to hit the road, right?

00:52:35   And they're going to have to find ways to monetize whatever Google search is going to be.

00:52:43   Seeing this stuff was, like, maybe gave me the first time where I was, like, maybe I don't want to use Google for Google anymore.

00:52:52   Yeah.

00:52:53   I had the same thought.

00:52:54   And I think I'm not, like, I'm not, like, this is not super close to me, but I feel like, you know, I've explained this before.

00:52:59   If I'm Googling for me, usually, is I have an idea of where I want to go.

00:53:04   And I use ChatGPT for searching for things that I don't know about yet.

00:53:08   And so, like, when I'm Googling, I usually have a destination in mind.

00:53:15   That is very, like, in line with showing me links to stuff.

00:53:21   And if they're going to make that harder and harder, maybe I just need to find another search engine where I can just find links to stuff, you know?

00:53:29   Because that's what I'm looking for out of that product.

00:53:32   Because I don't find Google's AI overviews to be sufficient for the bigger stuff.

00:53:40   Because they try to serve it too quickly.

00:53:42   And I think that's, like, the fundamental issue with the AI overview.

00:53:45   They try to serve it immediately.

00:53:47   And that means the quality usually isn't so good.

00:53:49   That's my theory, anyway, of, like, why AI overviews are much worse than, say, if you were searching for something in Gemini.

00:53:56   So, yeah, I don't know what they're doing over there with Google Search.

00:53:59   It seems like a lot.

00:53:59   And just in general, Google's doing some stuff.

00:54:04   Yeah.

00:54:05   That sums it up.

00:54:09   They're doing some stuff.

00:54:12   This episode of Connected is brought to you by Steam Clock.

00:54:16   If you're like me, you have opinions about the quality of the apps you use.

00:54:20   You know the difference between an app that feels right and one that feels off.

00:54:25   Steam Clock software builds mobile apps for companies that care about taste.

00:54:30   They're a design and development studio based in Vancouver, Canada.

00:54:34   And they've been shipping iOS and Android apps for over 15 years.

00:54:38   Their clients are growing tech companies that care about mobile, but don't have the in-house team to build something great.

00:54:45   Steam Clock works with companies to level up their app so it can go from, it's holding us back, to, it's really pulling its weight.

00:54:55   Some of their clients discovered the hard way that vibe coding your way to the app store is not a product strategy.

00:55:00   Steam Clock has deep experience shipping apps for iOS and Android.

00:55:05   They get these platforms, and they're good at helping companies figure out the right technical approach for their situation.

00:55:11   Apps built by Steam Clock have been downloaded over 10 million times, and they've helped five of their clients through acquisitions.

00:55:20   So if you're building something and need a mobile team that cares as much as you do, Steam Clock is where to start.

00:55:26   Go to steamclock.com slash connected to get in touch.

00:55:31   That's steamclock.com slash connected.

00:55:35   Our thanks to Steam Clock for their support of the show and all of Relay.

00:55:39   So Federico, you have huge news.

00:55:44   You've been working on Shortcuts Playground for a long time.

00:55:47   It's been fun to follow that behind the scenes.

00:55:50   So tell us about this project.

00:55:53   Are we standing in the foothills of the shortcut right now?

00:55:56   Yes.

00:55:57   No, you are on the top of the hill, actually.

00:56:00   Oh, wow.

00:56:01   Wow.

00:56:01   Yeah.

00:56:02   Look around you.

00:56:04   One day, Mike, all of this could be yours.

00:56:06   Oh, yes.

00:56:07   Thank you.

00:56:08   Or as the light touches.

00:56:10   Yeah.

00:56:11   So, okay.

00:56:12   So I launched a bunch of things on Mac Stories today.

00:56:16   The highlight is the Shortcuts Playground.

00:56:19   It's a plugin for Cloud Code or OpenAI Codex that is free and open source.

00:56:25   You can go to the GitHub repo and you can download it.

00:56:29   You can take a look at the code yourself.

00:56:31   You can just ask Cloud or Codex to set it up for itself and it'll do it.

00:56:36   And this is a plugin that lets you create on your Mac, obviously, because these things are Mac only.

00:56:43   It lets you create real Apple shortcuts that you can open in the Shortcuts app and add them to your library.

00:56:51   That's the short version.

00:56:52   You just type a sentence, type whatever you want.

00:56:55   Like, hey, I want a shortcut that gets me directions to my next calendar event.

00:56:58   Or I want a shortcut that shows me how much caffeine I've had today.

00:57:02   You type anything, you get a shortcut back.

