Latest / AI Fire Daily / 🎙️ EP 299: $500M Biotech to Fight Common Cold & Codex Rewrites the Rules of Work
Transcript
- 0:00You know, you probably think of AI as purely
- 0:02digital, just code running on a distant server.
- 0:06Right. Totally detached from biology. Yeah, exactly.
- 0:09But right now, the paradigm is violently shifting.
- 0:13It really is. Tech billionaires are stepping
- 0:15away from just training massive digital models.
- 0:18Instead, they are actively trying to hack fundamental
- 0:21human biology. It is wild. They want to kill
- 0:23the common cold. I mean, it sounds completely
- 0:26surreal. Yeah. but it is actually happening right
- 0:29now. Welcome to the Deep Dive. I'm really glad
- 0:32you are here. Today, we are unpacking a truly
- 0:34strange reality. Yeah, we have a lot of ground
- 0:37to cover. We absolutely do. We are going to explore
- 0:39this massive biotech pivot from AI giants. Then
- 0:43we will look at how the government is throttling
- 0:46frontier AI rollouts. The policy shifts are pretty
- 0:49intense right now. They are. And finally, we
- 0:52have some leaked internal data from OpenAI. Oh,
- 0:55this part is fascinating. Data proving. that
- 0:57the very nature of human work is shifting. Autonomous
- 1:00agents are actually doing your work while you
- 1:02sleep. I am genuinely pumped for this first topic,
- 1:04though. Yeah, it is fascinating. So let us start
- 1:08with this biotech move. You have some of the
- 1:10biggest players in AI jumping ship. Right. Anthropix,
- 1:15Stripe, and the OpenAI Foundation just backed
- 1:18something called Intercept. Yeah, Intercept is
- 1:21this $500 million nonprofit initiative. Wow.
- 1:26And. Their goal is staggering. They want to make
- 1:30routine illnesses completely obsolete. They are
- 1:33specifically targeting the common cold and the
- 1:36flu. Exactly. They want to eradicate them entirely.
- 1:38OK, let me push back here for a second. Sure,
- 1:40go for it. $500 million sounds like a massive
- 1:43amount of capital to us. It does. But in the
- 1:45world of biology, isn't that just a drop in the
- 1:47bucket? I mean, Pfizer spends tens of billions
- 1:49of dollars on research every year. Oh, absolutely.
- 1:51So why do these tech guys think half a billion
- 1:54will solve the common cold? That is the exact
- 1:56right. question to ask. And it really comes down
- 1:59to their specific strategy. They are not trying
- 2:02to fund the entire pharmaceutical pipeline. They're
- 2:05just trying to aggressively de -risk. The earliest
- 2:08stages. OK, so explain how that mechanism actually
- 2:11works. Well, traditional pharma hates early stage
- 2:13risk. Because it's too expensive to fail. Exactly.
- 2:16It is a massive capital drain. Yeah. So Intercept
- 2:19is stepping in to fund that dangerous initial
- 2:22research. Got it. They are targeting broad spectrum
- 2:25shots and specialized nasal sprays. They are
- 2:29also developing pills that block. dozens of respiratory
- 2:32viruses simultaneously. So it is kind of like
- 2:35they are paying for a highly dangerous scouting
- 2:38mission. Yeah, that is a good way to put it.
- 2:40They send a small, agile team into the jungle
- 2:43to clear a path. Once that path is proven perfectly
- 2:46safe, traditional pharma drives the heavy machinery
- 2:49right through it. That is a brilliant analogy,
- 2:52yes. It is like stacking Lego blocks of data.
- 2:55Right. They build the precarious foundation.
- 2:57And once the technology is demonstrably de -risked,
- 3:00the dynamic completely flips. Traditional pharma
- 3:03companies finally feel safe enough to step in.
