Latest / Today’s AI News / Claude Beats ChatGPT at Work, Amazon’s Alexa Goes Agentic, AI Cyberattacks Are Doubling Monthly

Transcript

Claude Beats ChatGPT at Work, Amazon’s Alexa Goes Agentic, AI Cyberattacks Are Doubling Monthly · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by Nynex
• 0:02
  Productions. Imagine watching a marathon
• 0:05
  where the undisputed champion suddenly looks over their
• 0:07
  shoulder, only to see the quiet runner in the practical shoes
• 0:10
  blowing right past them. That's exactly what's happening
• 0:14
  in the AI enterprise race today, as a new spend report shows
• 0:18
  Anthropic officially overtaking Open AI.
• 0:21
  Let's unpack what this flip means for the industry.
• 0:24
  It really is a stunning reversal when you look at the raw
• 0:26
  numbers, right? Yeah, I mean, we are seeing a
• 0:29
  foundational shift here. AI is moving from just acting as
• 0:33
  a clever brainstorming partner to actually functioning as the
• 0:36
  load bearing infrastructure for modern businesses.
• 0:39
  Absolutely. And you know, to kick off our
• 0:41
  mission for this deep dive, we're going to examine that
• 0:44
  anthropic takeover. Then we'll look at Amazon's
• 0:46
  aggressive new AI shopping agent, some crazy breakthroughs
• 0:49
  in automated AI training, and then rounded out with our quick
• 0:52
  hits and community workflows. There's a lot of ground to cover
• 0:55
  today. There really is.
• 0:56
  So let's dive right into the data driving this enterprise
• 0:59
  flip because frankly, it is staggering.
• 1:02
  We are looking at the newest AI index from the fintech firm
• 1:05
  Ramp. Right.
• 1:06
  And if you're listening and aren't familiar with Ramp, they
• 1:08
  process corporate cards and invoice payment.
• 1:10
  Exactly, across more than 50,000 US businesses.
• 1:14
  So we aren't looking at some survey where executives just,
• 1:17
  you know, claim they use AI. Yeah.
• 1:19
  Surveys can be wildly inaccurate.
• 1:21
  Right. We are looking directly at the
• 1:23
  expense reports, the hard capital.
• 1:25
  Which is the ultimate lie detector in the tech industry,
• 1:28
  really. You can issue all the press
• 1:30
  releases you want, but the corporate credit card tells the
• 1:32
  truth about what tools employees actually rely on.
• 1:34
  So true, and the aggregate data shows overall AI use in business
• 1:38
  has climbed to 50.6%. But the real shocker is the
• 1:42
  changing of the guard. The anthropic numbers.
• 1:45
  Yes. In April, Anthropic rose 3.8% to
• 1:47
  capture 344% of paid business adoption.
• 1:51
  Meanwhile, Open AI actually fell 2.9%, dropping down to 32.3%.
• 1:56
  And Anthropic has seen a four times adoption surge since 2025.
• 2:00
  OK, let's unpack this. Wait, is Open AI actually cooked
• 2:03
  here? Or is this just one specific
• 2:05
  slice of the pie since Ramp isn't tracking massive custom
• 2:09
  enterprise deals? What's fascinating here is that
• 2:12
  Open AI is certainly not obsolete.
• 2:14
  I mean, ChatGPT remains the undisputed heavyweight champion
• 2:17
  in the consumer brand space. Sure, and you make a great
• 2:20
  point. Ramp's data doesn't capture the
• 2:22
  bespoke deals negotiated directly in corporate
• 2:24
  boardrooms, however. The trend is really the big
• 2:27
  story here. A. 400% adoption surge over a
• 2:31
  single year is massive. It is.
• 2:33
  It represents A profound behavioral shift.
• 2:35
  It means individual developers and department heads are
• 2:39
  specifically choosing to expense anthropic products over the
• 2:42
  default open AI options. But why though?
• 2:46
  What is the actual mechanism driving that preference?
• 2:49
  Because for a long time both models felt relatively
• 2:52
  comparable to the average user. Well, it comes down to
• 2:54
  structural utility in high stakes environments.
