Latest / Connecting the Dots / GitHub Security Breach, Gemini 3.5 Flash Launch, And AI Model Pricing Shifts

Transcript

GitHub Security Breach, Gemini 3.5 Flash Launch, And AI Model Pricing Shifts · Connecting the Dots

• 0:00
  Today's tech news moves fast, but understanding
• 0:03
  it shouldn't be hard. That's why Snarfle Solutions
• 0:06
  Group brings you Connecting the Dots, where we
• 0:09
  break down what's happening in tech and why it
• 0:11
  matters to you. Now, let's dive in with your
• 0:14
  hosts. Imagine finding out that the core software
• 0:19
  running like... 90 % of the Fortune 100 was just
• 0:23
  quietly breached. Right, and the craziest part
• 0:25
  is it wasn't even the work of some, you know,
• 0:28
  ultra -sophisticated hacking syndicate. Exactly.
• 0:30
  No million -dollar firewall is being dispatched.
• 0:32
  It literally happened just because someone somewhere
• 0:36
  wanted a slightly more convenient browser tool
• 0:39
  and just clicked install. Yeah, just trying to
• 0:43
  save a few seconds. Today, we are unpacking the
• 0:46
  true cost of our digital convenience, plus the
• 0:49
  arrival of Google's new era of AI interns. And,
• 0:52
  of course, the exhausting reality of what people
• 0:54
  are calling the zombie internet. Welcome to Connecting
• 0:56
  the Dots, broadcasting straight from Snarfle
• 0:58
  Solutions Group in Sacramento, California. I'm
• 1:01
  Alex. And I'm Morgan. And today is Wednesday,
• 1:03
  May 20, 2026. We've got a really packed deep
• 1:07
  dive into the source material today. But before
• 1:10
  we get into the tech, let's get you caught up
• 1:12
  with your morning star updates. Always a good
• 1:14
  place to start. Looking out the window here in
• 1:16
  Sacramento, it is a gorgeous sunny 93 degrees
• 1:18
  with just a really light like one to five mile
• 1:21
  per hour breeze. Nice. Yeah, but nationally,
• 1:25
  the map looks quite a bit more turbulent. There
• 1:28
  are actually 400 active weather alerts across
• 1:31
  the US right now. Wow, 400. Yeah, 400. So depending
• 1:35
  on where you are tuning in from today, please
• 1:37
  stay safe and, you know, keep an eye on your
• 1:40
  local advisories. Definitely. And looking over
• 1:43
  the financial markets, it was a pretty bumpy
• 1:45
  ride to close out Tuesday, the Dow closed down
• 1:48
  about 332 points. Ouch. Yeah. And the S &P 500
• 1:52
  also took a slight dip, dropping by, I think,
• 1:54
  22 points. Right. But hey, if you happen to be
• 1:57
  holding crypto, you're probably having a much
• 1:59
  better morning. Oh, for sure. Bitcoin is having
• 2:01
  a really solid day. It's up over $580 to close
• 2:05
  just above $77 ,340. Not bad at all. And moving
• 2:10
  over to sports, it was kind of a mixed bag for
• 2:12
  our local Northern California fans. The Oakland
• 2:14
  A's absolutely crushed the Angels 14 -6. Nice
• 2:17
  win for them. Yeah. But on the other side of
• 2:19
  the bay, the Giants dropped their game to the
• 2:21
  Diamondbacks 3 -5. Yeah, tough break. And for
• 2:24
  the Hoops fans, the Kings are actually resting
• 2:26
  today, so no game on the schedule. Well, which
• 2:28
  gives us plenty of time to focus on our mission
• 2:30
  here at Connecting the Dots. Because every day,
• 2:33
  our goal is to break down complex tech news and
• 2:36
  explore its everyday impact on, well... You working
• 2:39
  professionals, families, completely free of the
• 2:43
  dense jargon. Let's unpack this. So I want you
• 3:05
  to think about the last time you were working
• 3:07
  on your computer. You probably needed like a
• 3:10
  quick tool to format a document or maybe block
• 3:13
  some annoying pop -up ads, right? Or like manage
• 3:16
  a PDF or something. Exactly. You found a little
• 3:18
  browser extension or an app plug -in and you
• 3:20
  just absentmindedly clicked install. Guilty as
• 3:23
  charged. Right. I mean, you didn't read the terms
• 3:25
  of service. You certainly didn't like audit the
• 3:27
  developer's background. Nobody has time for that.