- 3:05Exactly. They take over for that final, incredibly
- 3:08expensive stretch of human trials. They snap
- 3:10the final pieces into place. It is a really clever
- 3:13application of startup economics. They are not
- 3:16just treating the symptoms. They are trying to
- 3:18build a biological shield. Right. And they are
- 3:20even funding advanced indoor air cleaning tech.
- 3:23Oh, really? Yeah. They want these systems deployed
- 3:26in offices and schools nationwide. But, you know,
- 3:29it still makes you wonder why AI giants specifically
- 3:32care. Well, it is all about efficiency. Right.
- 3:35But why do they want to step out of the software
- 3:37realm at all? Why cure the cold? Because of raw,
- 3:40undeniable productivity loss. According to Intercept's
- 3:44data, people lose 15 to 25 days every single
- 3:48year. Wow. Yeah, you lose nearly a month of focus
- 3:51to routine respiratory infections. That is a
- 3:53staggering bottleneck for human progress. It
- 3:56absolutely is. And when you zoom out, the math
- 3:59gets even crazier. How so? That lost time translates
- 4:02to a $600 billion hit. A year? Yes. That is an
- 4:06annual loss to global economic productivity.
- 4:09So why do AI companies actually care about curing
- 4:12the cold? They need humanity operating at absolute
- 4:15peak efficiency to scale their own technological
- 4:17growth faster. So they are upgrading human hardware
- 4:21to maximize global output. Precisely. That is
- 4:25wild. So these AI giants are trying to optimize
- 4:28our biology. But while we wait for those physical
- 4:31upgrades, the government is stepping in. Oh,
- 4:33yeah. Big time. They are already throttling the
- 4:36software side of things. The pulse of the industry
- 4:38is definitely shifting toward heavier regulation
- 4:41right now. Look at what is happening with GPT
- 4:43-5 .6. Yeah. The White House is reportedly intervening
- 4:47with a very specific request. Right. They do
- 4:50not want an immediate, massive. public launch.
- 4:53They want open AI to release it incredibly slowly.
- 4:56Which makes sense from a policy standpoint. Yeah.
- 4:58They are asking for a measured customer by customer
- 5:01rollout. It is a clear attempt to monitor the
- 5:04impact in real time. You know, they want to contain.
- 5:07any potential fallout from these frontier models.
- 5:10Just strictly observing the facts here, it is
- 5:12a very cautious approach. It is. But while we
- 5:14wait for that slow rollout, current models are
- 5:17quietly improving. Oh, definitely. GPT -5 .5
- 5:20instance just got a major silent upgrade. It
- 5:23really did. It is handling detailed multi -step
- 5:26instructions much better now. Yeah, it is also
- 5:29excellent at shopping picks and mapping local
- 5:31recommendations. And honestly, OpenAI even stated,
- 5:35It is simply much more fun to talk to. Right.
- 5:39But the really massive news is happening away
- 5:42from the legacy giants. Yeah, the startup space
- 5:44is where the real heat is. We need to talk about
- 5:46what is happening there. Look at General Intuition.
- 5:49Oh, man. Their recent news is huge. They just
- 5:52hit a massive fundraising milestone. 320 million
- 5:57dollars raised yeah and they are sitting at a
- 5:592 .3 billion dollar valuation now and their training
- 6:03mechanism is what makes them so valuable it is
- 6:05entirely unique they are not just feeding it
- 6:07text from the internet no not at all they are
- 6:10taking a completely different path so what are
- 6:12they doing they train their ai on hundreds of
- 6:14millions of hours of gameplay data beat Whoa.
- 6:18Yeah. I mean, imagine training an AI on hundreds
- 6:21of millions of hours of gameplay. It is literally
- 6:23learning physics through play. Right. And that
- 6:26fundamentally changes how the model maps reality.
- 6:28How so? Well, text models just predict the next
- 6:31logical word in a sentence. But general intuition
- 6:33is predicting how objects actually interact.
- 6:36Why? They are teaching the AI deep spatial reasoning.
