It's strange to think about our current reality. Beat. We have robots learning to read physical thermometers now. Yeah, which is just incredible. Right. But at the exact same moment, executives are scheming. Yeah. They leak memos to sabotage billion -dollar company valuations. It's a wild split screen of technological progress. We are living in two completely different realities right now. One is pure science, and one is just cutthroat business. Welcome to the Deep Dive.
I am glad you are here with us today. Yeah, it's going to be a good one. We are exploring the rapidly maturing AI ecosystem, and we are going to start at the highest boardroom level. We have a massive OpenAI corporate leak to unpack. Oh, yeah, a really big one. From there, we will move into a major software shift. AI is going from simple chat to autonomous employees. Which totally changes everything. Exactly. And finally, we
will look at a very physical evolution. These digital agents are breaking out of traditional software. They are physically interacting with our tangible real world. It's a massive structural shift across the board. I mean, every layer of the technology stack is basically changing. Let's start with the aggressive business maneuvers currently happening. Because these corporate wars directly shape the tools you use. Right. They dictate the market. A leaked internal memo
from OpenAI just surfaced recently. It comes directly from their chief revenue officer. Her name is Denise Dresser. She claims Anthropic's $30 billion revenue run rate is inflated. She basically called their financial success a complete mirage. Calling it a mirage is a very aggressive stance. That's an incredibly heavy word in corporate strategy. It really is. She is... directly attacking their core financial narrative there. You don't
use the word mirage lightly. No, you don't. She also labeled Anthropic a single product company in a platform war. Ouch. Yeah. She argues that OpenAI has a very multi -pronged approach. This diversity makes OpenAI a much more resilient investment. Right. And the broader context here is really fascinating because Anthropic relies heavily on a single core product offering. But OpenAI is trying to build an entire operating system. They want to be the foundational layer
for everything. But here is the truly surprising part of this leak. Dresser actually offered a candid critique of OpenAI itself. Yeah, which you almost never see. Right. She criticized their highly publicized partnership with Microsoft specifically. That is a fascinating admission from a key executive. You just rarely see them critique their biggest financial backer. Dresser admitted the Microsoft deal has been quite limiting. It restricts their enterprise growth due to Azure
-linked bottlenecks. Azure is Microsoft's massive cloud computing platform infrastructure. So if Azure struggles, OpenAI essentially struggles right alongside them. Exactly. They are inherently tied to Microsoft's physical server capacity. Which explains the strategic deal they made in February. They brought OpenAI models directly over to Amazon Bedrock. Right. And the demand has been totally staggering since they made that move. It really looks like OpenAI is using Amazon
specifically. They want to bypass Microsoft's constraints entirely. Yeah, to capture a much wider enterprise market quickly, which is a wild corporate power move, honestly. It really is. They are playing the big cloud providers against each other. They are ensuring they don't get bottlenecked by one partner. She also slammed Anthropic's current internal compute shortage aggressively. Dresser called their server constraints
a massive strategic failure. Wow. Yeah. This shortage forces Anthropic to heavily throttle user access. Meanwhile, OpenAI just continues to scale its operations globally. Well, compute power is the ultimate currency in this industry. If you cannot give users access, you lose them instantly. Right. A mass evaluation means absolutely nothing without the server space. It is like uncoupling your train from the main engine. You are going 100 miles an hour. You just have to
hope the new engine catches you. Neat. But I still wrestle with filtering out the corporate spin myself when reading these leaked memos. Oh, totally. It's... incredibly difficult to separate fact from fiction. The spin is heavily engineered to sway public perception. You always have to ask who benefits from the leak. Right. Either OpenAI's internal security is an absolute disaster right now, or this supposed leak is a highly calculated strike. OpenAI is effectively
writing its S1 filing in public. An S -1 is the formal document to go public. You need a highly compelling narrative to sell to investors. Exactly. It's the story that justifies your multi -billion dollar valuation. So Dresser is strategically framing Anthropic as a fear -based company. She paints them as overly focused on rigid safety constraints. Meanwhile, she frames OpenAI as the positive message growth engine. In lead up