• 3:30
  No, you just wanted the immediate convenience
• 3:31
  to finish whatever you were doing. We all do
• 3:33
  it. It's basically the ultimate digital temptation.
• 3:36
  We willingly trade a little bit of security for
• 3:40
  a lot of immediate utility. And most of the time,
• 3:43
  nothing bad happens, which just reinforces that
• 3:45
  habit. Well, that exact human impulse just caused
• 3:48
  a massive, massive headache for GitHub. Yeah,
• 3:51
  it really did. And for the listener who might
• 3:53
  not be a developer, GitHub is basically the foundational
• 3:56
  library for modern software. And they just confirmed
• 3:59
  that roughly 3 ,800 of their internal repositories
• 4:03
  were breached. It's just a staggering number.
• 4:06
  And it happened because a single employee installed
• 4:09
  a malicious, trojanized extension for VS Code.
• 4:12
  Which is wild. And let's break down the mechanics
• 4:15
  of how that actually works because it really
• 4:17
  is fascinating. Please do. So VS Code Visual
• 4:19
  Studio Code is this massively popular text editor
• 4:23
  that programmers use to write software. And just
• 4:26
  like the regular web browser you use every day,
• 4:28
  it has this huge marketplace where developers
• 4:31
  can download extensions. Right, to add new features
• 4:33
  and stuff. Exactly. Maybe they want a tool that
• 4:36
  color -codes their text or, you know, an extension
• 4:39
  that autocompletes repetitive typing to make
• 4:42
  their life easier. Makes sense. But in this specific
• 4:44
  case, the extension that the employee downloaded
• 4:47
  was poisoned. It had a Trojan horse in it. See,
• 4:50
  I look at this and my mind immediately goes to,
• 4:53
  like, a physical analogy. Oh, I love your analogies.
• 4:56
  Let's hear it. Okay. So imagine a high -security
• 4:59
  bank vault. You've got the biometric scanners.
• 5:02
  armed guards, foot thick steel doors, you know,
• 5:06
  the whole nine yards. Totally impenetrable. Right.
• 5:08
  And this vault doesn't get compromised because
• 5:10
  someone blew the doors off with dynamite. It
• 5:13
  gets compromised because a bank teller bought
• 5:15
  a shiny new desk plant from some stranger on
• 5:18
  the street, walked it right past the guards,
• 5:20
  set it on their desk inside the vault. And guess
• 5:22
  what? The plant had a microphone hidden in the
• 5:24
  dirt. Oh, man, that is so spot on. The defenses
• 5:27
  were looking outward, not inward at what the
• 5:30
  employees were carrying inside. That paints a
• 5:32
  really clear picture of the supply chain vulnerability
• 5:35
  here. Because when an employee installs a tool
• 5:38
  inside their environment, that tool inherits
• 5:42
  certain permissions. Right. It gets their access.
• 5:44
  Exactly. If that tool is malicious, the hacker
• 5:47
  doesn't need to actually break into GitHub. They
• 5:49
  just ride the coattails of the employee who already
• 5:51
  has the keys. That is terrifying. And the scale
• 5:54
  of the fallout is what makes this so critical.
