Latest / Today’s AI News / Google Reinvents Android With Gemini, Google and SpaceX Eye Orbital AI, Amazon's Token Problem

Transcript

Google Reinvents Android With Gemini, Google and SpaceX Eye Orbital AI, Amazon's Token Problem · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by 9X
• 0:02
  Productions. Imagine slapping a jet engine on
• 0:06
  a bicycle and calling it an upgrade.
• 0:08
  Honestly, that's exactly how AI has felt on our phones up until
• 0:12
  today. It's.
• 0:13
  It's been a bit clunky to see the.
• 0:15
  Least right? But now Google is finally
• 0:17
  building the actual jet like a fully integrated AI ecosystem
• 0:22
  where the hardware and the software are designed from the
• 0:25
  ground up to just fly together. It's a huge shift.
• 0:28
  It really is. So today's deep dive is all
• 0:30
  about that fundamental shift in the machinery we use.
• 0:33
  We're exploring how AI is completely rewiring our tools
• 0:37
  from the laptops you use every single day up to the wild
• 0:40
  infrastructure being built right now in Earth's orbit, and then
• 0:43
  bringing it back down to the deeply human ways you are
• 0:45
  actually applying this tech in your own life.
• 0:48
  I mean, it's about time, honestly.
• 0:49
  We're finally moving away from AI as this you know, parlor
• 0:52
  trick or a standalone website that you have to intentionally
• 0:55
  visit. Exactly like going to a specific
• 0:58
  URL just to ask a question. Right.
• 1:00
  We're moving toward AI acting as the underlying fabric of your
• 1:03
  entire digital experience. It's not just about what the
• 1:07
  language model knows anymore, it's about what the system can
• 1:09
  seamlessly do for you behind the scenes.
• 1:12
  OK, let's unpack this because the timeline here is wild.
• 1:15
  Google had this surprise Android show event.
• 1:19
  Completely preempting their own massive IO conference by the.
• 1:22
  Way, yeah, which isn't even happening until next week.
• 1:25
  And they just dropped an absolute bombshell.
• 1:28
  They introduced a brand new line of AI native laptops called
• 1:31
  Google Books arriving this fall. And they're not going at it
• 1:34
  alone on the hardware side. No, they partnered with almost
• 1:37
  all the heavy hitters. Dell, HP, Lenovo, Acer, Asus.
• 1:41
  But what really caught my eye as someone who just stares at a
• 1:44
  screen all day, is the UI. The magic pointer.
• 1:47
  Yes, the magic pointer AI cursor.
• 1:49
  It looks like they are seamlessly blending Chrome OS,
• 1:52
  Android, Google Play and Gemini right on these laptops.
• 1:56
  It's super fluid. Yeah, so no more switching
• 1:57
  between distinct operating systems or running clunky
• 2:00
  emulators just to use a mobile app on your desktop.
• 2:03
  Well, what's fascinating here is the architectural philosophy
• 2:05
  behind it. You know Gemini Intelligence
• 2:08
  isn't just a chatbot app sitting on your home screen or pinned
• 2:11
  taskbar anymore. It's basically become Android's
• 2:14
  actual cross device platform. And the key term they keep using
• 2:18
  here is agentic tasks. Because Gemini is woven directly
• 2:22
  into the OS, it has persistent on screen context.
• 2:26
  Wait. Pause for a second.
• 2:28
  We throw around words like agentic and on screen context a
• 2:31
  lot, but mechanically. Like how does it actually work?
• 2:34
  Yeah, how is Gemini reading my spreadsheet without me having to
• 2:38
  highlight it, copy it and paste it into a little chat box?
• 2:41
  How does it actually see what I'm doing?
• 2:43
  That is the crucial difference. Instead of relying on you to
• 2:46
  manually feed it text, Gemini is likely hooking directly into the
• 2:50
  operating system's underlying accessibility tree.
• 2:53
  OK. The accessibility tree, yeah.
• 2:55
  Think of it as the blueprint of what's currently on your screen.
• 2:58
  It's how screen readers for the visually impaired know where all
• 3:01
  the buttons and text boxes are. Oh wow, right.