- 6:40It is predicting cause and effect in a three
- 6:42-dimensional space. Exactly. It has to understand
- 6:45gravity. momentum, and physical obstacles. So
- 6:48how does spatial reasoning change what an AI
- 6:50can actually do? It allows the system to successfully
- 6:54navigate unpredictable, messy, physical environments.
- 6:58It acts. It turns the AI from a talker into a
- 7:01physical doer. Exactly. OK, so we just talked
- 7:03about these massive models developing spatial
- 7:05reasoning. We looked at billions in funding.
- 7:08Right. But how does that power actually shrink
- 7:10down? I mean, how does it live inside the apps
- 7:13you use every single day? This is where the daily
- 7:15friction of work is actually getting solved.
- 7:17Let us unpack the Claude Tagg debate first. Okay.
- 7:20Former Tesla AI director Andrzej Karpathy just
- 7:24called it a brand new paradigm. Yeah, he praised
- 7:27how it completely flips the interaction model.
- 7:30You do not go to a chatbot anymore. Right. You
- 7:32tag the AI right inside your existing workspace.
- 7:35Exactly. It comes to you. But I have to push
- 7:37back a little here. Sure. Others in the community
- 7:39are calling this just another wave of UI hype.
- 7:42Is it really a new paradigm or just a new button?
- 7:45It is definitely more than a button, mostly because
- 7:48of the shared context. What do you mean? Look
- 7:50at a recent tutorial we saw about Slack integrations.
- 7:53It showed how teams use Claude directly inside
- 7:57a Slack channel. Oh, right. The AI reads the
- 8:00entire on. conversation between human employees
- 8:03it essentially becomes a shared second brain
- 8:06for the team exactly it synthesizes the team's
- 8:09messy human dialogue into actionable structure
- 8:12automatically speaking of Claude there's a fascinating
- 8:15leak recently yes The Fable 5 leak. Right, in
- 8:18Cloud Code version 2 .1 .190. People dug into
- 8:22the code and found hidden strings. And the strings
- 8:24hint that Fable 5 is coming to normal subscription
- 8:27tiers, but it is going to have strict weekly
- 8:30usage limits. Yeah, the compute costs for these
- 8:33frontier models are still astronomical. Weekly
- 8:35limits are the only way they can prevent massive
- 8:37server burns. Meanwhile, the enterprise giants
- 8:39are definitely not sitting still. No, Microsoft
- 8:42is moving fast. They just pushed a major architectural
- 8:44update for Copilot in Excel. They introduced
- 8:47a highly agentic feature they simply call skills.
- 8:49So how does an Excel skill actually work in practice?
- 8:54Well, it moves Excel away from being a static
- 8:57grid of numbers. Okay. Skills allow you to build
- 8:59repeatable, autonomous workflows right in the
- 9:02spreadsheet. It uses built -in financial connectors
- 9:05to pull live data continuously. And it has improved
- 9:08traceability so you can actually audit its math.
- 9:11Right. Google is also pushing hard. specifically
- 9:13into the education space yeah the gemini app
- 9:16just launched free study notebooks worldwide
- 9:18this takes any complex topic and autonomously
- 9:21builds an interactive learning space right it
- 9:23generates quizzes and interactive diagrams tailored
- 9:26to how you learn best the enterprise tools are
- 9:29evolving so fast but the independent api level
- 9:32tools are moving even faster They really are.
- 9:35Look at a platform like Oxlo Point AI. This one
- 9:38is wild. It gives you access to over 35 different
- 9:41frontier models. Yeah, you get DeepSeek V4 Pro,
- 9:44Kini K2 .6, GLM5. Kuhn, Lama, Mistral, all through
- 9:50one unified API. The mechanism here is intelligent
- 9:53routing. How does that work? You send a prompt,
- 9:57and Oxlo analyzes the complexity of your request.
