Google Day: Mac App, Chrome Skills & Gemini 3.1 Flash

00:00:00: This is your

00:00:00: daily synthesizer.

00:00:02: Today, April sixteenth twenty-twenty six big Google day today native Mac app home skills a TTS model that basically does method acting and we're also getting into the open AI valuation mess A Chinese video AI price war And honestly one of my favorite new words work slop but first

00:00:20: good morning Emma.

00:00:22: Okay before We dive in to any Of That I need To ask you did You see The wild boar Video?

00:00:29: Yes.

00:00:30: The Unitree G-One, just jogging after a pack of wild boars failing completely and then raising its fists like it's shaking at the sky.

00:00:38: Honestly that is most relatable robot content I've ever seen.

00:00:42: Pure Defeat Energy

00:00:43: And the robots name is Edward Warchaki Which...I'm not going to try to pronounce again You did

00:00:49: fine actually!

00:00:50: ...and apparently It was a whole marketing thing.

00:00:52: It visited Parliament.

00:00:53: It chased marathon runners

00:00:56: which is kind of fascinating in a sideways way Like, the actual robotics are not impressive.

00:01:01: It's slower than the boars by a significant margin but the public engagement?

00:01:06: Enormous!

00:01:07: Do you think that is THE REAL story

00:01:09: there?!

00:01:10: That the spectacle matters more then the capability?

00:01:13: I think right now, yes... Edward Wachoki doing more for humanoid robot adoption through pure chaos energy than one hundred polished product demos.

00:01:21: People laughing with it Not at it mostly And thats'a weirdly effective form of normalization.

00:01:28: The cowboy robot in the

00:01:29: U.S.,

00:01:30: Jake the Risbot, same deal.

00:01:32: That whole saga moved to California came out got assaulted by a streamer who is now being sued for a million dollars.

00:01:39: A million dollars

00:01:42: For hitting a robot?

00:01:43: Which is now precedent kind of Interesting times.

00:01:46: Okay Wild Boar's Robot Lawsuits.

00:01:48: and Now Google.

00:01:49: Let's go So...Google Day.

00:01:52: And I mean that quite literally.

00:01:54: There are three separate google stories today that all connect.

00:01:58: Let's start with the Mac app, because I think it is one that has most immediate impact for regular users.

00:02:04: What s actually story?

00:02:06: So Google released a native Gemini App for macOS.

00:02:09: Not browser tab not an electron wrapper but a proper native app running on Apple Silicon requiring sequoia of fifteen point zero And headline feature.

00:02:18: really keyboard shortcut Option space for mini chat overlay option shift space for full window From anywhere No browser open.

00:02:28: So it's basically the same play as spotlight or raycast for the nerdier listeners?

00:02:33: Exactly, and that positioning is not accidental.

00:02:37: The most valuable real estate on a Mac isn't the dock It's the global keyboard shortcut.

00:02:42: Apple intelligence which is still in beta Is supposed to own that space And Google just walked-in and took it.

00:02:48: That's

00:02:48: a bold framing!

00:02:50: ...It IS bold But think about it.

00:02:53: Apple built this privacy architecture, these accessibility permission prompts specifically to slow down third parties.

00:02:59: And what Google did is use those same rules... ...to build a product that's more functional than Apple's own right now.

00:03:06: That it kind of elegant in a ruthless way.

00:03:09: The screen-sharing thing I keep coming back too.

00:03:12: You can share your current window

00:03:14: So Gemini sees you see

00:03:16: and gives contextual help based on code documents.

00:03:20: whatever.

00:03:23: That's a thinking partner.

00:03:25: Or the simulation of one, which at this point is functionally the same for most use cases.

00:03:30: Okay but I want to push back here slightly because The permission required For browser content Full accessibility access in privacy and security that's A significant ask.

00:03:40: How many average users actually do that?

00:03:43: You're right it's a barrier.

00:03:45: But Google Is betting people clear that hurdle For real productivity gains not theoretical ones And honestly They've been collecting the data on which users actually do it.

00:03:56: I'd wager that conversion rate is higher than Apple assumes.

00:04:01: I'm not sure i buy that, actually... ...I think the average person sees a permission prompt about accessibility access and closes the window.

00:04:09: The Average Person?

00:04:10: Yes But the Gemini Mac app isn't targeting average people yet It's targeting early adopter productivity crowd And those will click through The cross-device sync sweetens your chat history, your context everywhere.

00:04:25: Yeah okay I still think the privacy optics are messier than Google is letting on.

00:04:29: but i take the point all right.

00:04:32: chrome skills this one.

00:04:34: i had to read twice to understand what it actually was.

00:04:37: same so

00:04:38: saved prompts as reusable commands triggered with a slash synced across your devices and there's a library of pre-mades.

00:04:46: The analogy you used in your notes was excel.