• 2:57
  The data points to Claude code as the anchor for much of this
• 3:00
  swing. Oh, interesting.
• 3:01
  Yeah, historically, Claude was adopted by highly technical
• 3:04
  software teams. But what's happening now is that
• 3:07
  utility is bleeding over into finance, legal and deep research
• 3:11
  workflows. So department's dealing with
• 3:13
  highly complex, dense documentation.
• 3:16
  Exactly. They don't just need a chatbot
• 3:18
  to write a polite e-mail. Claude's architecture
• 3:21
  specifically. It's extremely reliable.
• 3:23
  Recall within a large context window means a legal team can
• 3:26
  upload a 500 page merger contract and ask intricate
• 3:30
  questions. Without the AI hallucinating
• 3:32
  clauses that don't exist. Right, it's incredibly reliable.
• 3:35
  I see, because if you are parsing a legally binding
• 3:38
  document, accuracy isn't just a nice feature, it's the entire
• 3:41
  product. Precisely and clearly, this
• 3:44
  bleed into the B to B market share hasn't gone unnoticed by
• 3:47
  Open AI's leadership. In fact, this specific data
• 3:50
  trend is exactly why Open AI Executive Fiji Simo initiated an
• 3:54
  internal Code Red pivot about two months ago.
• 3:56
  A cobread. Yeah, you don't reorganize your
• 3:59
  company's priorities unless you feel the tectonic plates
• 4:01
  shifting beneath you. They are aggressively
• 4:04
  prioritizing enterprise pushes like Codex to try and reclaim
• 4:07
  that mindshare. I have to play devil's advocate,
• 4:10
  though. Anthropic's momentum is
• 4:11
  impressive, but Ramp highlighted some serious vulnerabilities for
• 4:15
  them too, right? They did, yeah.
• 4:17
  They noted recent clawed outages.
• 4:19
  And frankly, if you are positioning yourself as the core
• 4:23
  infrastructure for a finance team pulling quarterly earnings,
• 4:27
  you cannot go dark. Uptime is the unglamorous
• 4:30
  reality of enterprise software. You are absolutely right.
• 4:34
  If a legal department score analysis engine goes down during
• 4:37
  a critical negotiation, they will switch back to a competitor
• 4:40
  instantly. Not to mention the rising costs.
• 4:42
  Right Anthropic is facing rising operational costs compared to
• 4:46
  rapidly improving open source alternatives and open AI.
• 4:50
  To hold on to that 34% market share, their reliability has to
• 4:54
  be bulletproof. Yeah, definitely.
• 4:55
  And, you know, we're seeing this theme of deep structural
• 4:58
  integration everywhere today. Anthropic is embedding itself
• 5:01
  into how businesses operate, but Amazon is attempting something
• 5:04
  arguably more ambitious. Just as Anthropic is winning by
• 5:08
  integrating deeply into specific corporate workflows like
• 5:11
  finance, Amazon is trying to dominate the consumer space by
• 5:15
  deeply integrating AI into the ultimate consumer workflow
• 5:19
  shopping. Yes.
• 5:20
  RIP Rufus. Yeah, the quiet execution of
• 5:23
  Rufus. Amazon officially killed off its
• 5:25
  standalone shopping chatbot, which is wild considering it had
• 5:28
  over 300 million users in beta last year.
• 5:31
  But they didn't scrap the underlying technology.
• 5:34
  No, they folded all that product knowledge into a sweeping new
• 5:37
  agent simply called Alexa for shopping.
• 5:39
  Which is a classic consolidation play.
• 5:41
  Having multiple fragmented AI personalities just confuses the
• 5:45
  everyday shopper. By rolling this under the Alexa
• 5:47
  umbrella, Amazon is signaling that this isn't an experimental
• 5:50
  feature anymore. It's the new default, and the
• 5:52
  mechanics of this new Alexa are intense.
• 5:55
  It shares memory across all your devices.
• 5:57
  It remembers your past purchases, your specific brand
• 6:00
  preferences, prior conversations.
• 6:02
  It can even run side by side product comparisons right in the
• 6:06
  search bar. Plus the new auto buy feature
• 6:08
  for when items hit a target price.
• 6:11
  But here's where it gets really interesting.
