Latest / Today’s AI News / Sundar Pichai Says Today's AI Will Look Like a Flip Phone in 3 Years, Codex Gets Locked Computer Use

Transcript

Sundar Pichai Says Today's AI Will Look Like a Flip Phone in 3 Years, Codex Gets Locked Computer Use · Today’s AI News

• 0:00
  Imagine holding a chunky plastic flip phone from the early 2000s.
• 0:05
  I mean in just three years. That is exactly how primitive
• 0:09
  today's cutting edge AI is going to look to you.
• 0:12
  It's a It's a pretty wild thought.
• 0:14
  Right. Well, welcome back to today's AI
• 0:16
  news brought to you by 9X Productions.
• 0:18
  We are on the precipice of a massive evolution from AI that
• 0:22
  just chats with us to, you know, a genic AI that actively does
• 0:27
  things for us in the background. Today, we're unpacking exactly
• 0:31
  how that leap is happening right now.
• 0:33
  We are looking at a stack of today's latest AI news to
• 0:36
  understand this massive shift. And we really have a lot to
• 0:39
  cover. We do.
• 0:40
  We're covering Google's futuristic vision straight from
• 0:42
  the top. Open AI is brand new tools that
• 0:44
  are dropping onto your desktop today.
• 0:46
  The profound socioeconomic ripples hitting California and
• 0:51
  how everyday people are already using this technology as a
• 0:54
  literal personal coach. It's a huge day for news.
• 0:57
  So if you are listening to this on your commute or at your desk,
• 1:00
  prepare to look at your devices a little differently by the end
• 1:02
  of this deep dive. Let's start with the big
• 1:04
  picture, right? Google IO 2026.
• 1:07
  Yeah, Google IO just happened and there was an exclusive
• 1:10
  interview with CEO Sundar Pichai.
• 1:14
  He was the one who actually made that bold claim about the flip
• 1:17
  phone. Right.
• 1:17
  And coming from the CEO of Google, I mean, that isn't just
• 1:20
  idle speculation. That is a well, it's a strategic
• 1:23
  road map. Yeah, when Pichai says that
• 1:25
  today's technology will look primitive in 36 months, he is
• 1:28
  signaling A fundamental rewiring of how their entire ecosystem
• 1:32
  operates. The keyword he used over and
• 1:35
  over during that interview was seamless.
• 1:37
  Seamless, right? Yeah, for the everyday user, the
• 1:40
  shift to their Gemini model isn't about having like a
• 1:44
  smarter chat, but window open on your browser.
• 1:47
  Right, it's deeper than that. Exactly.
• 1:49
  It is about seamless integration into your daily life, with
• 1:52
  agents handling background tasks across your phone, your laptop
• 1:55
  and your home constantly running 24/7.
• 1:58
  And it's not just for everyday consumers setting reminders or,
• 2:01
  you know, summarizing emails. The way Pichai described the
• 2:04
  future for developers was man, it was wild.
• 2:06
  Oh, the agentic coding part. Yes, he stated the metric for
• 2:11
  success is shifting completely away from AI written code.
• 2:15
  We are apparently already past the era of having a really smart
• 2:19
  auto complete that finishes your sentence for you.
• 2:22
  Right, which was the standard just yesterday basically.
• 2:24
  Exactly. The new metric is agentic
• 2:26
  coding. So the way I picture this, it's
• 2:29
  like hiring a tireless junior developer who works while you
• 2:33
  sleep. It's a really good analogy.
• 2:35
  Right, you don't just ask it to write a single function, you
• 2:38
  give it a massive complex software project and it just
• 2:41
  grinds away at it for hours handling all the long running
• 2:44
  tasks. That is a highly accurate way to
• 2:47
  visualize it. If we connect this to the
• 2:49
  broader software development life cycle.
• 2:51
  Think about how code is, you know, traditionally made right.
• 2:54
  You write the code, you test it, it inevitably fails.
• 2:57
  You read the error logs, you debug, you rewrite.