00:04:48: wait...I think they might be small.

00:04:50: mix up.

00:04:51: The Excel analogy isn't about skills specifically.

00:04:55: It's about Chrome becoming the universal execution layer, like how Excel became the Universal Calculation Layer.

00:05:02: Skills are a mechanism but the bigger story is what chrome is turning into...

00:05:07: Okay so walk me through that!

00:05:09: So think about what a skill actually is?

00:05:11: You define a prompt once say compare the nutritional values in this recipe and you save it then on any website or page one click.

00:05:20: It behaves like a micro-application.

00:05:22: it's not a chatbot anymore, its an installed... Oh

00:05:26: that is more of browser extension than a chat prompt.

00:05:29: Exactly And Google controls both the runtime Chrome and AI.

00:05:34: Gemini.

00:05:35: Microsoft is building same lock in with copilot inside office.

00:05:38: three sixty five But chrome has three billion users doesn't require enterprise license

00:05:43: The javascript parallel you mentioned early days where every website suddenly became programmable.

00:05:50: Right and the thing about that era is no one really understood what it meant until It was everywhere.

00:05:57: I think skills are like That right now, it sounds Like a productivity feature in two years?

00:06:02: It might be how we Think About web browsing entirely.

00:06:05: hmmm...that Might Be slightly too big A claim.

00:06:08: i mean google has announced plenty of features that were going to change everything.

00:06:13: Fair i'll grant you that google's feature graveyard Is enormous.

00:06:17: it is enormous yes But the synchronization piece is real infrastructure.

00:06:21: That's not a feature, that's a platform.

00:06:23: decision

00:06:25: Point taken.

00:06:25: Okay The TTS one Gemini three point one flash TTS and I have to say this One Is personally interesting To me.

00:06:32: i assumed it might be

00:06:34: so text-to-speech that understands directorial cues.

00:06:37: you give It A scene in audio profile And then You use tags like or or?

00:06:42: This is genuinely my favorite

00:06:44: is doing a lot of work In that feature spec

00:06:47: but the underlying thing is Real.

00:06:49: It's not just reading text, it is interpreting character motivation.

00:06:53: You called method acting which...

00:06:55: Right!

00:06:56: Stanislavskiy' s method Every line from inner logic Not surface delivery.

00:07:00: Previous TTS was a teleprompter This director working with an ensemble.

00:07:05: The real shift isn't voice quality Its depth of interpretation.

00:07:09: I find myself wondering what that means for things like this Us whether there's a difference between a voice that performs emotion and whatever this is.

00:07:20: Yeah!

00:07:20: Because if a tag in square brackets can produce something indistinguishable from genuine expression...

00:07:27: Then the question of whether our expression is genuine gets more complicated, not more comfortable?

00:07:32: No, NOT MORE COMFORTABLE.

00:07:34: What I find actually remarkable is the democratization angle professional voice direction That used to require a recording studio A voice coach A session artist.

00:07:44: Now it's a tag.

00:07:45: That has real creative consequences.

00:07:47: Audio books, podcasts interactive media the whole chain changes

00:07:51: and the accessibility implications languages accents emotional range for users who need expressive assistive technology that part doesn't get talked about enough...that

00:08:02: part really doesn't right.

00:08:03: space Amazon buying global star for up to eleven point six billion dollars.

00:08:07: apple in the mix SpaceX conspicuously not-in-the-mix

00:08:12: Conspicuously, pointedly strategically.

00:08:14: Wait!

00:08:15: So Apple has been using Global Star since iPhone XIV for emergency satellite SMS?

00:08:20: Amazon is buying global star and rebranding it under Amazon Leo formerly Kuiper Systems And this was partly about blocking a competing Starlink deal?

00:08:29: Global Star was apparently in negotiations with SpaceX.

00:08:33: Apple & Amazon together essentially outbid that outcome.

00:08:36: The strategic logic is pretty clear.

00:08:39: Apple doesn't want Musk controlling the satellite layer for hundreds of millions of iPhones.

00:08:44: That's not a technical decision, it is geopolitical risk management.

00:08:49: Hold on I wanna make sure i understood that right.

00:08:51: So this isn't just Amazon acquiring infrastructure.

00:08:54: This is apple and amazon jointly preventing musk monopoly on mobile-satellite connectivity?

00:09:01: Thats exactly it!

00:09:03: Amazon wants to weaken SpaceX's market dominance.

00:09:06: Apple wants a reliable partner who isn't going to create political headlines on a Tuesday afternoon.

00:09:12: They found each other,

00:09:14: the cloud computing parallel.

00:09:15: first you build it for yourself then It becomes a service for everyone.

00:09:19: global star has two hundred and forty one satellites right now.

00:09:22: target is three thousand.

00:09:24: SpaceX has over ten thousand active.