00:57:05   Obviously, there's a bit of latency involved because you're having an agent create a shortcut for you.

00:57:12   Depending on the model that you use, the complexity of your idea, you're either going to wait a minute.

00:57:19   Or if it's really complex, you can wait, I don't know, 10, 15 minutes.

00:57:23   Depends on what you want to make.

00:57:24   At the end, you get a .shortcut file on your desktop.

00:57:28   You double-click it, you add it to the Shortcuts app, and that's it.

00:57:30   This was a lot of work.

00:57:32   Yeah, because, like, how does this even work?

00:57:34   Yeah, yeah.

00:57:35   So, it's like, it doesn't feel like it should, you know?

00:57:42   Yeah, shouldn't work.

00:57:44   So, this all started in January when I came across this skill.

00:57:49   Not a plugin, a skill for CloudCode.

00:57:52   It's basically just a bunch of markdown files from this person, Drew, on GitHub.

00:57:58   And the skill, I was fascinated by the idea because it was basically a bunch of markdown files that taught CloudCode how to put together shortcuts.

00:58:09   And this is, like, the core of the whole thing.

00:58:12   Shortcuts behind the scenes have always been fancy XML files.

00:58:20   Aren't we all, really?

00:58:22   Aren't we all, really, if you think about it?

00:58:24   Technically, an RSS podcast is a...

00:58:27   I was going to say, our podcast is an XML file.

00:58:29   It's always been XML, and realistically, it'll always continue to be.

00:58:36   So, this has always been the case since the workflow days.

00:58:40   And even after workflow was acquired by Apple, a shortcut is essentially an XML file that gets signed into a proprietary Apple plist format that itself gets saved as a .shortcut file.

00:58:57   Right now, you can export any shortcut from your shortcuts app as an XML file.

00:59:04   That skill that I discovered at the beginning of 2026 essentially contains the information for Cloud to learn how the XML syntax worked behind the scenes for different actions, variables, parameters, conditional blocks, repeat blocks, everything.

00:59:26   And Cloud code could say, well, okay, I know the syntax, now I'm going to put together a new XML file for a new shortcut.

00:59:32   I thought it was fascinating.

00:59:35   And I started testing it.

00:59:42   It was hallucinating actions.

00:59:44   It was creating actions without variables.

00:59:48   The shortcuts didn't work.

00:59:50   But there was something about it that I thought was worth exploring.

00:59:54   So, I took that foundation and started working on my own version of the skill.

01:00:00   And I realized, well, okay, hold on a second.

01:00:03   If this is based on XML knowledge, and if I can export my entire shortcuts library as XML,

01:00:12   why don't I just teach Cloud code myself with my own shortcuts?

01:00:18   So, starting from a modern foundation of my own shortcuts from my own app,

01:00:25   maybe it'll find a way to update its knowledge from that skill.

01:00:30   And it did.

01:00:32   So, first, I gave it my entire shortcuts library.

01:00:35   And it updated the skill based on that.

01:00:39   And then, so it started snowballing from there.

01:00:42   The next step was, I remembered that the entire shortcuts database, if you have a Mac,

01:00:49   is actually freely available for you to copy, backup, and inspect.

01:00:55   It's an SQLite database that's on your Mac.

01:00:58   And you can just copy that and give it to an agent and say, hey, take a look at this.

01:01:02   And I did that.

01:01:04   So, the agent took a look at that and understood, actually, how variables worked,

01:01:11   how app intents were named, how parameters worked, all of those things.

01:01:18   And then, I said, well, okay.

01:01:20   So, he took a look at my library, took a look at my shortcuts database on Mac OS.

01:01:25   What if I give it the entire Mac Stories shortcuts archive and see what happens?

01:01:32   And so, from there, it scraped my own entire archive on Mac Stories,

01:01:39   downloaded 300 shortcuts, and repeated the same process over and over.

01:01:44   And in all of this, so while all of this is happening,

01:01:47   I'm creating shortcuts, right, with the skill.

01:01:50   It wasn't a plugin.

01:01:51   It was still an improved skill.

01:01:54   I'm creating shortcuts, and it's working.

01:01:57   It's starting to make shortcuts, and I'm giving it feedback.

01:02:00   So, for three months or so, every day, I would sit down, type in some ideas,

01:02:07   create shortcuts, install those shortcuts, go back, test them.

01:02:12   This is all manually done by me.

01:02:13   Test them.

01:02:14   Pass the feedback.

01:02:16   Back to, at that point, I moved from Cloud Code to Codex.