to a public debut, narrative is everything. Investors are buying the vision just as much as the tech. If investors start doubting Anthropic's $30 billion financial figure today, it could shift billions in valuation toward OpenAI overnight. The financial stakes are astronomically high at this moment. A single -linked memo can literally alter market dynamics. We also see these platform tensions with Apple recently. Apple threatened it could remove Grok from the App Store. Right, over the
deepfakes. Yeah, this was over concerns about generating highly sexualized deepfakes. This issue was raised in a January letter to U .S. senators. We're just looking at the facts of that tension here. Platform control is truly the ultimate corporate battleground today. Gatekeepers like Apple hold immense power over distribution channels. They essentially dictate what software reaches billions of consumers. So are these platform
wars ultimately hurting the end user? The fight for dominance pushes companies to innovate rapidly, but when they lock you into their specific ecosystem, it hurts. It creates massive roadblocks for actually getting work done. So competition forces innovation, but platform locks create friction. Exactly. It's a very delicate balance for the industry. That friction pushes us right into our next phase because these companies are fighting to justify massive valuations. Right. They have to evolve
their products from basic chat bots. They become actual digital workers inside our systems. They need to integrate deeply into our daily software. A simple chat window isn't enough to justify a massive subscription. Microsoft is considering a truly massive shift right now. They may start charging for AI agents like human employees. They would give them logins, seats, accounts, and permissions. You'd onboard an AI just like a new hire. We should clarify what that terminology
actually means here. AI agents are basically programs that act independently to complete multi -step tasks for you. That is the core definition of these AI agents. Perfect. And companies could literally pay per digital worker. They would manage them in their traditional HR software. Paying for software per seat. Like a human employee. That fundamentally changes the economics of running a modern business. You are essentially hiring a digital staff member now. You are not just
buying a static software tool anymore. You are buying labor, not just a bunch of code. Microsoft is working on yet another open -claw -like digital agent. The goal is a highly secure, always -on assistant tool. It handles multi -step tasks with much stronger enterprise controls. Enterprise security is the biggest hurdle for widespread adoption. Companies need absolute certainty before granting agents system access. Exactly. An autonomous
agent with a login is a huge security risk. A mysterious leak claims Anthropic is planning something similar. Early hints suggest a major shift toward built -in workflows. Everyone is trying to escape the basic chat window. They want the AI to do the work in the background. We see these workflow tools expanding everywhere right now. Google added new AI skills directly to their Chrome browser. Yeah, that's a big deal. You can save your absolute favorite workflows
very easily. Then you can run them on any website later. Gemini Personal Intelligence also just expanded operations into India. So the global footprint of these tools is growing incredibly fast. is actually going viral currently. Yeah. It is heavily inspired by the developer Andrej Karpathy. Oh, I saw that. Yeah. It fundamentally controls how your coding AI behaves locally. Think of it like a strict constitution for your AI. Before it writes a single line of code, it
checks this text file. It ensures it isn't breaking your specific local rules. It forces much clearer logic and reduces coding mistakes. It also prevents the AI from making unintended file edits. It's about giving the human much tighter control. The AI needs boundaries to be truly effective for you. We are seeing a new wave of empowered builders everywhere. New tools treat AI as highly independent software workers. Rather than listing them all out separately today, let's look at
how they create a unified ecosystem. Figma for agents is a perfect example of this. Yeah, that one is fascinating. You can design directly on the standard Figma canvas. You guide agents with context about your team's intent. It is like stacking Lego blocks of data visually. You are giving the AI visual parameters to work within. Then there is a tool called CatDoze version 4. It is essentially a powerful no -code application builder tool. Right. It runs in the cloud and
has its own computer. You can literally close your laptop browser tab completely. And the system just keeps working in the background. That is the real shift toward autonomous digital employees right there. It handles the tedious labor while you do other things. Soft AI co -builder is another incredibly powerful new tool. It builds custom portals and complex internal operation systems. You just describe what you need to the system, then you refine it with prompts or edit visually.