• 3:00
  The endless loop. The endless loop.
• 3:01
  Exactly. Agentic coding means the AI is
• 3:05
  running that entire loop autonomously.
• 3:07
  It is executing the code, reading its own errors and
• 3:10
  fixing them without you ever clicking a button.
• 3:12
  But this shift raises a massive question about the human
• 3:15
  element, doesn't it? It does, and Pichai was very
• 3:17
  keen to address this, particularly regarding platforms
• 3:20
  like YouTube. Yeah, he explicitly noted that
• 3:23
  YouTube will remain creator first, right?
• 3:26
  It's going to prioritize that authentic human to human
• 3:29
  connection, even though these incredible new models like Omni
• 3:34
  are there to help creators express themselves better.
• 3:36
  Yeah, that human core stays. But hearing all that, I mean, if
• 3:41
  agents are writing the code, fixing the bugs, and doing all
• 3:43
  the heavy lifting 24/7, it sounds like my job entirely
• 3:47
  shifts. How so?
• 3:48
  Well, I am no longer a creator in the trenches, right?
• 3:51
  I am essentially a manager of digital labor.
• 3:54
  That framing is spot on. You transition from creation to
• 3:56
  curation, from writing the boilerplate, which is all that
• 4:00
  standard repetitive code you have to type out before doing
• 4:03
  the actual fun stuff, to architecting the entire system.
• 4:07
  So I'm the architect now. Exactly.
• 4:08
  You aren't in the weeds writing every line.
• 4:10
  You are setting the parameters, defining the goals, and
• 4:13
  evaluating the final output. The human provides the taste,
• 4:16
  the strategic direction and the ethical boundaries.
• 4:19
  While the agents provide the brute force execution.
• 4:23
  Precisely. That three-year road map is
• 4:25
  fascinating, but if we look at what's dropping onto your
• 4:28
  desktop today, Open AI is already pushing us into that
• 4:32
  manager. Role.
• 4:32
  Oh definitely, They just released a massive wave of
• 4:35
  upgrades to codecs. And looking at the broader
• 4:37
  industry, this is clearly a strategic move, right?
• 4:40
  To close the gap with Anthropic and maintain their lead over X
• 4:46
  AI and Google. It is, yeah.
• 4:47
  But the specific features, man, they are intense.
• 4:51
  They are actively bridging the gap between the AI model
• 4:54
  residing in the client and the actual local operating system on
• 4:58
  your physical device. We should break down how they
• 5:01
  are doing this. Let's start with appshots.
• 5:03
  Yeah, Appshots. So if you're working on a Mac,
• 5:05
  you just hit command a command on your keyboard and codecs
• 5:07
  instantly attaches whatever application window you currently
• 5:10
  have open directly into the chat thread.
• 5:12
  It's so fast. Right.
• 5:13
  It grabs the screenshot, the underlying text, the content,
• 5:15
  everything. It's instant context for the AI.
• 5:18
  What is mechanically fascinating here is that it is not just
• 5:21
  taking a static picture like a standard screenshot.
• 5:24
  Oh, really? No, by capturing the underlying
• 5:27
  text and the content state of the application.
• 5:29
  Codecs actually understands the semantic structure of what you
• 5:33
  are working on. OK, what does that mean in
• 5:36
  practice? It means it knows exactly where
• 5:38
  the buttons are, it knows what code is currently highlighted in
• 5:41
  your IDE, which is your coding software, and it knows what
• 5:45
  specific cell in a spreadsheet is active.
• 5:48
  That is wild. It is, but capturing that
• 5:50
  context is just the input side. The execution side is where it
• 5:54
  becomes highly advanced. Right, you're talking about goal
• 5:56
  mode. Exactly.
• 5:58
  This is available right now in the Codex app, the IDE
• 6:01
  extension, and the command line. You literally just set a goal
• 6:04
  for the AI and the agent works toward it for hours or even
• 6:08
  days. And this pairs with another new
• 6:10
  feature called Advanced Annotation Mode.