• 6:13
  They introduced a buy for me feature that handles checkouts
• 6:16
  on non Amazon stores. That specific feature is a
• 6:20
  massive paradigm shift. It's like having a hyper
• 6:23
  attentive personal assistant. Think about your own habits for
• 6:26
  a second. Imagine an AI that knows you
• 6:28
  bought your favorite toothpaste three months ago.
• 6:31
  Right, so it knows you're running low.
• 6:32
  Exactly. It tracks your cadence.
• 6:34
  It also knows exactly what your maximum price limit is, and if
• 6:38
  Amazon's price happens to be too high, this AI will actually
• 6:41
  leave the Amazon ecosystem, navigate to a competitor's
• 6:45
  website, and buy that toothpaste for you.
• 6:47
  And with schedule actions, it just automates that restocking
• 6:50
  entirely. It's wide.
• 6:52
  When you look under the hood of how Biferme actually functions,
• 6:54
  it is a marvel of agentic web scraping.
• 6:57
  Amazon has a Moat that is practically unassailable.
• 7:00
  Their sheer volume of customer history.
• 7:02
  Yes, the deepest reservoir of consumer purchase history,
• 7:06
  delivery timing, and verified reviews on the planet.
• 7:08
  They are using that to train Alexa to be a highly autonomous
• 7:12
  purchasing agent. I have a major reservation about
• 7:14
  this though. You and I, and probably most
• 7:17
  people listening, are already building habits with other AI
• 7:21
  platforms. Right for research and things.
• 7:23
  Yeah, if I use Claude or ChatGPT to research a new camera for an
• 7:27
  upcoming trip, I want that agent to just buy it for me.
• 7:31
  Will Alexa actually play nice and integrate with outside
• 7:34
  ecosystems? Or is Amazon just trying to
• 7:36
  build the ultimate walled garden?
• 7:38
  This raises a critical tension in the industry right now.
• 7:41
  Interoperability versus ecosystem lock in.
• 7:45
  Historically, Amazon prefers closed loops.
• 7:48
  They want you in their house. Using their tools.
• 7:50
  Exactly. But consumers increasingly
• 7:52
  expect their various AI agents to talk to each other.
• 7:55
  If Alexa for Shopping refuses to interface with external research
• 7:58
  agents, Amazon really risks throttling its own long term
• 8:02
  adoption. But while Alexa is automating
• 8:04
  our checkout carts, power users are figuring out how to use AI
• 8:07
  to automate the very creation of AI content.
• 8:10
  We are moving past the days of typing a prompt and crossing our
• 8:13
  fingers. Way past it, there's a new
• 8:16
  workflow the community's been using that combines Claude code
• 8:18
  with the Higgs field CLI. Now, before we go further, if
• 8:21
  you're listening while commuting and wondering what ACLI or Node
• 8:25
  or NPM even are, let's break that down.
• 8:27
  Sure, think of Node and NPM as the basic plumbing and building
• 8:32
  permits required to run raw code on your computer.
• 8:34
  OK, and ACLI or command line interface is just a way to talk
• 8:38
  to your computer by typing text commands rather than clicking on
• 8:41
  shiny buttons. Perfect.
• 8:43
  So by installing this Higgs field scale into that
• 8:46
  environment, you can tell Claude Code to take a single image
• 8:49
  prompt and autonomously run it across six different AI image
• 8:53
  models simultaneously. Mid journey, stable diffusion
• 8:56
  all at once. Exactly.
• 8:58
  And Claude Code collects the generated images from all six
• 9:01
  models and saves them directly into a project folder on your
• 9:04
  machine. Which is a huge time saver.
• 9:06
  But it gets better. It then takes it a step further
• 9:09
  and automatically generates a comparison markdown file.
• 9:11
  So basically the text file with notes and detailed critiques on
• 9:15
  how each model interpreted your prompt.
• 9:17
  That is brilliant. Yeah, so your job as the human
• 9:20
  is just to open one file, read the AI's notes, pick the best
• 9:24
  image, and ask Claude to refine it.
• 9:27
  Oh, and Hinksfield supports video clips now, too.