- 10:00Then it automatically routes your task. to the
- 10:03specific model best equipped to handle it. It
- 10:05optimizes for speed and cost without you doing
- 10:07anything. Exactly. Then there's Xero. This tool
- 10:10is built specifically for our chaotic modern
- 10:12workflows. Xero is incredible. It maps your scattered
- 10:15brain across different applications. Yeah. It
- 10:17reads your Slack messages, your chaotic Gmail
- 10:20threads, your random Notion documents. Honestly,
- 10:23I still wrestle with prompt drift myself. Oh,
- 10:25we all do. Just trying to manually get an AI
- 10:27to organize my scattered notes is exhausting.
- 10:30Xero removes that cognitive load. It synthesizes
- 10:34your intent from the background context and actually
- 10:36builds working software out of it. It keeps itself
- 10:38updated automatically as your Slack conversations
- 10:40evolve. It is brilliant. Another really fascinating
- 10:44breakthrough is Browser Act. This is built for
- 10:47agents navigating the open web. Web browsing
- 10:49has always been notoriously difficult for AI
- 10:51agents to handle. Right. Websites block automated
- 10:54scrapers. Pop -ups ruin the navigation path.
- 10:58So many obstacles. So how does Browser Act actually
- 11:00bypass those blocked pages safely? It uses advanced
- 11:04vision models to literally see the screen like
- 11:07a human. Wow. It mimics human cursor movements
- 11:10and solves visual captchas natively. It adapts
- 11:14to real, messy scenarios without breaking the
- 11:16workflow. And finally, for designers, we have
- 11:19Figma Motion. This changes the entire design
- 11:21canvas. Yeah. The motion variables live directly
- 11:24alongside your team and component libraries.
- 11:27Your static designs can literally come to life
- 11:29with physical physics from day one. It is a game
- 11:31changer. But with all these different agentic
- 11:33platforms integrating deeply into our daily routines...
- 11:36What is the biggest risk of having so many empowered
- 11:38tools? You might easily spend more time managing
- 11:42your tools than doing the actual deep work. We
- 11:45risk becoming exhausted middle managers to our
- 11:47own software. Exactly. We're going to take a
- 11:50quick break for our sponsor here. Stick around.
- 11:52You do not want to miss the second half. And
- 11:55we are back. So we have looked at the theoretical
- 12:00tools. We understand how the integrations actually
- 12:02work. Right. But how is this changing your actual
- 12:05job right now? This is where the story shifts
- 12:07from theory to reality. We have a massive new
- 12:10leaked research paper to look at today. It analyzes
- 12:13OpenAI's own internal usage data. And it is revealing.
- 12:18It tracks exactly how individuals and organizations
- 12:20used Codex between August 2025 and June 2026.
- 12:25The expanding horizons of this autonomous work
- 12:28are genuinely shocking. Look at the sheer volume
- 12:31of work being delegated. By May 2026, over 70
- 12:34% of individual users deployed Codex for major
- 12:37tasks. These are not quick email drafts. No.
- 12:39These are complex tasks estimated to take a human
- 12:42over an hour to complete. But the next metric
- 12:44is what really broke my brain. Yeah, this is
- 12:46the wild one. A massive 25 .6 % of users went
- 12:50much further. They ran requests that exceeded
- 12:53eight hours of human labor. They triggered an
- 12:55entire workday of execution with a single prompt.
- 12:59And the demographic driving this massive explosion
- 13:02in usage is really surprising. It is not the
- 13:05software engineers driving this growth. Who is
- 13:07it? It is the non -developers. Since August 2025,
- 13:11non -developer adoption completely exploded.
- 13:14Oh, big time. It grew 137 times for individual
- 13:17users, and within external organizations, it
- 13:21grew 189 times. The non -technical enterprise
- 13:24landscape is migrating to these tools at blinding
- 13:27speed. Look at OpenAI's own internal operations
- 13:30as a leading indicator. For their internal teams,
- 13:33standard chatbots are officially dead. Codex
- 13:36has entirely replaced chat as their primary interface.