• 6:12
  Right where you just describe the changes you want on a web
• 6:15
  page and it instantly previews the results without you having
• 6:18
  to touch a single line of code. It's basically a real time
• 6:22
  preview of your thoughts. It is, but here's where it gets
• 6:25
  slightly terrifying to me. They introduced something called
• 6:27
  locked computer use. A massive technical leap,
• 6:31
  without question. I mean, sure.
• 6:33
  So you can trigger codecs from a second device like your phone,
• 6:37
  and it can use your desktop applications even when your Mac
• 6:41
  is completely locked and the screen is totally black.
• 6:43
  Yep, it's running in the dark, basically.
• 6:45
  I mean, I love the idea of my computer working for you while
• 6:48
  I'm out of dinner, but letting an agent go rogue on my desktop
• 6:52
  for 48 hours while I'm not looking it?
• 6:53
  Requires a lot of trust. It's like hiring a house sitter,
• 6:57
  giving them your wallet, and trusting them to not only water
• 7:00
  the plants, but also successfully negotiate your
• 7:02
  mortgage rate while you're out. That's that's a very vivid way
• 7:06
  to put it. But seriously, what are the
• 7:08
  actual guardrails here? If an agent is running for days
• 7:11
  on end, couldn't it just accidentally delete a database
• 7:14
  or e-mail my boss something crazy?
• 7:17
  It is a vital concern and it really highlights the friction
• 7:20
  of this transition period we are in.
• 7:23
  The mechanics of how this locked computer use functions is
• 7:26
  actually quite clever. How does it work?
• 7:29
  The agent is interacting with the operating system's
• 7:31
  accessibility APIs. These are the same internal
• 7:35
  protocols originally designed to help visually impaired users
• 7:38
  navigate a computer. Oh wow, I wouldn't have guessed
• 7:41
  that. Yeah, it allows the software to
• 7:43
  essentially see the screen elements and simulate human
• 7:48
  clicks and keystrokes. Well, what about the?
• 7:49
  Guardrails. Right now the primary safety net
• 7:52
  is the extreme specificity of your initial prompt and the
• 7:56
  sandboxing of the agents permissions.
• 7:58
  So it's basically on you to not give it the keys to the castle.
• 8:01
  Exactly, but you are touching on the central anxiety of agentic
• 8:05
  AI. We are moving from a
• 8:07
  deterministic system, where the computer only does exactly what
• 8:10
  you explicitly type to a probabilistic system.
• 8:13
  Where it's guessing. Where the agent is making
• 8:15
  localized, unscripted decisions to achieve a broader goal.
• 8:18
  OK, so we have these incredibly powerful agents running on our
• 8:21
  local machines, but a desktop is a walled garden.
• 8:26
  It is. If they can't navigate the
• 8:27
  messy, unpredictable Internet to gather data or interact with
• 8:30
  other services, their utilities actually pretty capped.
• 8:34
  How do they break out of that local environment?
• 8:36
  Historically, if you wanted an AI to interact with a specific
• 8:39
  website, say checking ESPN for live sports scores or scanning
• 8:44
  Google Flights, you had to manually connect the AI to that
• 8:48
  website's API. It's application programming
• 8:52
  interface, and that's a massive bottleneck, right?
• 8:54
  Because APIs are constantly changing, or they're poorly
• 8:57
  documented, or the website just decides to shut them down.
• 9:00
  Or put them behind a massive paywall.
• 9:02
  Exactly. So they each it just hits a
• 9:03
  brick wall. Precisely.
• 9:04
  They are incredibly brittle, but developers are using a new
• 9:08
  method to solve this with a tool called printing press.
• 9:11
  Printing press OK. It allows an agent to generate
• 9:13
  its own native command line interface or CLI from any
• 9:17
  website. Wait, hold on, let me make sure
• 9:19
  I'm wrapping my head around this because it sounds like science
• 9:21
  fiction. Walk me through it.
• 9:22
  The AI is actually installing a software binary like an
• 9:27
  executable file onto my machine to build its own custom tool to
• 9:31
  scrape a website. Am I getting that right?