• 5:57
  Like you mentioned, GitHub isn't just a regular
• 5:58
  tech company. It's used by 90 % of the Fortune
• 6:01
  100. It's where the biggest companies in the
• 6:04
  world store their private code. Now GitHub has
• 6:07
  officially stated that they don't have any evidence
• 6:10
  that customer data stored outside of those specific
• 6:14
  internal repos has been impacted. Which is a
• 6:17
  relief, I guess. Yeah, they caught the breach,
• 6:19
  they removed the malicious extension, isolated
• 6:21
  the device, and started their response protocols.
• 6:24
  But I mean, the damage to those 3 ,800 internal
• 6:27
  repositories is done. The desk plant was already
• 6:30
  inside the vault. Yeah, the microphone was already
• 6:32
  recording. And the group claiming responsibility
• 6:35
  for planting that microphone is a hacker group
• 6:37
  called Team PCP. Right, Team PCP. Yeah, they
• 6:42
  recently posted on this cybercrime forum called
• 6:44
  Breached, claiming they have access to GitHub
• 6:47
  source code along with thousands of repositories
• 6:50
  of private code. The ransom note they posted,
• 6:53
  though, is what I find so bizarre. Oh, it's weird.
• 6:55
  It's incredibly casual. They're asking for 50
• 6:57
  grand for the data, but they explicitly state
• 6:59
  to the forum like, this is not a ransom. We do
• 7:02
  not care about extorting GitHub. They're just
• 7:05
  throwing it up for auction. Literally optioning
• 7:07
  it to the highest bidder on the dark web. They
• 7:09
  actually wrote, it looks like our retirement
• 7:12
  is soon, so if no buyer is found, we will leak
• 7:14
  it free. The casual tone is just chilling, but
• 7:17
  it reveals a really calculated strategy. Hackers
• 7:19
  know that targeting the developer platforms directly
• 7:22
  is, well, it's highly efficient. Yeah, why hack
• 7:25
  the user when you can hack the tool, right? Exactly.
• 7:28
  If they compromise the building blocks that the
• 7:31
  engineers use, the code editors, the package
• 7:33
  managers, they bypass the end user entirely.
• 7:36
  So bringing this back to you, the listener, if
• 7:39
  you are a working professional or just trying
• 7:41
  to manage your family's digital footprint, The
• 7:43
  takeaway here is pretty stark. The weakest link
• 7:46
  in digital security is rarely the complicated
• 7:49
  backend code. No, not at all. The weakest link
• 7:52
  is human convenience. It really demands a shift
• 7:55
  in how we evaluate the tools we download. We
• 7:57
  have to start treating plugins and extensions
• 8:00
  with the exact same skepticism that we apply
• 8:03
  to suspicious email attachment from an unknown
• 8:06
  sender. Absolutely. The convenience of a quick
• 8:08
  fix just isn't worth the risk if you can't verify
• 8:11
  exactly who built it and what permissions it's
• 8:14
  quietly requesting in the background. Well, speaking
• 8:16
  of our constant desire for convenience, while
• 8:18
  humans are over here making simple errors and
• 8:20
  accidentally installing Trojan horses just to
• 8:23
  save a few keystrokes, Google is rolling out
• 8:25
  AI designed to just do the typing for us entirely.
• 8:28
  Which ties perfectly into the major news out
• 8:30
  of Google I .O. 2026. Yeah, let's talk about
• 8:33
  that. For anyone who doesn't follow it closely,
• 8:36
  Google I .O. is their massive annual developer
• 8:38
  conference, where they basically set the stage
• 8:41
  for the next year of tech. And the entire spotlight
• 8:44
  this year was dominated by the immediate release
• 8:47
  of Gemini 3 .5 Flash. And the rollout strategy
• 8:51
  alone is noteworthy. Google completely skipped
• 8:54
  the usual preview phase. Yeah, they didn't even
• 8:56
  put it in a sandbox for a few months. Right,
• 8:57
  they just dropped it globally, making it available
• 9:00
  immediately in the Gemini app and in Search's
• 9:03
  AI mode. And they're touting this as their strongest
• 9:07
  agentic encoding model yet. It's a very aggressive
• 9:10
  push to integrate this into the daily workflow
• 9:13
  of everyday users. The focus is entirely on bringing
• 9:17
  what they call frontier intelligence directly
• 9:19
  to the consumer. OK, I have to stop the train
• 9:21
  for a second because we need to define some of
• 9:23
  these terms. Sure. Google keeps throwing around
• 9:25
  the word agentic, you know, welcoming us to the
• 9:28
  agentic era, promising agentic workflows. What
• 9:31
  does that actually mean for you listening right
• 9:33
  now? How is an agent different from the standard
• 9:36
  AI chatbots we've all been playing around with?