01:02:20   So, pass the feedback to Codex and say, hey, look at this action.

01:02:23   It was not configured properly.

01:02:25   This other action, you actually hallucinated a variable over and over and over.

01:02:31   So, this for about three months.

01:02:33   Then...

01:02:35   You did some model training.

01:02:37   Yes.

01:02:38   Yes.

01:02:39   I actually did a kind of lightweight version of reinforcement learning, basically.

01:02:45   Then, I had another idea.

01:02:51   Once I reached the point where the skill was, like, really good.

01:02:55   It was a really good skill.

01:02:57   But sometimes, it would still break something.

01:03:02   It would still hallucinate an action, a variable, stuff like that.

01:03:06   I remembered that one of the popular techniques in the AI programming community

01:03:14   was this so-called RALF loop.

01:03:18   I think we spoke about this on Connected before.

01:03:20   RALF is this technique, now also a plugin, that you can install, where you basically put

01:03:28   an agent in a loop, where it validates its own work.

01:03:32   And if the validator fails, it gives the feedback back to the agent and says, ah, you wrote some

01:03:40   code, but the code is failing here.

01:03:41   Fix it.

01:03:42   Over and over.

01:03:43   And it's called RALF after the character from The Simpsons, RALF, that does-

01:03:49   Oh, RALF Wiggum?

01:03:50   Yes.

01:03:51   Makes mistakes over and over, but persists no matter what.

01:03:55   So, I thought, well, wouldn't it be funny if I sort of adapted the RALF loop technique

01:04:02   for making shortcuts?

01:04:04   So, in the plugin, you will find my take on the RALF loop.

01:04:10   It's called a Kreg loop.

01:04:13   And who's that named after?

01:04:16   A popular character from the Apple.

01:04:21   From Apple.

01:04:22   The Kreg loop is this huge Python script that anytime the agent tries to say, okay, I think

01:04:32   I got it.

01:04:32   I wrote a shortcut draft to disk, the validator kicks in automatically and tries to validate

01:04:39   the shortcut.

01:04:40   If there are errors, it tells the agent, ah, yeah, you wrote a file, but there are such

01:04:47   and such errors.

01:04:48   Go back.

01:04:48   And this, this is the loop.

01:04:51   And it's why shortcut player gun can be so accurate for a amateur reverse engineered

01:04:57   project.

01:04:57   Because there's the Kreg loop behind the scenes that over and over, anytime Claude or Codex

01:05:04   say, I think I got it.

01:05:05   I wrote something to disk.

01:05:06   The validator comes in and like, let me take a look at this.

01:05:09   Does it validate?

01:05:10   No.

01:05:11   Here's the feedback.

01:05:12   Go back.

01:05:12   Until the validator says, yes, it passed.

01:05:15   At that point, the plugin uses the actual Apple made native shortcuts CLI that signs a shortcut.

01:05:29   And that's why you're able to install it in the shortcuts app.

01:05:35   The final, so that was my big breakthrough was that idea of putting a validator and putting

01:05:42   the agent in a loop.

01:05:43   The second breakthrough was using Codex in a loop for multiple days with computer use to

01:05:57   create a thousand shortcuts.

01:05:58   Codex, Codex thought about, Codex took a look at the entire documentation.

01:06:07   There's a lot of documentation in the skill.

01:06:09   Took a look at the entire skill, made a plan to create a thousand shortcuts for all the possible

01:06:16   categories and all the possible actions.

01:06:19   And it worked in a loop to create a thousand shortcuts.

01:06:25   Then it used computer use to install those shortcuts, actually double click them, open them in the

01:06:32   shortcuts app, take a look at them, test them.

01:06:35   And if something was visually broken or functionally, like if something wasn't working, it would see

01:06:43   it on screen, go back and repeat that process over and over until that particular shortcut out

01:06:51   of a thousand shortcuts was working.

01:06:54   So it worked for multiple days on this, on its own.

01:06:57   And every once in a while I would screen share, log in and see what was going on.

01:07:02   Just scream.

01:07:06   And that's, and that's, and that was the other breakthrough.

01:07:09   Like actually putting the agent in a long running loop for multiple hours at a time.

01:07:14   And then about a month ago, I thought I had it.

01:07:20   So I started putting together everything that you see on Mac stories today.

01:07:24   the redesigned shortcuts, the redesigned shortcuts archive, the dedicated mini website for, for, for shortcuts playground.

01:07:32   And there's a companion shortcut that's for paying members only.

01:07:38   That, that's also like the crazy part.