• 9:33
  You are because instead of a human developer hard coding the
• 9:36
  integration, the agent investigates the target
• 9:39
  environment itself. How does that even work?
• 9:42
  The process works like this. First you install a basic
• 9:45
  starter kit. Then you tell your agent whether
• 9:48
  you are using codecs, Claude code or Hermes to inspect the
• 9:52
  tools and run a safe read only command to see what works on the
• 9:55
  target website. Then you actually ask the agent
• 9:58
  to install the printing press binary itself.
• 10:01
  OK, let me try an analogy here to see if I'm tracking.
• 10:03
  Let's hear it. It's like instead of giving your
• 10:06
  AI agent a map of a fortress and a specific key that might not
• 10:10
  even work tomorrow, you're handing it a 3D printer, right?
• 10:14
  So it can just walk up to whatever door it encounters on
• 10:16
  the Internet, analyze the lock, and manufacture its own lock
• 10:19
  picks on the spot. That is a remarkable way to put
• 10:22
  it. Once printing press is
• 10:23
  installed, the agent looks at a site with public data and
• 10:27
  dynamically generates and tests its own commands to extract what
• 10:30
  it needs. It just figures it out.
• 10:32
  Completely. If an agent struggles with the
• 10:34
  standard API integration, you simply handed an API key and
• 10:38
  tell it to use printing press to build its own customized
• 10:40
  solution. It is entirely self tooling.
• 10:43
  The AI encounters an obstacle, realizes it doesn't have the
• 10:47
  right wrench, so it just fabricates the wrench.
• 10:49
  Exactly, but there's a reason they emphasize read only
• 10:52
  commands in the starter kits. That goes back to the guardrails
• 10:55
  we were talking about with the locked computers.
• 10:56
  Oh, absolute. You want the agent to read the
• 10:59
  flight data, not accidentally book 40 non refundable tickets
• 11:03
  to Tokyo while trying to figure out how the website works.
• 11:06
  Exactly. That read only limits the blast
• 11:09
  radius while the agent is experimenting, but consider the
• 11:12
  massive acceleration this provides.
• 11:14
  Oh for sure. When agents can build their own
• 11:16
  tools to navigate the web, the bottleneck of human developer
• 11:20
  time is completely removed from the equation.
• 11:22
  And when agents can bypass human bottlenecks like that, the
• 11:26
  acceleration is incredible. But that kind of hyper
• 11:28
  efficiency creates immediate economic triction.
• 11:31
  It does. It has real world consequences.
• 11:33
  Which brings us to the human cost and what is happening right
• 11:37
  now in California. We have to look at the tangible
• 11:40
  impact of this technology moving from the lab to the corporate
• 11:43
  world. California Governor Gavin Newsom
• 11:46
  just signed an executive order specifically aimed at protecting
• 11:50
  workers from AI driven job losses.
• 11:53
  And the timing of this is very stark.
• 11:55
  Very just a day before this order, Mehta laid off 8000
• 12:00
  employees and they explicitly stated this was to offset their
• 12:04
  massive investments into AI infrastructure.
• 12:07
  Yeah, the context here is grim. Over 70,000 tech jobs have
• 12:11
  already vanished in 2026. That's a huge number.
• 12:14
  It is, and California's Ground Zero for this shift.
• 12:18
  They're home to 33 of the top 50 AI companies in the world.
• 12:22
  The concentration of both the incredible innovation and the
• 12:25
  severe workforce disruption in one geographic area is just
• 12:29
  unprecedented. What the state is doing here is
• 12:31
  essentially trying to build a regulatory parachute while the
• 12:34
  economic claim is already in freefall.
• 12:36
  Right. And they've established a highly
• 12:37
  aggressive timeline. Very aggressive.
• 12:39
  Within 90 days, the state is launching a dashboard just to
• 12:42
  track a is ongoing impact on jobs.
• 12:44
  Just to see where the bleeding is.