• 9:38
  It's a really vital distinction and it represents
• 9:40
  a massive leap. Think about a standard chatbot.
• 9:44
  It operates a bit like a highly advanced conversational
• 9:47
  dictionary. Okay. You give it a single prompt,
• 9:50
  you ask a question, or you tell it to draft an
• 9:52
  email, and it gives you a direct one -off response.
• 9:55
  It's entirely reactive, it only moves when you
• 9:58
  push it, and it only takes one step at a time.
• 10:00
  Right, so if I ask for a recipe, it prints the
• 10:03
  recipe, and then it literally goes back to sleep
• 10:04
  until I talk to it again. Precisely. And agentic
• 10:07
  AI fundamentally changes that dynamic. It's designed
• 10:11
  to handle complex long -horizon tasks without
• 10:15
  needing you to hold its hand every step of the
• 10:17
  way. So more autonomous. Yeah, it acts much more
• 10:20
  like a digital intern. Instead of just answering
• 10:22
  a question, you give an agent a broad goal. You
• 10:25
  might say, research our top three competitors'
• 10:28
  pricing for the last three quarters, build a
• 10:30
  comparative spreadsheet, draft an email summarizing
• 10:32
  where we are losing our margin, and send that
• 10:34
  email to the sales team by four o 'clock today.
• 10:37
  Wow, so it's not just generating text, it's executing
• 10:40
  a full multi -step project independently. Yes.
• 10:44
  The AI has to break your large goal down into
• 10:47
  subtasks, navigate different software tools to
• 10:50
  find data, make judgment calls, format the spreadsheet,
• 10:54
  open the email client, and deliver the outcome.
• 10:57
  That's incredible. It possesses a level of autonomy
• 10:59
  or agency. That's what the tech industry means
• 11:02
  by agentic. If this technology truly works the
• 11:06
  way Google is promising, it is going to fundamentally
• 11:08
  shift the day -to -day reality of the professional
• 11:11
  world. We're basically transitioning from a workforce
• 11:14
  that does granular tasks to a workforce that
• 11:17
  manages AI agents doing those granular tasks.
• 11:20
  It requires a completely different skill set.
• 11:22
  Your value won't be in knowing how to build the
• 11:24
  spreadsheet yourself, but in knowing how to instruct
• 11:27
  and quality check the digital intern building
• 11:29
  it. You become an editor and a manager. But of
• 11:32
  course, hiring a fleet of shiny new digital interns
• 11:34
  isn't exactly cheap. No, it's really not. And
• 11:37
  as we dig into the deeper realities of this Gemini
• 11:40
  rollout, we're finding out that sometimes these
• 11:42
  digital interns don't even show up for work when
• 11:43
  they're supposed to. Let's look closer at the
• 11:46
  hitting price of the future. The capabilities
• 11:48
  of Gemini 3 .5 Clash are impressive, sure, but
• 11:51
  that agentic power comes with a very steep price
• 11:54
  tag. It's currently three to six times more expensive
• 11:57
  than its predecessors. Let me put the actual
• 11:59
  numbers on the table so we can see the scale
• 12:00
  here. The pricing is set at one dollar fifty
• 12:03
  cents per one million dollar input tokens and
• 12:06
  nine dollars per one million output tokens. Yeah,
• 12:09
  those tokens add up. Before we go further, let's
• 12:11
  clarify that. If I'm a small business owner trying
• 12:14
  to budget for this. What exactly am I paying
• 12:16
  for when I buy a token? That's the core metric
• 12:19
  of the AI economy. A token is basically how an
• 12:22
  AI reads and generates language. You can think
• 12:25
  of it as a chunk of a word. So 1 million tokens
• 12:27
  translates to roughly 750 ,000 words. Okay. Input
• 12:31
  tokens refer to the data you feed into the AI,
• 12:34
  like your prompts or the documents you upload.