01:07:42   The companion shortcut is called shortcuts playground remote, because I wanted to convey the idea that it's an optional, like remote experience thing.

01:07:50   That is a shortcut that creates shortcuts on device.

01:07:57   You actually run it on your iPhone and iPad and you type in an idea.

01:08:01   And a few minutes later, you see a new shortcut import window on your iPhone or on your iPad.

01:08:08   And you may wonder how that's possible.

01:08:10   It's actually talking behind the scenes to your plugin installed on your Mac via SSH.

01:08:18   And it's getting the shortcut file back as, drum roll, base 64.

01:08:25   Oh, no, no, wow.

01:08:27   Yes.

01:08:28   And that's why it can, it can transfer that shortcut from your Mac back to your iPhone or iPad.

01:08:35   That's a companion shortcut.

01:08:37   It was very difficult to put together.

01:08:39   There's an auto, there's an automation academy guide and sort of story about how I did it.

01:08:44   It's very complex.

01:08:46   But since I started from an open source skill, and since I, throughout this process, I drew from a lot of previous open source projects for shortcuts.

01:08:59   That's why I decided that the plugin should be free and open source for everybody to use.

01:09:06   There are a lot of nice things about the plugin.

01:09:11   Like, for example, you can ask for shortcuts to be installed and to have a specific color or to have a specific icon glyph.

01:09:22   And there's actually support for icon synonyms.

01:09:25   So it actually supports the official shortcuts glyph library.

01:09:30   But if you don't want to be using the exact name of an icon that Apple is using, you can ask for a synonym of that icon.

01:09:38   The shortcut knows how to make health kit actions.

01:09:42   Sorry, the plugin knows how to make health kit actions.

01:09:46   It knows how to make complex actions with conditional blocks, API calls.

01:09:53   Anything you can imagine, you can ask.

01:09:56   And the final thing I will say is, since it's running in an agent on desktop, this thing is running on macOS.

01:10:07   So it can use a terminal.

01:10:08   So one of the nice things that I've discovered is, if you ask shortcuts playground to, like, can you make me a Todoist shortcut for, I don't know, getting my tasks for Todoist due today?

01:10:22   And let me choose which ones I want to reschedule to another day?

01:10:26   The agent will actually test the Todoist API on its own before building the shortcut.

01:10:36   This was actually one of the best practices that I sort of infused myself into the plugin by telling it,

01:10:47   anytime the user asks you to make a shortcut that involves a web API, try to test the web API yourself so that you don't hallucinate URLs or API responses.

01:11:01   So, I don't know.

01:11:03   I'm pretty proud of the work that I've done.

01:11:05   And like I wrote in the conclusion, I fully expect to be Sherlocked in a few weeks by Apple.

01:11:13   Well, maybe, right?

01:11:14   Maybe.

01:11:15   I don't know.

01:11:16   Like Mark Garman said, in the updated shortcuts app, users presented a prompt asking what do you want your shortcut to do, along with a text field to describe the request.

01:11:26   The system then automatically builds and installs the shortcut on device.

01:11:29   So, sounds pretty good, but you never know, right?

01:11:32   Like with this stuff, you never know how good it's going to be or not.

01:11:36   So, maybe there'll still be a case for shortcuts playground even in a 27 release here.

01:11:43   I think this is something that is definitely very impressive and also hard to get my head around, right?

01:11:50   Because it's very complex.

01:11:52   Maybe your craziest project yet, honestly, of which you've done many.

01:11:59   I've done many.

01:12:00   But I think what's different about this one is that it's actually useful for other people.

01:12:07   Whereas in the past, when I did things like the MacPad, for example, it was more, I don't want to say self-referential, but it was always to prove a point to myself and share a story about myself.

01:12:20   Whereas I think what I'm trying to do now and what I will continue doing is release more free and open source things for other people to use.

01:12:33   Because I feel like I want to be able to, when people ask me the question of like, yeah, sure, but what have you built with AI that's actually useful or that's actually making you money?

01:12:47   I want to be able to point at stuff and I'm very happy today because when people, now when some people ask, oh yeah, but what have you built with AI that's actually useful?

01:12:58   I can point them to something practical, the new shortcuts archive on Mac stories, the plugin, the shortcut for members, like that is real stuff that is helping our business.

01:13:10   And of course, the backdrop of it all, I learned a lot.

01:13:16   Like I learned things I had no idea about and it was the first time, this was the real first test for this new way of working that we have a Mac stories where there's me or John and we have our agents.