• 12:46
  Exactly. Within 180 days they are
• 12:49
  pitching updates to the Warren Act.
• 12:51
  For anyone unfamiliar, that's the federal law requiring
• 12:54
  companies to give advance notice of mass layoffs so workers get
• 12:57
  faster alerts. Right.
• 12:59
  And by October 15th, they were reviewing how unions negotiate
• 13:03
  AI adoption, updating workforce training, and exploring ways to
• 13:07
  direct AI revenue toward public benefits.
• 13:10
  And if you look at the specific policies they are exploring, it
• 13:13
  reveals A fundamental rethinking of our social contract.
• 13:17
  Like what? Well, they are examining
• 13:19
  severance standards, stock compensation adjustments and
• 13:22
  worker ownership models. But the most radical concept
• 13:25
  currently on the table is Universal Basic Capital.
• 13:28
  I have to stop you there because that caught my eye too, and I
• 13:30
  want to make sure we parse the difference.
• 13:31
  We've all heard of Universal Basic Income Ubi, which is the
• 13:35
  idea of giving to everyone a flat monthly cash stipend.
• 13:38
  But what is universal basic capital?
• 13:41
  How does a citizen actually get capital from an AI model?
• 13:45
  It is a critical distinction. Universal Basic income is about
• 13:48
  distributing cash. Universal Basic Capital is about
• 13:51
  distributing ownership. OK, ownership of what?
• 13:54
  The underlying economic theory is that in a highly agentic
• 13:58
  economy, human wage labor loses its market value because agents
• 14:02
  can do the work cheaper and faster.
• 14:03
  Right. Obviously, however.
• 14:05
  Capital, specifically the ownership of the compute the
• 14:08
  data centers, the models skyrockets in value.
• 14:12
  Universal basic Capital suggests that citizens should receive an
• 14:15
  actual equity stake in the economic output of these
• 14:19
  automated systems. How would that even work in
• 14:21
  practice? Mechanics for this could look
• 14:23
  like a sovereign wealth fund funded by a tax on compute usage
• 14:27
  where citizens receive dividends, or literally granting
• 14:31
  citizens shares and state licensed AI infrastructure.
• 14:35
  So instead of just handing people a check to survive, you
• 14:37
  make them shareholders in the automated economy.
• 14:39
  Exactly. Whether that is an actual
• 14:41
  scalable solution for a world of autonomous AI or just a
• 14:45
  temporary Band-Aid of probably the defining economic question
• 14:48
  of this decade. I would agree with that
• 14:50
  completely. And if you want proof that
• 14:51
  capital, specifically the ownership of compute, is
• 14:53
  skyrocketing, just look at the physical infrastructure being
• 14:57
  built right now. It's math.
• 14:58
  The sheer amount of money and hardware being pumped into
• 15:01
  making these agents smarter is staggering.
• 15:03
  Let's run through some of the massive deals happening in AI
• 15:06
  today. The physical infrastructure is
• 15:08
  the invisible engine driving all of this software innovation.
• 15:11
  Without the silicon and the power, none of these agents
• 15:14
  function. Case point, SpaceX recently
• 15:16
  filed an IPO prospectus and hidden in that document it was
• 15:20
  the revelation that Anthropic is paying SpaceX $1.25 billion.
• 15:25
  And that's not per year. No, not per year, per month.
• 15:28
  Every single month through 2029 for access to compute capacity
• 15:32
  on their Colossus and Colossus to Zek and supercomputers.
• 15:35
  It's hard to even wrap your head around that number.
• 15:37
  A billion and a quarter dollars every 30 days.
• 15:40
  Why does an AI company need that level of continuous
• 15:43
  infrastructure? Well, when Sundar Pichai talked
• 15:45
  about the flip phones of today, this is what the future actually
• 15:48
  costs. To build a gentic AI requires
• 15:51
  continuous, persistent compute. Think about your own
• 15:54
  interactions with AI. Right now a chat bot only uses
• 15:57
  processing power for a few seconds when you ask it a
• 16:00
  specific question. And then it stops.