• 12:37
  Output tokens are the words or code the AI generates
• 12:39
  in response. Okay, so $1 .50 to have the AI read
• 12:42
  a massive stack of documents, and $9 for every
• 12:45
  million tokens it spits back out. At first glance,
• 12:48
  that might sound incredibly cheap. A million
• 12:50
  tokens is a lot of text. It sounds cheap until
• 12:52
  you realize how it works. Right. The problem
• 12:54
  is how agentic workflows actually consume those
• 12:57
  tokens. This is where the math gets dangerous.
• 13:00
  When you ask a normal chatbot a question, it
• 13:03
  uses a few tokens to read your prompt and a few
• 13:06
  hundred to give you an answer. Simple. But an
• 13:08
  agentic AI operates in loops. If you tell it
• 13:11
  to build a software feature, it generates a chunk
• 13:14
  of code that uses output tokens. Then it reads
• 13:18
  its own code to check for errors that uses input
• 13:20
  tokens Then it rewrites the code to fix a bug
• 13:24
  More output tokens it loops over and over again
• 13:27
  in the background Burning through tokens at an
• 13:29
  exponential rate before it ever shows you the
• 13:31
  final product which leads to a hilarious But
• 13:35
  very telling anecdote from the rollout I know
• 13:37
  exactly what you're gonna say a tech blogger
• 13:39
  decided to test the new model by asking it for
• 13:41
  a simple image The prompt was just to generate
• 13:44
  an SVG image file of a pelican riding a bicycle.
• 13:48
  A very noble use of frontier intelligence. The
• 13:50
  ultimate stress test, right. So the AI went to
• 13:52
  work and instead of just pulling an image, it
• 13:54
  generated a massive complex block of raw code
• 13:58
  to manually draw this pelican from scratch. Someone
• 14:01
  joked that the resulting image looked like a
• 14:03
  pelican attending a Miami crypto conference because
• 14:07
  it had these blocky retro aviator sunglasses.
• 14:10
  But the real ticker was the cost. That single
• 14:13
  image generation looped so many times that it
• 14:16
  burned through over 14 ,000 output tokens and
• 14:20
  cost the user 13 cents. goofy image of a pelican.
• 14:26
  Now imagine scaling that up to an enterprise
• 14:28
  level. Exactly. If you have a company running
• 14:30
  automated agents constantly in the background,
• 14:32
  parsing databases, grafting hundreds of emails,
• 14:36
  those pennies turn into thousands or even millions
• 14:38
  of dollars, astonishingly fast. The tech labs
• 14:41
  are definitely testing the price tolerance of
• 14:43
  the market right now. They absolutely are. But
• 14:46
  the financial cost is only half the frustration.
• 14:48
  There's a growing, very vocal exhaustion within
• 14:51
  the tech community regarding the psychological
• 14:53
  costs of these massive announcements. Yeah, specifically
• 14:56
  the feeling that we are just drowning in vaporware.
• 14:59
  The reaction to Google I .O. this year was surprisingly
• 15:01
  muted. And it wasn't because the tech wasn't
• 15:04
  impressive. It was because almost none of the
• 15:06
  truly groundbreaking features are actually ready
• 15:09
  for the public to use today. It really felt like
• 15:11
  an endless parade of coming soon. They announced
• 15:14
  their absolute top -tier model, Gemini 3 .5 Pro,
• 15:18
  but admitted it won't be available until next
• 15:20
  month. They showed off this mind -blowing video
• 15:23
  generation tech, but told everyone it's limited
• 15:26
  to a select group of paid users sometime later.