01:13:34   And at the end of the process, we have our developer that's review, who's reviewing all the work of these agents and who's actually taking the output of our ideas and putting them live on Mac stories.

01:13:48   So like it's this new way of like hybrid work.

01:13:52   I don't know how to describe it, but if I like the truth, like if I didn't have agents, there would be no shortcut archive.

01:14:00   There would be no dedicated website.

01:14:02   There will be no plugin.

01:14:03   I don't know.

01:14:05   It's, it's all very different for me, but people seem to like it.

01:14:10   And, uh, and I don't know, maybe Apple won't.

01:14:15   I do wonder.

01:14:17   I mean, this, obviously we know you've been doing this for a while and there's definitely been some conversations about like, is this thing going to last?

01:14:25   Will this see out the week?

01:14:29   We'll find out, you know, one way or another.

01:14:32   Uh, but it is for as long as it exists, it is very, very impressive and complex.

01:14:39   Thank you.

01:14:39   Um, and is a, this is a true Federico Vatici project.

01:14:46   Uh, yeah, my next one.

01:14:49   Uh, so I'm trying to release a bunch of these things before WWDC.

01:14:53   And there's a, I have two more projects that are ready to go.

01:14:57   So I think realistically, I'm going to have time for one.

01:15:00   So my next one is, I mean, I can just mention it on the show.

01:15:03   I built a, uh, a command line interface for reminders, uh, on the Mac.

01:15:09   And it's the, the, uh, I mean, not to brag.

01:15:14   It's, it's the best reminders CLI, um, on available.

01:15:19   I mean, once it will be available, uh, because it's the only one that supports all.

01:15:24   And I mean, all of the reminders features, including the private API ones.

01:15:30   And yes.

01:15:31   I can't say that out loud.

01:15:33   Well, private.

01:15:35   It's private.

01:15:36   It'll be free and open source.

01:15:38   And that's what the Mac lets you do.

01:15:42   Yeah, man.

01:15:43   When you have, when you have stuff available in Finder for you to inspect and tinker with,

01:15:49   that's what happens.

01:15:50   But you, you promise you're not going to bring the singularity.

01:15:54   Are we standing in the foothills of the Federico singularity?

01:15:59   You can stand in my foothills.

01:16:05   That's all I've ever wanted.

01:16:06   Okay.

01:16:07   All right.

01:16:08   Well, I think that about does it for this episode of Connected.

01:16:14   If you want links to the stuff we spoke about, there are in, they're in your podcast players.

01:16:19   Like where are my voices coming out of?

01:16:21   There's links in there.

01:16:22   Go check them out.

01:16:23   They're also on the web at relay.fm slash connected slash 604.

01:16:27   A couple of other links you should know about.

01:16:29   You can get connectedfeedback.com.

01:16:31   Head over there.

01:16:32   You can leave us a note.

01:16:33   You can leave us a nice note.

01:16:35   You can leave us an anonymous note.

01:16:36   You can tell us your WBC predictions.

01:16:39   Just have at it in there.

01:16:41   Or tell us what's going to happen at WBC.

01:16:44   If you know that.

01:16:45   If you know.

01:16:45   Let us know.

01:16:46   We'll take it.

01:16:48   You can join and get longer ad-free versions of the show.

01:16:52   We do those each and every week.

01:16:53   There's extra content at the beginning of the show.

01:16:55   No ads.

01:16:56   And then we put titles at the end.

01:16:58   You can get that for just seven bucks a month at getconnectedpro.co.

01:17:03   And again, that link is in the show notes.

01:17:04   If you want to find more of us, we're online.

01:17:08   Prince Flexi Federico is at maxstories.net.

01:17:10   We just talked about his new project.

01:17:13   And the promise of many more to come checks calendar in three weeks.

01:17:17   So, make Max Stories your homepage, I guess.

01:17:20   That's what I'm saying.

01:17:22   You can find Mike across a bunch of other shows here on Relay and his work at Cortex Brand.

01:17:27   And he blogs at theenthusiast.net.

01:17:29   I've been your attorney, General Flexi.

01:17:32   And you can find my writing at 512pixels.net.

01:17:36   I'd like to thank our sponsors this week for sponsoring the show, making it possible.

01:17:40   Squarespace and Steam Clock.

01:17:42   And until next time, guys, say goodbye.

01:17:44   Arrivederci.

01:17:45   Goodbye, f***s.

01:17:46   No!

01:17:48   No, at the end, too.

01:17:50   I've got to get it in.

01:17:52   Come on, we'll give him one.

01:17:53   Yeah, okay.