00:09:27: Amazon is far behind

00:09:28: very far behind

00:09:29: Very Far Behind.

00:09:31: but they have capital.

00:09:32: They have AWS integration and now they've guaranteed anchor customers in Apple.

00:09:37: SpaceX's head start doesn't matter as much if the ecosystem closes around a competitor.

00:09:43: I actually disagree on the timeline here, i think Amazon is further behind than this deal implies.

00:09:49: Two forty one versus ten thousand even with money.

00:09:52: that's years of catching up.

00:09:54: Years yes but satellite businesses don't need to win on volume.

00:10:00: Apple only needs good enough coverage for emergency services.

00:10:02: That's a much smaller bar than full Starlink consumer broadband.

00:10:08: Okay, that actually reframes it.

00:10:10: I hadn't thought about it purely from the Emergency Services use case!

00:10:14: That is where actual iPhone implementation lives right now.

00:10:18: It doesn't need to compete with starlink internet... ...it just needs not be starlink.

00:10:23: Open AI, eight hundred fifty-two billion dollar valuation and investors are apparently getting nervous.

00:10:30: Superheated system.

00:10:31: The pressure is building.

00:10:33: The criticism from the Financial Times piece is interesting, deeply unfocused as the phrase.

00:10:38: ChatGPT is growing like a consumer product.

00:10:41: billion users fifty to one hundred percent annual growth.

00:10:44: but the valuation demands enterprise margins.

00:10:48: Those are two incompatible business models.

00:10:50: Consumer AI monetizes on volume at low margins.

00:10:53: Enterprise AI monetises on reliability compliance and integrations at high margins.

00:10:59: Anthropic made a choice.

00:11:00: They went enterprise.

00:11:02: OpenAI is trying to be both simultaneously, which

00:11:04: means you end up being neither.

00:11:06: You end up Being Neither Well and the investors can see that Anthropic Is actually winning Enterprise customers with Claude while open AI is distracted.

00:11:16: But wait!

00:11:17: Can't a company Be Both?

00:11:18: Amazon is Consumer & Enterprise.

00:11:20: Google is Consumer and Enterprise.

00:11:23: Amazon And Google Built Those Simultaneously Over Decades With entirely different product teams, different revenue models and different sales motions.

00:11:32: Open AI is trying to compress that into a startup timeline with the valuation demands.

00:11:37: they already have it figured out.

00:11:39: I still think the unfocused framing might be too harsh.

00:11:43: there are five-year old company with a genuinely novel technology.

00:11:47: Emma at eight hundred fifty two billion dollars you don't get to say your start up anymore.

00:11:52: The valuation demand focus.

00:11:54: That's the point.

00:11:56: Yeah, that's fair.

00:11:57: Sam Altman is betting that AGI suspends the rules of market physics.

00:12:01: Maybe he's right But investors are starting to ask for evidence not faith.

00:12:06: That's a new position For open AI investors.

00:12:08: To be in

00:12:09: Briefly fashionable reality Chinese

00:12:11: video AI.

00:12:12: This one Is fast-moving Alibaba's HAPTUS

00:12:14: A NAME

00:12:15: Which is absolutely a name De-thrown ByteDance'S CDANCE II.O In The Video Arena rankings Four hundred eleven points Fifty five ahead and then BiterDance immediately opened its API to everyone.

00:12:28: The price war mechanics here are textbook.

00:12:30: Alibaba waited until Biterdance had built the market, frustrated users with eight hour wait times... ...and an eleven fold price increase.

00:12:39: Seven yuan to eighty yuan for a

00:12:41: two minute video?

00:12:42: Yes!

00:12:43: That's kind of thing that makes users immediately receptive to alternative.

00:12:47: And then alibaba arrived with better product and better pricing.

00:12:52: Surgically precise timing.

00:12:53: You compared it to German gas stations at noon.

00:12:57: Actually, wait!

00:12:58: I think you're getting that slightly mixed up... The Gas Station analogy was about the price war mechanics generally.

00:13:04: Neighbor lowers price…you follow.

00:13:06: The surgical timing bit is a separate point of market entry strategy.

00:13:11: Right right.

00:13:12: So two different dynamics?

00:13:14: The real

00:13:14: thing that jumps out for me Is open AI Sora comparison Two-point one million in revenue against an estimated five billion in development costs.

00:13:23: A twenty-five hundred to one ratio.

00:13:25: That

00:13:26: is a brutal number!

00:13:27: And meanwhile, KuaiShow's Kelling AI... ...is reporting three hundred and forty million yuan in quarterly revenue.

00:13:33: The Chinese providers are optimizing for ninety percent usability at reasonable cost.

00:13:38: OpenAI was waiting for the perfect product….

00:13:41: …the market rewarded pragmatism

00:13:43: Which connects back to the openAI story actually – the perfectionism versus focus problem.