• 15:29
  The generative UI for Search, coming this summer.
• 15:32
  The agents and mini -apps, also this summer.
• 15:35
  They even teased new intelligent eyewear, but
• 15:38
  suggested it might be this fall or maybe even
• 15:40
  next year. This strategy creates a very specific
• 15:43
  type of fatigue. When a massive company holds
• 15:46
  an annual keynote to hype up products that don't
• 15:48
  actually exist on the market yet, it feels, well,
• 15:52
  manipulative. It totally does. The goal isn't
• 15:54
  just to excite the consumer. The goal is to freeze
• 15:56
  the market. They want to convince you to hold
• 15:59
  off on buying a competitor's product today because
• 16:01
  Google's vastly superior version is supposedly
• 16:05
  right around the corner. As a professional trying
• 16:07
  to navigate this space, you really have to ask
• 16:09
  yourself, Are we paying premium subscription
• 16:12
  prices today just to act as beta testers for
• 16:15
  features that are perpetually stuck in the pipeline?
• 16:17
  And that fatigue bleeds right into another phenomenon
• 16:20
  that's deeply impacting how we experience the
• 16:23
  internet right now. There's a term gaining a
• 16:26
  lot of traction lately that perfectly captures
• 16:29
  this feeling. the zombie internet the zombie
• 16:32
  internet it sounds like the title of a bad sci
• 16:34
  -fi movie but it is incredibly descriptive of
• 16:37
  what is happening and we should clarify for the
• 16:38
  listener this is distinctly different from the
• 16:41
  dead internet theory yes the dead internet theory
• 16:43
  is the idea that the web is entirely devoid of
• 16:47
  humans just automated bots talking to other automated
• 16:50
  bots to generate fake traffic right the zombie
• 16:53
  internet is much more insidious because humans
• 16:55
  are still involved But their interactions are
• 16:58
  entirely mediated by AI. It's humans using AI
• 17:02
  to talk to other humans who are using AI. Think
• 17:04
  about how exhausting this is in a daily workflow.
• 17:07
  Let's say a colleague wants to send you an update.
• 17:10
  Instead of just writing three bullet points,
• 17:12
  they use an AI agent to draft a polished five
• 17:16
  paragraph email. They send it to you. You receive
• 17:19
  this massive wall of text and instead of reading
• 17:22
  it, you use your own AI agent to summarize the
• 17:25
  email back down into three bullet points so you
• 17:27
  can understand it. You have two machines performing
• 17:31
  a highly inefficient expensive dance. just to
• 17:34
  transmit a simple thought between two humans.
• 17:36
  It's absurd. And it goes far beyond emails. You
• 17:39
  have influencers spinning up automated YouTube
• 17:41
  channels to spam the algorithm for ad revenue.
• 17:44
  You have AI -generated summaries of books flooding
• 17:47
  online marketplaces. Drowning out the actual
• 17:49
  authors. Exactly. You even see it in social spaces.
• 17:52
  Someone will post a heartfelt, vulnerable story
• 17:54
  on a forum seeking human connection, and they
• 17:56
  get a deeply empathetic, multi -paragraph response.
• 17:59
  that was entirely generated by a marketing firm
• 18:01
  using a language model to build account engagement.
• 18:04
  Oh, that just hollows out the core of why we
• 18:07
  communicate in the first place. It creates this
• 18:10
  profound, underlying sense of distrust. You never
• 18:13
  truly know who or what is on the other side of
• 18:16
  the screen. You're constantly wondering if the
• 18:18
  warmth in an email is genuine, or if it was just
• 18:21
  an output token optimized by an algorithm. Then
• 18:23
  that constant second -guessing requires a massive
• 18:26
  amount of cognitive load. If we synthesize all
• 18:29
  of these elements together, the exponentially
• 18:32
  higher token costs of Eugentic AI, the endless
• 18:35
  vaporware promises that freeze the market, and
• 18:38
  the sheer mental exhaustion of navigating the
• 18:40
  zombie internet, it all points to a massive reality
• 18:43
  check. Definitely. For the everyday professional,
• 18:46
  the math simply has to make sense. The value
• 18:48
  proposition can't just be about speed anymore.
