Practical AI · Live Show Flow · Conversational

Ep46 Live Flow

The news block as a two-host conversation. Read from this on air. Time codes are guides, not hard cues. Every beat: say the plain news first, then riff.
The thread for the whole block
We treat our tools as permanent — you buy the hammer, it's yours. This week that illusion shattered. The government remotely switched off the best AI model. An AI-and-rocket company is buying the coding tool. The market leader is bleeding billions. Power is consolidating at the very top — and the one move that protects you is portability.
00:00 · Cold open / CallbackWe tracked when Anthropic's best model would open up. This week the government turned it off.
📰 The news — say this first
For months we've followed Anthropic's most powerful models. Last week Fable 5 finally went public. This week the US government ordered it shut down. On June 12 the Commerce Department issued an export-control directive, and Anthropic disabled both Fable 5 and Mythos 5 worldwide.
Why open here
It's the biggest story of the week and it's our own thread paying off. Chris called it on Ep40: "I'm bearish on anyone who gatekeeps models." Now there's a real fight over who the gatekeeper actually is — and both sides have a case.
Quick frame before the news. We've spent episodes tracking when Anthropic's most powerful model would open up. Last week, it finally did. This week, the US government turned it off.
And remember what I said back on Ep40 — I'm bearish on anyone who gatekeeps these models. I just didn't think the gatekeeper would be Washington.
And there are two very different stories about why. The White House says Anthropic left them no choice. Anthropic says Washington overreached. We'll give you both. That's where we start.
Producer note (backstage, not on air)
Thematic callback to Chris's Ep40 gatekeeping thesis — logged in the predictions log, stays open. Keep it quick, roll into Beat 1.
~01:45The government banned Claude's best models — and both sides are fighting over why.
📰 The news — say this first
On June 12, the Commerce Department's Bureau of Industry and Security issued an export-control directive citing national security, ordering Anthropic to cut off Fable 5 and Mythos 5 for foreign nationals. You can't reliably geofence that, so Anthropic disabled both models for every customer worldwide. The government's reason, per White House AI czar David Sacks: a trusted partner found a jailbreak of Fable's guardrails and, asked to fix it, Anthropic refused — so the ban was a last resort. Anthropic disputes that, calling the flaw narrow and the process rushed. Then 100+ cybersecurity experts — Alex Stamos, Katie Moussouris, Rachel Tobac — signed an open letter at freefable.org saying the ban backfires.
The frame that lands
Two real stories collide: the government says Anthropic wouldn't fix a dangerous hole, Anthropic says it was a tiny flaw and a rushed overreach. Either way, defenders lost their best engine and the attackers never used it.
Let me be clear up front — this is a politically charged government action, and we're going to give you both sides and let you decide. Operationally it was an earthquake: the government ordered Anthropic to cut off Fable 5 and Mythos 5 for foreign nationals, and since you can't verify the citizenship of every API call, they pulled the plug for everyone, everywhere.
Now, the government's side, and this matters. According to David Sacks — he's the White House AI czar, and yes, one of the All In guys — a trusted partner was testing Fable and found a jailbreak. Fable is the safer model, guardrails layered on top of the more powerful Mythos. Break the guardrails and you've exposed serious cyber capabilities to people who shouldn't have them.
And Sacks says they went to Anthropic first. The administration asked them to fix it or pull it, and he says Dario Amodei refused — said the jailbreak wasn't a serious risk. So in their telling, the export control was a reluctant last resort, not the first move.
Anthropic tells it differently. They say the jailbreak was narrow — basically asking the model to review code for flaws, something other top models do too — not a reason to recall a tool used by hundreds of millions of people. And they say the process was rushed and opaque.
Then the cybersecurity world came unglued. Over a hundred heavy hitters — Stamos, Moussouris, Tobac — signed an open letter saying the ban backfires no matter who's right.
Their logic is hard to argue with. Defenders use these frontier models to parse malware, simulate attacks, automate patching. Turn them off and you take the best engine away from the good guys — while the ransomware gangs and nation-states keep running open-weight models on their own servers. Nobody switches those off.
So three camps: the government says safety, and that Anthropic wouldn't cooperate. Anthropic says overreach on a narrow flaw. The security pros say it backfires either way. We're not telling you who's right.
But here's the part that touches you. The assumption we all live by — that the best model is always one tab away — just broke. Access is now a policy lever.