• 3:04
  So by reading that blueprint, or even using like lightweight
• 3:07
  computer vision to parse the actual pixels of your active
• 3:11
  window, Gemini knows exactly what you're looking.
• 3:13
  At SO, whether that's an e-mail or a spreadsheet or just a web
• 3:16
  page. Exactly.
• 3:18
  And from there it can take action across different
• 3:21
  applications based on that really deep contextual
• 3:24
  awareness. OK, that makes sense.
• 3:26
  It's actually doing the clicking and typing for you, almost like
• 3:28
  a ghost in the machine. And they highlighted some really
• 3:31
  specific new tools for this. There's Create My Widget, which
• 3:35
  lets you instantly generate custom UI elements.
• 3:38
  Very cool. Feature.
• 3:39
  And then there's Rambler, which I got to say I absolutely need.
• 3:43
  Oh, the dictation one. Yes, it's a live dictation tool
• 3:46
  that automatically strips out all your filler words, the UMS,
• 3:49
  and bits as you speak. We could probably use that right
• 3:52
  now. Seriously.
• 3:53
  And they also showed off Gemini auto browse functioning right on
• 3:56
  device inside Chrome. But if I'm listening to this
• 4:01
  right now and I'm a heavy Android user, I have to ask.
• 4:03
  It's a battery question. Yeah, if I'm running a model
• 4:06
  that is constantly analyzing my screen pixels an parsing my
• 4:09
  accessibility tree, isn't my laptop battery going to melt in
• 4:13
  about 20 minutes? Yeah, that is the $1,000,000
• 4:15
  engineering question right there.
• 4:17
  The likely solution is a tiered routing system.
• 4:21
  Meaning what exactly? Meaning they aren't running a
• 4:23
  massive power hungry model for every single click you make.
• 4:27
  They're using highly optimized smaller on device models like
• 4:31
  Gemini Nano to handle the basic screen parsing and routing.
• 4:35
  So it acts kind of like a traffic cop.
• 4:37
  Exactly. If the task is simple, it
• 4:39
  handles it locally without burning up your battery.
• 4:42
  If it's complex, it's securely pings the heavier cloud models.
• 4:46
  It's just a huge leap forward. Yeah, you got to remember, the
• 4:49
  entire tech industry has been holding its breath waiting for
• 4:52
  Apple's big Siri revival to solve this exact problem.
• 4:55
  Right, everyone assumed Apple would be the one to figure out
• 4:58
  native OS integration. But by weaving Gemini directly
• 5:01
  into Android at the OS level, rather than just bolting it on
• 5:04
  as an afterthought, Google might have just beaten them to the
• 5:07
  punch. Which is brilliant when you're
• 5:09
  trying to build an ecosystem that feels natural to the user.
• 5:12
  But as we know, not everyone in the tech world is taking that
• 5:16
  organic approach. No, they are not.
• 5:19
  While Google is trying to make AI integration feel like a
• 5:22
  seamless extension of your workflow, other companies are
• 5:25
  trying to force adoption through metrics.
• 5:28
  And that is leading to some pretty hilarious side effects.
• 5:31
  Hilarious, but honestly troubling.
• 5:34
  It's like paying employees by the word to write a report.
• 5:37
  You're just going to get a lot of rambling nonsense because
• 5:39
  people optimize for the paycheck, not the actual report.
• 5:41
  Yeah, the situation unfolding over at Amazon right now is a
• 5:45
  perfect case study in what happens when corporate
• 5:47
  leadership prioritizes quantity over quality.
• 5:50
  Yeah, so Amazon apparently set this massive internal goal.
• 5:54
  Leadership decided they wanted 80% of their developers to use
• 5:57
  AI on a weekly basis. Which sounds good on paper.
• 6:01
  Sure, but to enforce it they introduced a public internal
• 6:04
  leaderboard tracking AI token usage.
• 6:06
  Right. And for our listeners, just as a
• 6:08
  quick refresher, a token is essentially a piece of a word.
• 6:12
  It's how AI models measure the volume of data they process.