• 9:29
  And if that setup sounds tricky, the barrier to entry has really
• 9:33
  vanished. You don't need to know how to
• 9:34
  code, you just ask Claude code to check your node and NPM setup
• 9:39
  and it will provide a step by step walkthrough.
• 9:41
  It holds your hand through the whole thing.
• 9:43
  Exactly. It perfectly illustrates how our
• 9:45
  rules are shifting from operator to manager.
• 9:48
  You aren't prompting models manually anymore, you're hiring
• 9:51
  an AI to manage a team of other AIS.
• 9:54
  Wow. And if Claude code acts as a
• 9:56
  manager for the output of models, there's an even bigger
• 9:59
  shift happening with how models are actually built.
• 10:01
  What if? AI could automate its own
• 10:03
  training, right? Let's talk about adaption.
• 10:05
  This is a startup from Sarah Hooker, the former VP of
• 10:08
  Research at Cohere, and they just introduced a system called
• 10:11
  Auto Scientist. This is where we cross from
• 10:14
  workflow optimization into structural evolution.
• 10:17
  Auto Scientist automatically customizes AI models for
• 10:20
  specific jobs by adjusting what the model learns from and how it
• 10:24
  learns. But how does an AI actually do
• 10:27
  that? To understand the breakthrough,
• 10:29
  think about how fine tuning normally works.
• 10:32
  Customizing a model requires human engineers to manually
• 10:35
  adjust things called hyper parameters.
• 10:38
  You can think of hyper parameters as the dials and
• 10:41
  knobs on a complex sound board that dictate how a neural
• 10:44
  network learns, like adjusting its learning rate.
• 10:47
  I like to think of it like trying to tune a grand piano
• 10:50
  while wearing oven mitts. It is incredibly delicate,
• 10:53
  tedious work. That is a perfect analogy, and
• 10:56
  Auto Scientist completely automates that loop.
• 10:58
  The mechanism driving this is a secondary evaluator model.
• 11:01
  So an AI grading an AI? Yes.
• 11:03
  Auto Scientist tweaks the dials, generates a result, and then
• 11:06
  uses a secondary AI to grade that result.
• 11:08
  It iterates until it meets the user's goal.
• 11:10
  It's a closed loop system of trial and error that runs
• 11:14
  thousands of times faster than a human.
• 11:15
  And the results in their internal tests speak for
• 11:17
  themselves. Auto Scientist outperformed
• 11:20
  expert tuned models by 35% on average.
• 11:22
  That is massive. It bumped the overall success
• 11:25
  rate of customized models from 48% up to 64%, and this held up
• 11:31
  across 8 different industries, including finance, medical and
• 11:35
  legal fields. If we connect this to the bigger
• 11:37
  picture, the implications are sweeping.
• 11:40
  Right now, Fine tuning a frontier model is essentially a
• 11:43
  dark art understood by only a few thousand people globally,
• 11:47
  mostly at elite labs. Auto Scientist democratizes this
• 11:50
  expertise. It makes creating highly
• 11:52
  customized business models infinitely more practical and
• 11:54
  accessible. This rapid automation of AI
• 11:57
  training and deployment is sending shock waves across the
• 11:59
  entire tech landscape. Let's look at the ripple effects
• 12:02
  in our quick hits. Let's run through.
• 12:03
  Them so trending tools today we already discussed Alexa for
• 12:07
  shopping and the Higgs field supercomputer cloud agent.
• 12:10
  We're also seeing huge traction for Incognito chat.
• 12:14
  The private meta AI chats. Yes, directly on WhatsApp.
• 12:17
  And keeping with the enterprise theme, Entropic just released
• 12:20
  Quad for Small Business, which introduces tools for handling
• 12:23
  payroll, generating invoices and running campaigns.
• 12:26
  That is a direct offensive maneuver to solidify that 34%
• 12:30
  market share we talked about earlier.
• 12:32
  Moving to Industry 3 News, NVIDIA became the first company
• 12:36
  to hit a $5.5 trillion market cap today, and this coincided
• 12:40
  with CEO Jensen Wong arriving in China to join U.S.
• 12:44
  President Donald Trump in meetings with Xi Jinping A.