JUNE 12 · COMMERCE/BIS DIRECTIVEFABLE 5 + MYTHOS 5 OFF WORLDWIDEGOVT (SACKS): JAILBREAK FOUND · FIX REFUSEDANTHROPIC: NARROW FLAW · RUSHED100+ SIGNATORIES · FREEFABLE.ORG
Why this matters · for your audience
The best AI tool a defender could reach for vanished overnight by government order. Don't build a critical workflow on a single model you can't replace. Keep one real task tested across two providers so a sudden cutoff — commercial or political — is an inconvenience, not a crisis.
Producer note (backstage, not on air)
Sourcing for the govt side: David Sacks on X, June 13, corroborated by Fortune / Politico / Bloomberg (Amazon found the jailbreak; "last resort after hours asking Anthropic to work with them"). "Refused" is Sacks's word — Anthropic frames it as "we disagreed it was serious." Attribute to Sacks / the White House, don't state it as flat fact. Anthropic's side = their official statement at anthropic.com/news/fable-mythos-access.
~05:00Same week, the mirror image: China's DeepSeek raised $7.4B.
📰 The news — say this first
In the exact week the US forced its best model offline, China's DeepSeek raised roughly $7.4 billion — its first outside round, making it China's most valuable AI startup. China took 76% of all AI funding globally this week (DeepSeek alone was about 64% of it).
The contrast
The US legally throttled its frontier. China poured capital into theirs. Nature abhors a vacuum.
Here's the timing that should stop you. The same week the US pulled its best model offline, China's DeepSeek raised about seven point four billion dollars.
In a single week. China took 76% of all AI funding on the planet — DeepSeek alone was roughly two-thirds of that. While we throttle access, they accelerate.
We'll put the full money picture in the funding segment at the end. But the headline here is geopolitical, not financial — when you restrict your own companies, the only players who can fill the void are nation-states and spacefaring mega-corporations.
Producer note (backstage, not on air)
CORRECTED from the prep audio: it's China 76%, not "DeepSeek 76%." DeepSeek alone ≈ 64%. Full funding numbers are the closing segment — keep this beat short, it's the bridge.
~06:00The SpaceX week: a trillionaire, and a $60B grab for the coding tool.
📰 The news — say this first
SpaceX's IPO raised about $75 billion, making Elon Musk the world's first trillionaire (~$1.1T on paper). By June 16 its market cap hit $2.65 trillion — briefly bigger than Amazon. Same week, SpaceX agreed to buy Anysphere, the maker of Cursor — the AI coding tool a lot of developers live in — for $60 billion, all stock (announced June 16, expected to close in Q3). And here's the key context: SpaceX absorbed xAI, Elon's AI company, earlier this year, so Grok is now its in-house AI division. This is an AI company buying the top AI coding tool.
The "yeah but"
Why does a "rocket company" buy a code editor? Because it isn't one anymore — SpaceX now owns xAI and Grok. This is an AI company buying the most-used AI coding tool, and the landlord just changed.
The SpaceX IPO raised around seventy-five billion. That made Elon Musk the world's first trillionaire, on paper, and by June 16 SpaceX briefly passed Amazon by market cap.
And it immediately moved. That same week SpaceX agreed to buy Anysphere — the maker of Cursor, the coding tool a huge chunk of developers rely on — for sixty billion in all stock. The deal's announced, expected to close next quarter.
Why is a rocket company buying a code editor?
That's the trick in the question. It's not really a rocket company anymore. SpaceX swallowed xAI, Elon's AI company, earlier this year, and Grok is now its in-house AI division. So this is an AI company buying the most popular AI coding tool there is.
So what do they do with it?
Officially, pair Cursor with Grok and their supercompute and aim straight at OpenAI and Anthropic. Reading between the lines: Grok becomes the default engine inside Cursor, they build coding agents on it, and every keystroke from millions of developers trains a better model. Model, compute, distribution, and data, all in one house.
Which makes "the landlord changed" real. If you live in Cursor, your AI and your code may now run through Elon's stack. Know where you'd move tomorrow.
IPO ~$75BMUSK ~$1.1T · FIRST TRILLIONAIRESPACEX ~$2.65T · BRIEFLY > AMAZONCURSOR · $60B ALL-STOCK · AGREED, CLOSES Q3SPACEX OWNS xAI/GROK (SINCE FEB 2026)
Why this matters · for your audience
The tool you build on can be bought out from under you — and the new owner may run it on its own AI model. Spend 20 minutes confirming you could move your work elsewhere. Portability is the hedge against consolidation.