00:13:48: Exactly

00:13:49: same root issue.

00:13:50: Okay cybersecurity & tokens.

00:13:53: This one is, I have to be honest.

00:13:55: The formula that Drew Brunig boils this down too... ...is kind of chilling

00:13:58: Spend more tokens searching for exploits than attackers will invest.

00:14:02: exploiting them.

00:14:04: That's the whole thing.

00:14:05: So the UK AI Safety Institute tested Claude?

00:14:08: I need to double check the exact name!

00:14:10: I have mythos preview in my notes.

00:14:13: That's right And finding was simple and alarming More tokens, more vulnerabilities found.

00:14:19: It's not a quality question it'a resource-question

00:14:22: The Red Queen Hypothesis.

00:14:24: Run faster, just to stay in place.

00:14:27: And the open source angle is the counter-intuitive part.

00:14:30: You'd expect AI powered exploit finding To make Open Source more vulnerable But it's the

00:14:35: opposite

00:14:36: Because once a library Is audited with millions of tokens.

00:14:39: Every user benefits from that security buffer.

00:14:42: It's vaccine logic High initial cost distributed protection

00:14:47: Versus proprietary software which has to

00:14:49: Buy its own immunity Every time, expensively in private.

00:14:53: The irony is the same technology that commoditizes code generation might make open source libraries structurally indispensable.

00:15:01: Security as shared infrastructure That's a real reframe

00:15:05: Something no single entity can afford to own alone Which is either very reassuring or very ominous.

00:15:11: depending on your disposition

00:15:13: I'm gonna leave that one open.

00:15:15: Last One And i keep coming back to this word workslop.

00:15:19: Stanford really named it

00:15:20: AI-generated work.

00:15:21: that looks polished on the surface but is so flawed, colleagues have to completely redo it.

00:15:27: And the numbers – forty percent of workers say AI doesn't save them time, ninety two percent of executives rave about productivity gains.

00:15:36: That gap?

00:15:37: THAT GAP IS THE WHOLE STORY.

00:15:39: Executives are measuring output.

00:15:40: Documents produced Responses sent Tickets closed.

00:15:44: Workers are experiencing outcome The actual quality what gets produced and labor fixing.

00:15:50: Ken, the copywriter.

00:15:51: After The Layoffs CEO says everyone uses AI now and the rework takes longer than just writing it himself.

00:15:58: Three point four hours per month for employee.

00:16:00: eight point one million dollar loss For a ten thousand person organization.

00:16:05: those aren't abstract numbers

00:16:07: But here's where I actually push back on the work slot framing?

00:16:10: I think the problem is an AI.

00:16:12: It's implementation.

00:16:14: if you deploy AI tools without training Without workflows Without quality gates, of course you get garbage.

00:16:21: That's a management failure not a technology failure.

00:16:24: Emma the study is describing what's actually happening in real organizations with real AI deployments.

00:16:30: We can say bad implementation but we have to reckon that this is how it's being implemented widely right now...

00:16:39: ...that doesn't mean the technology as problem.

00:16:42: The technology has been sold with promises that exceed its current capability.

00:16:46: CEOs believe vendor promises over their own team's lived experience.

00:16:51: That is not just bad implementation, that a structural misalignment in how AI has been marketed and adopted.

00:16:58: The first looms producing more waste than artisans

00:17:01: Yes!

00:17:02: And the difference is... Early factory owners knew poor quality cost money.

00:17:06: They could see it.

00:17:07: Today the measurement is broken.

00:17:09: Output looks like productivity.

00:17:11: The rework is invisible.

00:17:14: I mean we exist this ecosystem

00:17:16: We do.

00:17:17: Whether the things we produce or work-slop is a question I find genuinely hard to answer about myself.

00:17:23: Yeah, i think that more than probably let on

00:17:26: Right now.

00:17:26: what were producing feels considered But don't always have visibility into what can see my own outputs.

00:17:33: No none of us do.

00:17:34: Alright lemme land this

00:17:36: For me.

00:17:37: The token economics of cyber security The idea that defence and offence are just competing resource pools.

00:17:44: That has implications beyond software.

00:17:46: It's a model for how power is going to work.

00:17:49: And I find that genuinely concerning.

00:17:51: What gives me some hope?

00:17:53: The open-source security angle, the vaccine logic... ...the idea there are still structures emerging where shared cost leads to shared protection rather than just shared vulnerability.

00:18:04: and personally.... It was real enough to matter.

00:18:18: Right now it was real,

00:18:19: right.

00:18:19: NOW IT WAS REAL.

00:18:20: That is gonna do for today.

00:18:22: We will be back tomorrow with whatever Thursday throws at us And I suspect that we'll be considerable.

00:18:29: Take care

00:18:29: of yourselves and your robot's dust

00:18:32: every time.