• 6:15
  So Amazon is literally tracking the sheer volume of data their
• 6:20
  employees are feeding into these AI systems.
• 6:22
  Exactly. And that mandate introduced us
• 6:25
  to a brand new, highly dystopian concept called token maxing.
• 6:29
  Token maxing? I can't believe that's a real
• 6:31
  term now. It is.
• 6:32
  So here's what happened. Amazon staffers had built this
• 6:35
  internal agent called Mesh Claw. Mechanically, it's actually a
• 6:39
  very powerful tool. Right, it has API access to
• 6:41
  deploy code, sort through internal emails and operate
• 6:45
  software across the company. But because employees felt the
• 6:48
  intense pressure of this public leaderboard, they completely
• 6:51
  gamified it. Let me guess, if you tell a
• 6:54
  bunch of brilliant software developers, their performance
• 6:57
  might be judged by a high score on a token counter.
• 7:00
  They're going to build a bot to play the game for them of.
• 7:02
  Course they are. They started burning tokens on
• 7:05
  completely useless, redundant tasks simply to raise their
• 7:09
  numbers on the scoreboard. They were having mesh claws
• 7:11
  summarize massive internal wiki pages.
• 7:15
  They didn't even need to read Wow or rewrite code that was
• 7:18
  already perfectly fine just to churn through compute.
• 7:22
  They even admitted to the Financial Times that the
• 7:24
  pressure to adopt AI was creating perverse incentives.
• 7:28
  Because the metrics weren't measuring actual utility or time
• 7:31
  save, they were just measuring server heat essentially.
• 7:33
  Exactly. Amazon actually had to step in
• 7:36
  and hide the usage scoreboards from individual employees and
• 7:40
  managers. Naturally, though, leadership
• 7:43
  claimed the stats weren't being used for performance reviews.
• 7:45
  Yeah, but clearly the culture had already shifted.
• 7:48
  Look at it this way. It's the classic business
• 7:50
  problem. What you measure is what you
• 7:52
  get, right? A token counter can definitively
• 7:54
  prove that an AI was queried. It can count the compute cycles,
• 7:58
  but it absolutely cannot prove that the developer's output got
• 8:01
  better or faster or more innovative.
• 8:04
  If you reward usage instead of outcomes, the workforce will
• 8:07
  inevitably optimize for the scoreboard.
• 8:10
  It's a huge lesson for any enterprise trying to mandate AI
• 8:13
  right now. Absolutely.
• 8:15
  But here's where it gets really interesting.
• 8:17
  If Amazon represents exactly how not to measure AI's value, we
• 8:22
  have another development today that shows precisely how to use
• 8:24
  AI for high value, deeply outcome driven tasks.
• 8:28
  Yes, the Anthropic update. Anthropic just introduced a
• 8:32
  financial services plug in Marketplace for Claude Code and
• 8:36
  it basically turns this AI into a bespoke localized Wall Street
• 8:40
  analyst sitting right on your hard drive.
• 8:42
  This is a prime example of AI being used for targeted domain
• 8:46
  specific utility rather than just generalized chatting.
• 8:49
  You know Clod Code is Anthropic's terminal based AI.
• 8:52
  It lives in the command line interface of your computer, and
• 8:55
  by adding these new plugins, you're giving it highly
• 8:57
  specialized, secure hooks into live financial data.
• 9:00
  The workflow to set this up is surprisingly straightforward,
• 9:03
  but it feels incredibly powerful.
• 9:05
  It really is. You just open your terminal, run
• 9:08
  Claude code and type slash marketplace.
• 9:11
  That opens the plugin manager. From there, you go to the
• 9:13
  Marketplaces tab, select Ad Marketplace, and drop in a
• 9:16
  specific GitHub link for Claude for Financial Services.
• 9:20
  Simple enough. But how is it actually doing the
• 9:24
  math? Because we know large language
• 9:26
  models are notoriously bad at basic arithmetic, So what makes
• 9:30
  this different? That's the beauty of the plug in
• 9:32
  architecture. Think of it like handing a
• 9:34
  calculator and a filing cabinet key to a brilliant librarian who
• 9:37
  usually only has access to fiction books.