- 13:39By April 2026, the non -technical departments
- 13:42completely crossed the threshold. Yeah. Finance
- 13:44hit 91 % Codex usage. Recruiting reached 89%.
- 13:48And legal hit 88%. The traditional boundaries
- 13:50between these corporate roles are completely
- 13:52blurring away. They really are. Over a quarter
- 13:54of the work done by these business teams radically
- 13:57changed. Right. How exactly did it change? Like,
- 14:00what are they doing? Well, these non -technical
- 14:02teams are using codecs to do deep technical engineering.
- 14:05Really? Yes. They are executing complex coding.
- 14:09heavy automation, and advanced data transformation.
- 14:12So you have a recruiter building their own custom
- 14:15software automation. Exactly. And you have a
- 14:17finance manager writing complex data pipelines
- 14:19without knowing Python. Exactly. The AI acts
- 14:23as the translator between their business intent
- 14:25and the machine code. Wow. And this leads to
- 14:29a specific group the report calls the power users.
- 14:32The top 1 % of active users? Right. These power
- 14:35users regularly orchestrate multiple parallel
- 14:38agents at the exact same time. They are essentially
- 14:41running their own autonomous digital departments.
- 14:43They are. And because these agents run in parallel,
- 14:46the math gets wild. How wild? These users accumulate
- 14:49over 60 hours of autonomous runtime in a single
- 14:52day. Two sec silence. 60 hours. In a single 24
- 14:57-hour day. It literally breaks the physical limits
- 15:00of human time. As these agentic tools become
- 15:03ubiquitous, the standard expectation shifts.
- 15:06Completely. The basic unit of knowledge work
- 15:09is fundamentally changing. We are moving away
- 15:12from real -time chatting and prompt engineering.
- 15:14Yes. We are moving entirely toward asynchronous
- 15:17delegation. That is the crucial transition. It
- 15:20changes everything about human labor. Let us
- 15:23define that specific jargon for a second. asynchronous
- 15:27delegation giving the AI a complex job and walking
- 15:31away while it works yes you provide the high
- 15:33-level intent and you completely detach from
- 15:35the execution right you only return hours later
- 15:38to review the final completed output so what
- 15:41happens to traditional job titles if non developers
- 15:44are doing the coding Titles will describe the
- 15:46specific problem you solve, not your keyboard
- 15:48skills. You will be judged entirely on outcomes,
- 15:51not technical execution. Precisely. So if we
- 15:54pull all of these disparate threads together,
- 15:56from curing the common cold to agents coding
- 15:59while you sleep. What is the fundamental shift
- 16:02you are watching unfold here? We are witnessing
- 16:05the absolute death of the AI chatbot. Yeah. We
- 16:08are moving from an AI that just politely answers
- 16:10your questions. We are moving to an AI that acts
- 16:13as an autonomous worker. It is funding cures
- 16:16for biological diseases. It is actively navigating
- 16:19blocked web pages. It is doing eight hours of
- 16:22complex data pipelines while you sleep. The era
- 16:25of the digital assistant is gone. The era of
- 16:27the digital employee has arrived. It really forces
- 16:30you to change how you view your own time. I want
- 16:34to leave you with a final thought today. It is
- 16:36a good one. Something to deeply mull over as
- 16:38you look at your own workflow. If a digital agent
- 16:40is clocking 60 hours of autonomous runtime for
- 16:43you in a single day, at what point do you stop
- 16:46being an employee and start being a manager of
- 16:49digital experts? That is the big question. Thank
- 16:52you for joining us on this deep dive. Tomorrow
- 16:54morning, take a hard look at your daily task
- 16:55list. Yeah, evaluate it. Ask yourself which tasks
- 16:58take one hour and which take eight. Then honestly
- 17:01ask yourself if an agent could just do them for
- 17:03you. We will see you next time.