• 18:51
  If your shiny new AI coding agent helps you write
• 18:54
  software twice as fast, you'd better be absolutely
• 18:57
  certain that it also cut your maintenance costs
• 18:59
  in half. Because if you double your output, but
• 19:01
  the AI generates messy, convoluted code that
• 19:04
  takes your human engineers twice as long to debug,
• 19:07
  you haven't actually saved any time or money.
• 19:09
  You've just created a larger, more expensive
• 19:11
  mess. You are trading a temporary burst of speed
• 19:14
  for permanent maintenance debt. The adoption
• 19:17
  of these tools needs to be deeply intentional.
• 19:19
  It's a cautionary tale to avoid AI FOMO, the
• 19:23
  fear of missing out. You shouldn't adopt an agentic
• 19:26
  workflow just because it was announced at a flashy
• 19:28
  developer conference, or because you feel pressure
• 19:31
  to have an AI strategy. Right. You should only
• 19:34
  adopt these tools if they demonstrably solve
• 19:36
  a specific problem in your life without creating
• 19:39
  a larger, more exhausting problem in its place.
• 19:42
  Absolutely. Before we wrap up, a quick thanks
• 19:44
  to Old Glory, an iconic music and sports fan
• 19:47
  merch store. Perfect for the techie in your life.
• 19:50
  Use promo code SNARFL for 15 % off at oldglory
• 19:54
  .com. Great stuff. So let's take a step back
• 19:57
  and look at the broader journey we unpack today.
• 19:59
  We started by looking at the very real dangers
• 20:01
  of our own digital habits, seeing how a single,
• 20:04
  casually installed plug -in brought a massive
• 20:06
  headache to the internal systems at GitHub. It
• 20:08
  was a stark reminder that the strongest vault
• 20:11
  in the world is useless if we willingly carry
• 20:13
  the microphone inside. Very true. From there,
• 20:17
  we explored the arrival of the agentic era, dissecting
• 20:20
  how Google's new Gemini 3 .5 flash is attempting
• 20:23
  to transition the workforce from doing our own
• 20:26
  digital chores to managing complex digital interns.
• 20:29
  But we also face the harsh reality of that transition.
• 20:32
  Those digital interns are surprisingly expensive
• 20:35
  to run, many of their best features are stuck
• 20:38
  in the vaporware pipeline, and we are all feeling
• 20:41
  the growing exhaustion of navigating a zombie
• 20:44
  internet where humans use machines to talk to
• 20:46
  other machines. Which really leaves us with a
• 20:49
  fascinating paradox to consider. We are pouring
• 20:51
  billions of dollars into building technology
• 20:53
  designed to perfectly mimic human output. at
• 20:56
  scale in order to make our lives easier and more
• 20:58
  efficient. Right. But as the internet fills up
• 21:00
  to the brim with these polished, perfectly articulate
• 21:03
  AI agents talking to other AI agents, it really
• 21:06
  makes you wonder, as everything becomes artificially
• 21:08
  perfect, will genuine, unpolished human connection,
• 21:12
  you know, the kind with typos, raw emotion, and
• 21:15
  actual vulnerability, become the ultimate priceless
• 21:18
  luxury of the future? We're here to help at Snarfles,
• 21:21
  so reach out on our website if you have any more
• 21:22
  questions. That's a wrap on today's top stories.
• 21:29
  Remember, tech doesn't have to be complicated,
• 21:31
  and we're here to help connect the dots. If you've
• 21:34
  got questions or want to learn more, visit us
• 21:37
  at snarful .com. Thanks for listening, and we'll
• 21:40
  catch you next time.