These tools democratize software creation for non -technical business users. You don't need an engineering degree to build complex applications anymore. Caveman is another tool that optimizes AI token usage. It cuts 75 % of Claude's output tokens easily. Which is huge. It manages to do this without losing any factual accuracy. That represents a massive cost saving for heavy users. Efficiency is just as important as raw computing power today. It is a simple one -line install
for popular coding tools. It works with quad code, cursor, windsurf, and copilot. AI is becoming the foundational layer of modern work. It is weaving itself into every tool we actively use. So if an AI has its own seat, who takes the blame when it fails? That is a very complicated legal and operational issue. Legally, the system absolutely requires a responsible human overseer. Right. You cannot fire an AI for making a critical error. You must trace the failure back to the human
deployer. Accountability stays with the human manager, even for digital staff. Yeah, the human is always the ultimate safety net. Sponsor. We just discussed the rapid rise of workflow autonomy. If an AI can run while you close the PAB, What happens when it gets physical eyes and hands? It breaks out of the digital screen entirely then. It starts interacting with the messy, unpredictable real world. Exactly. Gemini Robotics ER 1 .6 is Google's answer to this. Most robots are great
at following a pre -written script. But they are terrible at understanding a dynamic, changing environment. They struggle when the physical world does something unexpected. If a coffee cup is moved two inches, they fail completely. Gemini 1 .6 addresses this massive physical reasoning gap directly. It is a total overhaul of how a robot perceives. Yeah. It allows the machine to verify its own physical work. This is a massive
milestone for physical AI systems. They are finally moving past rigid, pre -programmed robotic movements. Let's examine instrument reading as a prime example here. This is a major win for the industrial inspection field. Oh, absolutely. The model can now interpret complex circular pressure gauges. It can accurately read traditional thermometers and basic digital displays. Reading a physical dial is notoriously difficult for machines. Lighting, angles, and reflections usually confuse basic
computer vision. The real world is just full of messy visual noise. The system actually zooms in on the physical instrument. It runs custom code to estimate the specific tick marks. It uses its broad world knowledge to evaluate things. It inherently knows if the recorded pressure is too high. This process relies heavily on a concept called agentic vision. It is fundamentally different from traditional image recognition software. Agentic vision is truly the secret
sauce here today. The model successfully combines visual reasoning with real -time code execution. Yeah. It literally performs complex math on what it currently sees. Whoa. Imagine robots literally writing and executing code in real time just to understand the physical object they're looking at. That is an unbelievable leap in robotic cognitive ability. It really is. It is a fundamentally new paradigm for machine perception. It bridges the gap between digital logic and physical reality.
Two sec silence. Let's discuss its multi -view success detection capabilities now. A robot with a wrist cam and an overhead camera view. It can finally triangulate its operational success. Very reliably. Right. And it uses multiple camera angles to verify a completed task. It doesn't just assume it did the job correctly. It synthesizes multiple data streams to confirm the physical reality. It can verify that a blue pen is actually in a holder. Yeah. It could do this even if one
view is entirely blocked. It understands object permanence in a very human -like way. It knows an object exists even when completely hidden. It can now handle incredibly complex relational logic commands. You can instruct it to perform highly specific spatial tasks. Right. For example, moving every object that can fit inside the blue cup. That level of spatial awareness is incredibly difficult to program manually. The system has to understand volume, shape, and physical relationships
intuitively. The collaboration with Boston Dynamics is the real signal here. They are integrating Gemini ER 1 .6 into the orbit platform. By doing this, Google has effectively provided a prefrontal cortex. It acts as a brain for the world's most advanced hardware. Boston Dynamics makes some of the most capable robotic bodies out there. Now those agile bodies finally have a dynamic brain to match. We should also pivot slightly
to the microscopic physical world. Amazon just recently launched their new Amazon Biodiscovery platform. This is a major development for the global pharmaceutical industry. It takes this autonomous reasoning down to the cellular level. It is specifically designed to accelerate modern drug discovery. It can strategically create and test thousands of molecule ideas fast. Yeah. Early tests successfully generated 300 ,000 unique
medical antibodies. Generating 300 ,000 antibodies represents an unbelievable scientific scale. Humans could never physically test variations at that specific speed. That scale of physical world interaction is truly staggering today. It moves AI from digital abstraction into biological manipulation capabilities. It changes the entire timeline for developing life -saving medications. What used to take decades might only take a few months now. Does this visual reasoning mean robots
finally understand context? Well, they don't experience understanding the way humans naturally do, but they process enough variables to effectively mimic situational awareness. They don't feel context, but they can finally calculate it. Right. It is a mathematical approximation of actual human awareness. Let's step back and synthesize this entire deep dive. The overarching narrative arc today is actually very clear. AI is no longer just a clever little party trick. It's certainly
no longer just a simple chat window. It has completely outgrown its original digital boundaries now. It is touching every single aspect of our physical and corporate world. It is the absolute center of ruthless corporate warfare today. Yeah. Companies are battling for dominance over... massive platform distribution channels. It is actively demanding its own employee login in our software. It wants a dedicated seat at the modern enterprise table. Exactly. And it is finally getting the necessary
digital prefrontal cortex. It needs this dynamic brain to manipulate the tangible physical world. It is even beginning to actively manipulate our microscopic biology. The evolution from text generation to physical action is staggering. I mean, if an AI can get its own employee seat in a Microsoft database. and a robotic body with a prefrontal cortex from Google, at what point will our legal definition of a worker have to fundamentally change? Thank you for joining us
on this deep dive. We hope this exploration gave you some valuable new perspectives. Out to row music.