• 12:46
  Massive geopolitical moment. It is, and I need to pause right
• 12:50
  here and state this explicitly to you, the listener.
• 12:53
  We are simply reporting the facts of this meeting, as
• 12:56
  detailed in the sources, without endorsing any political
• 12:59
  viewpoints. Right, exactly.
• 13:01
  Now back in the US, Sam Altman testified today in the Elon Musk
• 13:04
  versus Open AI legal battle. Altman revealed that Musk's
• 13:08
  specific safety plans worried him so much that Altman proposed
• 13:11
  passing the company to his children to safeguard it.
• 13:14
  The courtroom drama over superintelligence is just
• 13:16
  surreal. It really is.
• 13:18
  In other news, David Silver's Ineffable Intelligence partnered
• 13:21
  with NVIDIA. They are building training
• 13:23
  pipelines for reinforcement learning agents targeting
• 13:26
  Nvidia's Vera Rubin hardware. Which is hardware explicitly
• 13:29
  designed to handle the massive data transfers those agents
• 13:32
  require. Exactly.
• 13:34
  And Microsoft introduced MDAH, which chains together more than
• 13:37
  100 specialized AI agents into a single swarm to hunt bugs.
• 13:42
  A defensive swarm. Yes, and it caught 16 flaws in
• 13:45
  Windows. And we need those defenses
• 13:47
  because EU KS AI Safety Institute warned today that AI's
• 13:51
  ability to complete cyberattacks is doubling every few months.
• 13:54
  They noted Mythos Preview and GPT 5.5 successfully finished
• 13:59
  their simulated breaches. So what does this all mean?
• 14:01
  Well, looking at Microsoft's MDAHH defensive agents and the
• 14:05
  UK Safety Institute's offensive warnings, it shows we are
• 14:08
  officially entering an era of AI versus AI cybersecurity.
• 14:12
  Human reaction time is just too slow now.
• 14:14
  It's an arms race at lightspeed. Between $5.5 trillion valuations
• 14:18
  and simulated cyber breaches, it's easy to feel overwhelmed.
• 14:22
  But let's look at how this tech is actually helping everyday
• 14:24
  people right now in our community AI workflows.
• 14:27
  It's so important to bring it back down to earth.
• 14:28
  Absolutely. Today's story comes from Rod RA,
• 14:31
  61 year old avid cyclist from Elizabethtown, KY.
• 14:34
  Rod needed coaching for grueling multi day events but couldn't
• 14:37
  afford a human coach. So he turned to ChatGPT.
• 14:40
  He did. He had a long conversation about
• 14:42
  his medical history, his goals, baseline fitness, event dates,
• 14:46
  and he's been on this custom plan for six weeks.
• 14:48
  That's great. But the best part is ChatGPT
• 14:51
  acts as an ongoing coach. It restructures the plan
• 14:54
  dynamically when life gets in the way, remembers his past
• 14:57
  chats, and gives him real Peace of Mind.
• 15:01
  Rod's final note to us was today's Arrest Day.
• 15:04
  I absolutely love this. You know, while corporations
• 15:06
  fight over enterprise market share, the most profound impact
• 15:10
  of AI is often deeply personal. It provides custom adaptive
• 15:14
  guidance that was previously gate kept by high costs.
• 15:17
  It really changes lives on an individual level.
• 15:19
  It does. And, you know, we've seen AI
• 15:21
  acting as a personal shopper, A-Team manager for image
• 15:24
  generation, a cybersecurity hacker and a cycling coach
• 15:26
  today. Yeah.
• 15:27
  If tools like auto scientists mean AI can now train itself to
• 15:31
  do all these things better, at what point do you and I stop
• 15:35
  being the pilots of this technology and simply become the
• 15:38
  passengers? Man, that is a heavy,
• 15:41
  fascinating thought to sit with. Well, if you want to keep riding
• 15:45
  shotgun with us through all these structural shifts, make
• 15:47
  sure you hit that subscribe button to stay updated.
• 15:49
  We'd also really invite you to rate the show 5 stars.
• 15:52
  It helps us out more than you know.
• 15:53
  Thank you so much for joining us, and we'll see you back for
• 15:56
  tomorrow's deep dive.