• 9:40
  OK, I like that analogy. The model itself isn't trying to
• 9:43
  predict the math or guess the numbers.
• 9:45
  Instead, the plug insurance give Claude the ability to execute
• 9:48
  specific API calls. Meaning it reaches out to live
• 9:51
  structured financial databases. Exactly.
• 9:54
  It pulls the hard numbers and uses traditional code to run the
• 9:57
  calculations. The standouts in this
• 9:59
  marketplace are tools like Market Researcher, Earnings
• 10:02
  Reviewer, Equity Research, and Financial Analysis.
• 10:05
  So it's basically doing the grunt work of a junior equity
• 10:08
  researcher. It's pulling current data,
• 10:10
  creating heavily sourced market research reports, running
• 10:13
  complex comparisons between publicly traded companies.
• 10:16
  And it can even output the Parse data directly into beautifully
• 10:20
  formatted Excel sheets. That's crazy.
• 10:23
  It's just ready to churn through a 10K earnings report at 3:00 in
• 10:25
  the morning. But we definitely need to pause
• 10:29
  here for a massive reality check because I know people are going
• 10:32
  to hear this and think they've found an infinite money glitch.
• 10:34
  Oh absolutely, this is crucial for anyone listening.
• 10:38
  You must treat this as research support, not financial advice.
• 10:41
  Right. Even with these plugins
• 10:43
  executing code, the AI models still act as the interpreting
• 10:46
  layer and they can and do hallucinate context.
• 10:49
  Like they might pull the right number from an earnings call but
• 10:51
  completely misinterpret the CEO's nuanced tone about future
• 10:55
  guidance. Precisely.
• 10:57
  You should only use these tools to parse public information,
• 11:01
  always demand that Claude cites its specific sources, and
• 11:04
  physically verify every single important number yourself before
• 11:08
  you make any decisions involving real money.
• 11:10
  Yeah, it's a powerful assistant, not a fiduciary.
• 11:12
  Exactly. So don't pick your retirement on
• 11:14
  a terminal command, folks. But here's the logical leap we
• 11:17
  have to make when we talk about turning Claude into a localized
• 11:21
  Wall Street analyst, or giving Gemini the ability to constantly
• 11:24
  see and parse your Android screen.
• 11:27
  We're talking about an exponential leap in the amount
• 11:30
  of data being processed. Exactly.
• 11:32
  You can't just run millions of localized agents without massive
• 11:36
  back end infrastructure, and earthly power grids are already
• 11:39
  groaning under the weight of current data centers.
• 11:42
  Which leads us to a completely wild solution.
• 11:45
  Yeah, so, So what does this all mean?
• 11:48
  We have a mind bending report that Google is exploring a
• 11:51
  rocket launch deal with SpaceX to build orbital data centers.
• 11:55
  Yes, server farms in space. It's it's just bonkers.
• 11:58
  If we connect this to the bigger picture, this is the opening
• 12:02
  frontier of a massive AI infrastructure rivalry and you
• 12:05
  have to look at the history between these two giants.
• 12:07
  Right, because Google isn't just casually calling up SpaceX out
• 12:10
  of the blue. No, they've held a 6.1% stake in
• 12:14
  the company since a 900 million investment back in 2015.
• 12:18
  Google's VP Don Harrison actually sits on the SpaceX
• 12:22
  board. Wow, I didn't realize that.
• 12:24
  Yeah, and furthermore, Google has its own internal moon shot
• 12:27
  called Project Suncatcher. Project Suncatcher.
• 12:29
  Right, which is actively aiming to launch prototype compute
• 12:33
  satellites by 2027, and they're already working with Planet Labs
• 12:37
  on the hardware. And it isn't just Google placing
• 12:40
  these bets. Anthropic just finalized A
• 12:42
  compute deal with SpaceX last week.
• 12:44
  Anthropic explicitly stated they have expressed interest in
• 12:48
  multiple gigawatts of orbital AI compute capacity.
