This week power consolidated hard at the top: the government switched off the best AI model, a rocket company bought the coding tool, the leader is bleeding billions. The news is what's happening TO you. The three mini deep dives are what you can do about it: build loops, protect your brain, and build a dashboard you can talk to.
SAY IT RIGHT ON AIR:China took 76% of AI funding, NOT DeepSeek (DeepSeek alone ≈ 64%). Anysphere makes Cursor, NOT Fyxer. It's Katie Moussouris. OpenAI's $21B is the OPERATING loss (net is bigger). ChatGPT <50% = Sensor Tower (reported). Xiaomi "reportedly" beats Claude Code on long coding. On the brain dive: "AI brain fry" is a survey term, not a diagnosis; burnout IS WHO-recognized. Keep all client names OFF on the loops story.
ON THE BAN — GIVE BOTH SIDES. Govt side = David Sacks (White House AI czar, also an All In host), on X June 13: a trusted partner (Amazon) found a jailbreak of Fable's guardrails; Anthropic/Amodei was asked to fix-or-pull and refused, so the export control was a reluctant last resort (corroborated by Fortune / Politico / Bloomberg). Anthropic's side (official statement): the jailbreak was narrow ("review code for flaws," exists in other models too), not worth recalling a model used by hundreds of millions, and the process was rushed. "Refused" is the govt's word; Anthropic frames it as "we disagreed it was serious." Tell it as a fight, not a villain.
0 · Cold Open
The Callback — they pulled Claude's best model, and both sides are fighting over why
10-second callback, then into news. Tee up the fight, don't pick a villain.
We tracked for months when Anthropic's most powerful model would open up. Last week Fable 5 finally did. This week the government forced it offline. The White House says Anthropic was warned about a security hole and wouldn't fix it; Anthropic says the flaw was tiny and the move was rushed. Callback to Chris Ep40: "I'm bearish on anyone who gatekeeps models" — now there are two sides arguing over who the real gatekeeper is. "Last week it opened. This week it's gone. And nobody agrees on why. That's where we start."
1 · News Block
6 stories + 3 quick hits
Say the news first, then riff. ~20 min. Lead the ban.
The Fable 5 / Mythos ban (lead) — govt directive June 12. BOTH sides: White House (David Sacks) says Anthropic was asked to fix a jailbreak and refused, so the ban was a last resort; Anthropic says the flaw was narrow and the process rushed; 100+ security pros sided with Anthropic.
The SpaceX week — bought Cursor $60B, Musk first trillionaire, briefly > Amazon.
ChatGPT below 50% + OpenAI's $21B operating loss.
AWS lets you charge AI bots — your site becomes a toll booth.
Bezos's Prometheus — $12B at $41B to build a robot engineer.
Quick hits — Grok in PowerPoint · Xiaomi beats Claude Code · Meta AI in Facebook.
Bridge to funding: same week the US banned its model, China's DeepSeek raised $7.4B.
2 · Deep Dive 1 your storyscreen-share, talk over it
Stop Prompting, Start Building Loops
Framework credit: Dharmesh Shah (simple.ai). Boris Cherny: "I write loops, the loops do the work."
A loop = objective + metric + boundary. Your story: not a developer, asked Claude for 5 loops per business, got 20, ran two — a BD reactivation loop and the title-naming loop.
The beats
The hook: "Something changed how I work this week, and it can change yours." Credit Dharmesh.
The framework: objective / metric / boundary. Runs vs learns.
The honest beat (best point): these weren't autonomous — I was the metric each time. The human's job is the objective, not the keystrokes.
Example 1 (BD): the reactivation loop caught its own wrong assumption and fixed it before I saw it.
Example 2 (juicy, screen-share the slides): the title-naming loop — watch a weak title climb from 2/10 to 10/10 against my value-first rule.
Try this week: take one task, write its objective, metric, boundary. That's a loop.
I Built My Recruiting Dashboard By Talking To My Website
The recruiter-scorecard loop, made visible. A live dashboard on revenuehire.com, built on PageMotor, no code.
This is the loops idea you can SEE. My recruiting numbers used to be scattered across Loxo. Now there's a live dashboard on my own site that pulls the data, scores every recruiter, and refreshes every morning — and I built it by talking to it.
The beats
The pain: the numbers I run the business on were buried in the recruiting system, pulled by hand.
The build: I described what I wanted to PageMotor. It pulls Loxo, scores the funnel, renders the dashboard. No code, no developer.
The result: a live, gated dashboard on my own domain, fresh every morning — the recruiter-scorecard loop turned into something I can look at.
The tie-back: Deep Dive 1 was loops you run; this is a loop you can SEE. Your website can be the place your business reports to itself.
Try this week: name one number you check by hand every week. That's your first talk-to-it dashboard.
⚠ On air
Show the ANONYMIZED demo only (fake names). NEVER the real board — it has live client/recruiter/candidate names behind the admin wall.
$11.64B AI funding (65% of all venture). DeepSeek $7.4B. China 76%.
The cold open's mirror. The same week the US turned off America's best model, China funded its best one at the largest round we've tracked. Strip DeepSeek and the real market actually cooled. The money is voting on who controls AI.
Don't trail off. Tie it back: news is what's happening to you; loops + protecting your brain + a dashboard you can talk to are what you do about it. Then ask for the subscribe, clearly. Judge the episode by conversion, not the clock.