Producer note (backstage, not on air)
CORRECTED from the prep audio: Anysphere makes Cursor, NOT Fyxer. (Fyxer is the tool that recorded the prep audio — that's where the mix-up came from.)
Producer note (backstage, not on air)
Accuracy: say "agreed to buy" / "is buying," NOT "bought" — announced June 16, expected to close Q3 2026 (Reuters, Bloomberg, Guardian). SpaceX acquired xAI in Feb 2026 and folded it in as its AI division (Grok) in May — that's why this is an AI play, not a rocket company's whim. The Grok-default / coding-agents / training-data angle is our analysis, framed as "reading between the lines," not officially announced. Skip the "founders each made ~$2.7B" line (rumor).
~08:00Bezos's Prometheus: a $12B bet to engineer the physical world.
📰 The news — say this first
Jeff Bezos's startup Prometheus (co-run with ex-Google X's Vik Bajaj) closed a $12 billion Series B at a ~$41 billion valuation. The goal: an "artificial general engineer" — AI that designs physical things (jet engines, chips, EV chassis, drugs), compressing decades of R&D into weeks.
The shift
For a century the bottleneck to innovation was human design time. If AI removes that, the leverage moves to whoever owns the factories, the materials, and the regulatory approvals.
This isn't a chatbot. Prometheus raised twelve billion at a forty-one-billion valuation to build what they call an artificial general engineer — AI that designs physical matter. Jet engines, chips, pharmaceuticals.
They're training on fluid dynamics, stress tensors, molecular folding — simulate millions of combinations in a digital twin, then build only the best few. Decades of R&D compressed into weeks.
And here's the uncomfortable question. If AI can engineer a better jet engine, does the power leave the brilliant human engineers and go to whoever owns the titanium factory and the lawyers who get it approved?
That's the structural shift. Remove the design bottleneck and the only friction left is physical capital and legal permission. "I work with real things, so I'm AI-proof" just got a twelve-billion-dollar challenger.
PROMETHEUS · $12B SERIES B~$41B VALUATION"ARTIFICIAL GENERAL ENGINEER"
~10:30The king is slipping: ChatGPT below 50%, and a $21B loss.
📰 The news — say this first
Two OpenAI leaks. One: per Sensor Tower, ChatGPT fell below 50% of the AI-assistant market for the first time — about 46.4%, with Gemini ~27.7% and Claude ~10.3%. Two: leaked audited financials (Ed Zitron, verified by the Financial Times) show a 2025 operating loss of ~$21 billion on $13.07B revenue — which actually tripled year over year.
The frame
Falling below 50% is the psychological line: the default is no longer the majority. The brand moat is evaporating — while the leader burns cash to defend it.
The monolith is showing cracks. Sensor Tower says ChatGPT dropped below fifty percent of the assistant market for the first time ever — about 46 percent. Gemini's at 28, Claude's at 10.
And the money leak is brutal. Ed Zitron got leaked audited financials, the Financial Times verified them — a twenty-one-billion-dollar operating loss in 2025.
On thirteen billion in revenue that tripled. So how does that math ever work?
It's a subsidized land grab. Gigawatt data centers, subsidized compute on billions of API calls, licensing fees to publishers. Every user costs them money. Tripling revenue is real, but the costs scale right alongside it.
Which means blind loyalty to ChatGPT as "the AI" is a strategic error. Pick your AI per task, not per brand.
CHATGPT 46.4% · FIRST TIME <50%GEMINI 27.7% · CLAUDE 10.3%OPENAI ~$21B OPERATING LOSS (2025)REV $13.07B · TRIPLED YoY
Why this matters · for your audience
The market is splitting three ways — more competition, faster gains. Run the same real task through ChatGPT, Gemini, and Claude once and pick per job. The "obvious winner" assumption is exactly what broke this week.
Producer note (backstage, not on air)
Say "operating loss" — the $21B is the operating loss; net loss is larger (~$38.5B, includes a one-time conversion charge). Market share = Sensor Tower (third-party). Financials = leaked/FT-verified.
~12:50Quick hits: AI is moving into the tools you already use.
📰 The news — say this first
Three fast ones. Grok is now free inside Microsoft PowerPoint — research and build decks without leaving the app. Meta put AI Mode inside the Facebook feed. And Xiaomi released an open-source model, MiMo, that reportedly beats Claude Code on long-horizon (200+ step) coding jobs.
The signal
When a phone company's open-source model competes with Anthropic on 200-step coding, frontier intelligence is no longer a scarce resource. It's getting embedded everywhere and commoditized.