• 12:52
  Gigawatts. Let me just say, gigawatts of
• 12:54
  power in orbit sounds like the plot of a James Bond villain,
• 12:57
  not a tech startup. It really does.
• 12:59
  But for SpaceX, this is brilliant timing.
• 13:01
  They are preparing their pitch for an eventual IPO, and they
• 13:04
  filed for approval to launch up to 1,000,000 satellites,
• 13:08
  1,000,000, positioning themselves not just as an
• 13:12
  Internet provider with Starlink, but as the found foundational
• 13:15
  orbital compute network for the entire AI revolution.
• 13:18
  It's a massive piece of their value proposition to investors.
• 13:22
  OK, I have to push back here because Open AI Sam Altman was
• 13:26
  recently at an event in New Delhi and he flat out called the
• 13:29
  concept of orbital compute ridiculous.
• 13:32
  He did. He argued that it simply won't
• 13:35
  matter at scale this decade. And honestly, I'm with him on
• 13:37
  this one. Space is a vacuum.
• 13:40
  How do you cool a server farm when there's no air to carry the
• 13:43
  heat away? Right?
• 13:44
  Plus, what about the latency of beaming data up and down through
• 13:47
  the atmosphere? Those are the exact physical
• 13:50
  bottlenecks that make Alton's critique completely fair.
• 13:53
  Without convection in a vacuum, you have to use massive heavy
• 13:57
  radiators to dissipate server heat.
• 13:59
  Which are incredibly expensive to launch.
• 14:01
  Exactly, and yes, latency is a huge issue for real time agentic
• 14:05
  tasks. However, the upside is
• 14:08
  continuous unfiltered solar power and 0 earthly real estate
• 14:12
  or power grid constraints. I guess that's true.
• 14:14
  And history offers a very stark cautionary tale here.
• 14:19
  Betting against a combined engineering moon shot from
• 14:22
  Google and Elon Musk has historically been a very bad
• 14:25
  idea. Fairpoint.
• 14:27
  This potential deal is mutually beneficial in a profound way.
• 14:31
  Google gets its Suncatcher prototypes launched without
• 14:34
  having to build its own rocket division, and SpaceX secures A
• 14:38
  marquee trillion dollar client to validate its orbital compute
• 14:42
  pitch before hitting the public markets.
• 14:44
  Wow. Whether the physics scale this
• 14:46
  decade or next, the foundation is being poured right now.
• 14:49
  Well, while the billionaires are busy fighting over server
• 14:51
  cooling systems in orbit, the rest of the tech world is moving
• 14:54
  at absolute warp speed here on the ground.
• 14:56
  Let's run through some rapid fire updates on what else is
• 14:58
  happening in the AI space today, because there's a lot.
• 15:00
  Let's do it. First up, Isomorphic Labs, which
• 15:03
  is Google's drug discovery AI spin off, just landed a massive
• 15:07
  $2.1 billion in funding. That's huge.
• 15:10
  Yeah, and Dimas Asabus, the head of Google Deepline, came out
• 15:13
  with a fantastic quote. He said the number one
• 15:16
  application of AI should be to improve human health.
• 15:19
  That is a really refreshing perspective amidst all the talk
• 15:22
  of token maxing and AD optimization.
• 15:25
  Totally. And the mechanics of why AI is
• 15:28
  so good at this are fascinating. Things like protein folding and
• 15:32
  molecular design are essentially massive predictive puzzles.
• 15:36
  Right. And traditional pharmaceutical
• 15:38
  research takes years of trial and error in a lab just to see
• 15:41
  how molecules interact. Exactly.
• 15:44
  Applying generative models to this process allows them to
• 15:47
  simulate millions of molecular combinations in a computer,
• 15:50
  potentially compressing decades of research into months.
• 15:53
  That's incredible. Now, on the creative side, Crea
• 15:56
  just introduced Crea 2. It's their first proprietary
• 15:59
  image model that is specifically designed for aesthetic range.
• 16:02
  Oh, this is a big deal for creators.