• 16:02
  Exactly. But an agent working for two
• 16:04
  days straight on a locked computer is churning through
• 16:07
  matrix multiplications, the complex mathematical
• 16:10
  calculations neural networks use to process data constantly.
• 16:13
  It never sleeps. It never sleeps.
• 16:15
  Anthropic is paying for planetary scale physical
• 16:18
  infrastructure. Compute is the new oil.
• 16:21
  It really is. They are also reportedly in
• 16:23
  talks to use Microsoft's new Maya purpose built AI chips,
• 16:28
  adding to the deals they already have for Google's TP US and
• 16:31
  Amazon's Trinium chips. So they are heavily diversifying
• 16:35
  their silicon supply chain so they are never bottlenecked.
• 16:37
  They are buying up every drop of compute they can find.
• 16:40
  And Speaking of Microsoft, they just announced a $1 billion
• 16:44
  partnership with EY. They are pairing Microsoft
• 16:47
  Engineers in AI with Ey's 400,000 consultants to
• 16:51
  accelerate AI projects for enterprise clients.
• 16:54
  That signals the deployment phase.
• 16:56
  What do you mean? You have the models, you have
• 16:58
  the immense compute power, but deploying an agent into a messy
• 17:02
  legacy corporate database is incredibly difficult.
• 17:05
  You need an army of human consultants to integrate these
• 17:08
  agents into the real world. On the corporate drama side,
• 17:11
  agentic AI startup Manus is exploring a $1 billion raise.
• 17:17
  And the geopolitical context here is wild.
• 17:19
  It is. They're doing this to comply
• 17:21
  with an order from China forcing them to unwind their acquisition
• 17:25
  by Meta, which was previously worth over 2 billion.
• 17:28
  The geopolitical tug of war over these agentic startups is very
• 17:31
  real. It really highlights that.
• 17:33
  Yeah, the strategic value of autonomous agents is recognized
• 17:36
  at the nation state level now. It is a core component of the
• 17:40
  broader technological decoupling between the US and China.
• 17:43
  To round out the ecosystem, we saw Figure AI Co founder Brett
• 17:47
  Adcock's new startup Hark just raised $700 million at a $6
• 17:52
  billion valuation. Massive numbers again.
• 17:55
  Yeah, they are building personal intelligence, developing native
• 17:57
  models, software and hardware all in house.
• 18:01
  Plus Google announced a new next Gen.
• 18:02
  A genetic development platform called Anti Gravity.
• 18:05
  There's a new open source alternative to open clock called
• 18:08
  Nano Claw that's heavily focused on enterprise security.
• 18:12
  Essential stuff. And Harvey's agents are now
• 18:14
  being used for contract intelligence, basically
• 18:17
  deploying AI to take the first pass on inbound legal contracts.
• 18:22
  What all these rapid fire developments show is a complete
• 18:25
  multi layered ecosystem materializing right from the
• 18:29
  silicon supply chain with Microsoft's Maya up to the
• 18:32
  developer platforms like Antery Gravity, to the secure
• 18:36
  enterprise application layer with contract intelligence all
• 18:40
  the way to the hardware layer with Hark.
• 18:41
  It's all happening simultaneously.
• 18:43
  Exactly. The agentic infrastructure stack
• 18:46
  is being built from the ground up all at once.
• 18:48
  We've been talking about billion dollar infrastructure, sweeping
• 18:52
  government policies, and rogue agents coding unlocked Macbooks.
• 18:56
  I'm going to pull this back down to a human scale.
• 18:58
  I think that's a great idea. Because at the end of the day,
• 19:00
  you might be wondering how a real person is actually applying
• 19:03
  this in their daily life right now.
• 19:04
  Right, moving past the corporate theory.
• 19:06
  Exactly. We have a great community
• 19:08
  workflow today from a reader named Loic R out in London.
• 19:11
  This is a perfect example of what all this massive technology
• 19:14
  is actually capable of doing for an individual.