Quick hits. One — Grok is now native inside PowerPoint. Research, charts, full slide decks without switching tabs. Two — Meta folded AI Mode right into the Facebook feed.
And three, the wild one — Xiaomi, the phone company, dropped an open-source model called MiMo that reportedly beats Claude Code on 200-step coding tasks. When a smartphone maker is competing with Anthropic, the underlying intelligence isn't scarce anymore.
GROK IN POWERPOINT (FREE)META AI MODE IN FACEBOOKXIAOMI MiMo · BEATS CLAUDE CODE (REPORTED)
Producer note (backstage, not on air)
Xiaomi = self-reported benchmark, scoped to long-horizon agentic coding. Say "reportedly," don't say "beats Claude" flat.
~14:20Your website becomes a toll booth: AWS lets you charge AI bots.
📰 The news — say this first
This week AWS added AI traffic monetization to its Web Application Firewall. For the first time, publishers can detect AI bots and charge them to read your content — right at the network edge, no custom code. A toll booth for crawlers.
The second-order story
The web is splitting in two: a free layer for humans, a paid layer for agents. But who can afford the toll? It may pull the ladder up behind the giants.
For years the fight over AI scraping your content lived in courtrooms and angry threads. This week AWS turned it into a product — you can now charge AI bots to read your site, right at the edge, with a toggle.
It's an E-ZPass toll booth on your front lawn. And it reshapes the whole economy of the web — a free layer for human eyeballs, a paid layer for the agents.
Isn't that what creators have been begging for? Finally getting paid when AI trains on their work?
Partly. But who can afford the toll? The open-source community and small research startups can't pay AWS. If every good site charges a microtransaction, only the trillion-dollar empires can afford to harvest the open web. You solve copyright and accidentally pull the ladder up behind the incumbents.
AWS WAF · AI TRAFFIC MONETIZATIONCHARGE BOTS AT THE EDGE · NO CODE
Why this matters · for your audience
If you own any digital real estate, you now have a stance to pick: invite AI crawlers in (GEO), block them, or bill them. Knowing your position is ahead of most.
~18:15 · RecapThe week in one breath.
So, fast recap. Power consolidated hard at the top — the government shut down frontier models, SpaceX — now an AI company — is buying the coding tool, Bezos is funding an engine to build the physical world.
But the center isn't holding. ChatGPT dropped below half the market, OpenAI's bleeding twenty-one billion, and an open-source phone-maker is catching up. Volatility and commoditization at the same time.
And the tools to protect your own turf — like the AWS bot toll booth — are finally real. Your homework: 20 minutes confirming you could move your most critical workflow to a second provider. Portability is the defense.
Provocative close → hands to the deep dive
The most-bookmarked idea on X this week was a rallying cry: "stop prompting, start building loops." That's our deep dive today. Because if your website is now a toll booth for bots, and people are building agents that surf the web for them — are we about to become the minority of internet traffic? A digital tourist in a house built and run by bots. Let's get into it.
DEMO · after the deep diveTalk to your website, and it builds you a private team tool.
Lead here · the promise
The deep dive just said agents are taking over the web. Here's you putting them to work on your own site. Describe the tool your business needs, and your website builds it. No developer. This is PageMotor.
🎬 Run order
Slides 1–4 → click into the live demo on the admin side → Slide 5 closer.
Slide 1 · Hook. Describe the tool your business needs. Your website builds it. No developer, no new software.
Slide 2 · Menu of dreams. The tool you keep wishing you had: a team scoreboard, a client portal, an internal tracker, a page that gets you recommended by AI agents.
Slide 3 · The reveal. Your website has a private side, not just public pages. Your data lives in the back room, never on the public internet.
Slide 4 · Proof. Here's one, real and running — my team's scoreboard on the private side, refreshes itself. → Click into the live demo now.
Slide 5 · Closer. What would you build for your business? That's PageMotor.
Live demo · log in FIRST
Demo (anonymized, safe): revenuehire.com/admin/daily-recruiter-scoreboard-demo/. Log into revenuehire.com/admin/ first in the browser you screen-share, THEN open the demo link, or it shows the login wall instead of the board.
Producer note (backstage)
Slides file: Revenue Hire/Operations/Dashboard/scoreboard-show-slides.html (arrow keys to advance). Demo uses fake names only — zero real client/recruiter/candidate identities (verified). Ties straight into the deep dive's "site for humans vs site for agents" thread.