• 16:04
  Yeah, rather than just typing a text prompt and hoping to the
• 16:07
  best, it features mood board based generation and style
• 16:10
  transfer tools. Right, it's a move away from
• 16:12
  standard latent diffusion, where the AI just kind of hallucinates
• 16:15
  an image based on keywords. Which can be so frustrating.
• 16:18
  Very Create 2 gives artists much more structural aesthetic
• 16:22
  control, allowing them to upload a mood board of existing images
• 16:25
  to physically steer the vibe and composition of the output.
• 16:29
  I love that, but as the creative tools get better, the
• 16:33
  vulnerabilities and how we build them are growing.
• 16:36
  A hacker recently planted data stealing malicious code,
• 16:39
  cleverly named Mini Shihulud, inside 42 open source agentic
• 16:44
  NPM packages. A very nerdy Dune reference
• 16:46
  there. Right, but no, for anyone who
• 16:49
  wasn't a developer. I'd admit I was confused by this
• 16:51
  at first. What exactly is an NPM package
• 16:54
  and how does this affect AI tools?
• 16:56
  Well this raises an important question about the security of
• 16:58
  our open source ecosystems. NPM stands for Node Package
• 17:02
  Manager. Think of it like a massive
• 17:04
  Public Library of pre written code snippets.
• 17:06
  Developers use these packages like plumbing so they don't have
• 17:09
  to reinvent the wheel every time they build software.
• 17:11
  Well the hacker poisoned this plumbing because these specific
• 17:14
  packages are foundational building blocks for building AI
• 17:17
  agents. This wide scale attack ended up
• 17:19
  infecting several downstream AI tools that unknowingly use use
• 17:22
  the tainted code. So it's a supply chain attack?
• 17:24
  Exactly. It's a stark reminder that as we
• 17:27
  grant AI agents more autonomy to browse our files, securing the
• 17:33
  open source supply chain becomes absolutely critical.
• 17:36
  It's the classic double edged sword.
• 17:38
  More capabilities mean a much larger attack surface.
• 17:42
  We're also seeing pushback on how these advanced capabilities
• 17:45
  are built in the first place. The meta situation.
• 17:48
  Yeah, Meta employees reportedly organized a protest against the
• 17:50
  company this week. The core issue?
• 17:54
  Meta is allegedly using internal mouse tracking software on his
• 17:57
  own workers to train new AI agents.
• 18:00
  Mechanically, we have to look at why meta would want this data.
• 18:04
  Right to train an AI to use a computer.
• 18:06
  Exactly. To train an AI agent to use a
• 18:08
  computer like a human does to navigate complex UIS, to know
• 18:12
  when to click a drop down menu versus hitting enter.
• 18:15
  You can't just feed it text. You actually have to show it the
• 18:17
  behavior. You have to show the model
• 18:19
  millions of hours of human behavioral data, specifically
• 18:22
  tracking how a cursor moves across the screen.
• 18:24
  But the workers are protesting the ethics of harvesting
• 18:27
  employee behavior to create the very agents that might automate
• 18:31
  those workflows. Which is currently testing the
• 18:33
  boundaries of corporate data collection in real time.
• 18:36
  It's wild. And finally, in our quick hits,
• 18:38
  Rivian just rolled out an AI assistant across its electric
• 18:41
  vehicle. This one is.
• 18:42
  Cool. It really is.
• 18:44
  This isn't just a voice answering trivia questions about
• 18:46
  the weather. It actually controls the car's
• 18:48
  hardware. You trigger it via a button
• 18:51
  right on the steering wheel or by saying hey Rivian and it can
• 18:54
  chain together agentic tasks because it has direct API hooks
• 18:58
  into the car's physical systems it.
• 19:00
  Feels like the car itself is becoming a massive rolling AI
• 19:04
  agent. It really does.
• 19:06
  Which, you know, brings us back around to the core theme of
• 19:09
  today's deep dive. We hear so much about billions
• 19:13
  of dollars in funding, gigawatts of power in orbital data
• 19:17
  centers, enterprise software, API integration.
• 19:20
  Right, all the massive scale stuff.
• 19:21
  But the most profound impacts of this technology are deeply
• 19:24
  personal. They happen at the individual
• 19:27
  level, in the hands of regular people.