• 19:17
  So Loic is a former sprinter, but years ago he tore the ACL in
• 19:21
  both of his knees playing football.
• 19:23
  He has 0 long distance running experience, but in one month
• 19:28
  he's taking part in a grueling 5 hour multi sport adventure race.
• 19:32
  Wow, that is intense. Right, so he is staring down a
• 19:36
  massive physical challenge with a very tricky medical history.
• 19:41
  He didn't just ask ChatGPT for a generic Couch to 5K running
• 19:45
  plan, right? Which wouldn't work for him
• 19:46
  anyway. Exactly.
• 19:47
  He had to build a custom four week training program highly
• 19:51
  tailored to his specific bilateral ACL entry history and
• 19:54
  his explosive strength as a former sprinter.
• 19:57
  The extreme personalization here is key.
• 19:59
  A static plan downloaded off the Internet cannot account for
• 20:02
  those specific compounding physical variables.
• 20:04
  Not at all. But what he does next is where
• 20:06
  the agentic continuous feedback loop really shines.
• 20:09
  What does he do? After every single run, Lowick
• 20:12
  uses voice notes on his phone to log his performance.
• 20:15
  He just talks to the AI, explaining his times, but more
• 20:18
  importantly, how his body actually felt.
• 20:20
  Any stiffness in the knees, unexpected fatigue, whatever.
• 20:23
  That qualitative data is crucial.
• 20:25
  It is. The AI then takes that voice
• 20:27
  data and dynamically adjust the rest of his four week program on
• 20:31
  the fly. It constantly calculates how to
• 20:34
  maximize his fitness in the dwindling time left before the
• 20:38
  race, while actively minimizing his specific risk of re injury.
• 20:43
  That is incredible. Loic literally described it as
• 20:46
  having a personalized coach in my pocket.
• 20:48
  If we circle all the way back to the start of our deep dive, this
• 20:51
  is the exact manifestation of Sundar Pichai's vision.
• 20:54
  There really is. Pichai talked about seamless
• 20:56
  integration into daily life. He talked about technology being
• 21:00
  fundamentally human centric. Right.
• 21:02
  Loic isn't writing code. He isn't managing.
• 21:05
  AP is. He's using natural language,
• 21:07
  just his voice, to interact with an intelligent system that
• 21:10
  retains state, understands deep context, and continuously
• 21:14
  optimizes toward a specific goal.
• 21:16
  Yeah, the complexity of the billion dollar models in the
• 21:18
  server farms is completely abstracted away, leaving only
• 21:21
  the pure utility. Exactly.
• 21:23
  It's the ultimate counter narrative to the fear of the
• 21:25
  machine replacing us. It's technology enabling a human
• 21:29
  to push their own physical limits safely.
• 21:32
  But as we wrap up today's deep dive, it does leave us with a
• 21:35
  fascinating tension to consider. What's?
• 21:37
  That. Well, we are handing over the
• 21:39
  keys to AI agents that can work for days on our locked
• 21:43
  computers, navigate the messy Internet autonomously, and
• 21:47
  generate their own tools to solve problems.
• 21:50
  In a world where digital labor is infinite and virtually free,
• 21:55
  will the ultimate luxury of the future be doing things manually?
• 21:59
  Oh wow, that's a thought. Will human made, whether it's a
• 22:03
  piece of code, a handmade table, or piece of art, become the new
• 22:07
  premium? It is a profound thought.
• 22:10
  When the execution of any task becomes entirely frictionless,
• 22:14
  the friction itself might become the primary source of value.
• 22:17
  Exactly something for you to Mull over as you watch your
• 22:20
  computer screen work by itself tomorrow morning.
• 22:22
  Yeah, really. Thank you for joining us today
• 22:24
  on the Deep Dive. Make sure you subscribe to stay
• 22:26
  updated with today's AI news. And if you're enjoying these
• 22:29
  deep dives, please take a second to rate the show 5 stars.
• 22:32
  It really helps us out. It really does.
• 22:34
  We will see you back here for tomorrow's deep dive.