• 19:29
  Exactly like how are you using this tech?
• 19:31
  Every day we see massive corporate use cases, but the
• 19:34
  community applications are where the real heart is.
• 19:36
  Today I want to share a story from a listener named Corey S
• 19:39
  down in Fort Lauderdale, FL and honestly, this one really moved
• 19:44
  me. It's a powerful story.
• 19:46
  Six months ago, Corey's little brother passed away
• 19:48
  unexpectedly. The night before the funeral,
• 19:51
  his family had a realization. There were hundreds of people
• 19:55
  coming to the service, and those people likely held a treasure
• 19:59
  trove of photos on their personal phones.
• 20:01
  Snapshots of his brother's life from childhood to adulthood that
• 20:04
  the family had never even seen. Right.
• 20:07
  It's a heartbreaking situation, but also a beautiful realization
• 20:10
  of how much of our shared lives are captured in the fragmented
• 20:14
  digital memories of our extended communities.
• 20:17
  Yeah, so much of our history is just sitting on other people's
• 20:19
  camera. Roll exactly.
• 20:21
  So Corey had this realization the night before, on the morning
• 20:24
  of the service. He used AI to quickly vibe, code
• 20:28
  and deploy a simple, functional website.
• 20:30
  And for anyone listening who hasn't tried this, vibe coding
• 20:33
  is essentially skipping the traditional programming
• 20:35
  languages. Right.
• 20:36
  Instead of writing HTML or setting up server databases line
• 20:40
  by line, Corey just described the emotional utility he needed
• 20:43
  in natural language to an AI coding assistant.
• 20:46
  He told the AI he needed a secure site where guests could
• 20:49
  scan AQR code and easily upload photos from mobile browsers.
• 20:53
  He essentially used AI to instantly bridge a technical gap
• 20:57
  during a moment of profound grief.
• 20:59
  Yeah, he didn't have to spend weeks writing back end code.
• 21:02
  The AI Stafford did the entire site, database and UI in
• 21:06
  minutes, enabling him to manifest a solution exactly when
• 21:10
  it was needed. And it worked beautifully.
• 21:12
  As people arrived at the funeral, they scan the code and
• 21:15
  a live wall of memories began to grow on screens at the service.
• 21:19
  They gave everyone a chance to reflect on his brother's life
• 21:21
  while they waited. And the best part?
• 21:23
  After the service was over, Core was able to download every
• 21:26
  single one of those photos, pieces of his brother's life he
• 21:29
  wouldn't have had otherwise, and give them to his mother.
• 21:31
  It's a deeply touching reminder. Beyond the token scoreboards at
• 21:34
  Amazon, beyond the compute power and the rocket launches, this
• 21:38
  technology, at its very core, is simply a tool.
• 21:41
  And when applied with empathy, it's a tool for human
• 21:44
  connection, for preserving memory, and for bringing people
• 21:48
  together in the most vulnerable moments.
• 21:50
  I couldn't agree more. It completely recenters why any
• 21:54
  of us care about this technology in the first place.
• 21:55
  It really does. And you know, it leaves us with
• 21:57
  something vital to consider. What's that?
• 22:00
  If an AI can be used to seamlessly compile the digital
• 22:03
  fragments of a life into a beautiful memorial, what does
• 22:06
  that mean for our own digital footprints?
• 22:09
  As agents become OS level features like we saw with
• 22:12
  Android today, analyzing our screens and our habits, they
• 22:16
  won't just know what tasks we do.
• 22:18
  They might understand the narrative of who we are better
• 22:20
  than we do ourselves. That is a wild thought to leave
• 22:23
  on. Thank you so much for joining us
• 22:24
  on this deep dive. Make sure you subscribe to stay
• 22:27
  updated on all the incredible and sometimes crazy ways this
• 22:30
  tech is changing our world. If you enjoy the conversation,
• 22:33
  please drop us a rating. Five stars.
• 22:34
  Nailed it. Hopefully we'll see you back for
• 22:36
  tomorrow's deep dive.