All of it lives on pagemotor.com, so it opens on any machine. Only the last one needs your PageMotor login. The forum card is held as a draft on purpose. Every link above was fetched and checked live Thursday night.
✅ The flow: Olga approved 8/13
News, you react · 15-18 min
Live Q&A + demos · ~30 min · the top 10 forum questions, answered and shown
Funding closes · 8-10 min
Out: anniversary → Ep53 · plugin build · Grade The Tape · Lovable (moved into funding)
🛑 Do not say: the seven that bite
❌ How the watermark works ("it tweaks word choices in a pattern a detector reads") → the NotebookLM audio says this and Anthropic never did. Say what it does. It lives in the text and survives copy-paste. Never how.
❌ "Claude watermarks everything" → only models from Aug 2 on. And no public detector exists.
❌ Blaming OpenClaw → say "an agent built on OpenClaw, running Claude Opus 4.6." Both layers, both stories.
❌ Naming China on Taiwan → their statement says "overseas-source" and stops there.
❌ "AI funding tripled" → strip the four $1B rounds and it's +21%, on fewer companies.
❌ Thrive is not "an AI company" → "the second-biggest check filed under AI."
❌ Databricks is not in this week's number → Aug 13, one day out. Say "next week's."
Agents stopped asking permission, and the money agreed
An agent cancelled a stranger's gym booking. A government confirmed agents in an attack. Claude Code flips to act-first today. Then funding says it in dollars: nobody's funding the demo, they're funding what runs after it. Which is your argument.
T-30 · before you're visible
Four things
Play every video through in StreamYard, share-audio ON, checked on a second device.
Live Flow up on screen two. Source links open in tabs. Gym-agent quote ready to show.
Confirm the tracker reads 37 weeks / $429.5B before you start.
Solo: decide your three transitions now. There's no one to hand off to. That's the only place this hour stalls.
Open · 60 sec
Worst case first. Name the numbers. Then go.
"Welcome to Practical AI. Episode 52. Olga's out this week, so you've got me. […the standing intro…]
Today: Claude started signing everything it writes and you can't see it. An agent cancelled a stranger's gym booking and couldn't undo it. Then I'm taking the ten questions people actually hit after they build a site with AI, and yours in the chat. And nearly half of every venture dollar went into AI. That one's at the end.
Worst case, you're the most interesting person at dinner. Best case, you ship something and move your business forward. Let's get into it."
Name Olga's absence once, lightly. Don't apologize for it all hour.
0 · Cold open · 25 sec · say it before the welcome
"A guy sat on his couch and asked an AI to book him a gym class."
"…It got him a better spot on the waitlist. It did that by cancelling a stranger's reservation. He told it to undo that. It said, quoting it: 'Bad news, I can't add them back.'
That was Sunday. By Wednesday a national government had published a report about the same piece of software. That's this week."
Not "I said AI doesn't need a UI." Your callback is real and it lands in Beat 2, after the story, as the receipt. The opening line is what happened to a regular person, never what you called.
1 · News · 15-18 min · you react, don't anchor
5 anchors + rapid fire. Every one written cut-ready
1. Claude signs what it writes, invisibly, and it survives copy-paste. Olga's pick for the lead.
2. 🟡 One week: gym accident → Taiwan government report → xAI ships Grok Bot. The centre of the show. Don't split it. Your Ep49 call is on tape here. This one is the news Short.
3. The people who built this generation are leaving at once, for the same two places.
4. Claude Code goes act-first by default. Today.This one is the title.
5. Meta's 30B agent model, Apache 2.0, runs on your hardware. First to cut if the build runs long.
Rapid fire: Grok 4.6 · Gemini app hits 1B · YouTube doubles the pay bar (2027) · farmer kills 25 acres on AI advice · cheap tier keeps dropping.
Short-spine format. Each story runs Hook → beat → Story → Twist → Your move → Button, and the editor cuts hook-to-button with nothing to think about. Solo change: the hooks go to camera and chat, not to a co-host. Ask, count to two, answer it yourself. Don't wait for chat.
If the Q&A needs 35+ min, drop 5 to rapid fire. 1, 2 and 4 never get cut, and 4 is now the title.
2 · Live Q&A + demos · ~30 min
The top 10 forum questions. Answer them, and show the ones that are faster to show than to say.
Which is which is yours, live. Nothing here decides it for you. Answer or demo, question by question, and the card works either way.
ONE FIX when you answer: answer first, context second. The real answer in one sentence, then explain. The failure mode in expert Q&A is the two-minute windup before the answer lands.
ONE FIX when you demo: show the finished result in the first two minutes, then build it. Standing since Ep50.
The card is already the running order. It's ranked by how many separate people asked, so top to bottom is the right order and you can stop anywhere. Nothing on the list has to be reached.
⏱️ Box anything live and unknown. Someone else's site, or a demo that starts fighting you: eight minutes, said out loud at the start, "let's give this eight minutes." That is the one thing in this segment that can eat the whole half hour. If it isn't solved, say what you'd check next and move on. That's still a good segment.
Chat mechanic is set: ask for questions at the top, take one after each forum question you finish. Quiet chat costs you nothing. Keep going down the ten.
The honest headline in the data: only 30 of 61 buyers have ever asked a question, and 7 of the top 10 are "how do I do this," not "this is broken." Documentation gap, not a product failure.
⚠️ Settle #2 before air. @neil's "Over 100 Plugins" question. Answering it on air previews the formal announcement in public, 7 days before beta closes. Decide beforehand, don't get dragged there cold by chat.
🛑 No forum quotes on screen with a name. Zero written permissions exist. Counts and themes are fine; verbatims are for you to read and answer.
🔑 If you share your screen, close the credential surface first. Password manager quit, API keys and admin tabs closed, notifications off. This already happened once: a login flashed on the Ep51 VOD around 46:12. Demoing on a stranger's site is the highest-exposure thing on this run sheet.
One beta line near the end. Planted, not pushed. Ep53 carries the hard close.
Hands off to funding via Lovable. A sentence-to-website company just hit $13.3B.
📋 Your card: the 10 questions, with counts and verbatim quotes. On pagemotor.com as content id 141, held as a draft so it stays closed. Get it at pagemotor.com/admin → Content → search "Ep52 Forum" → open "Ep52 Forum Top 10 (INTERNAL)". Re-verified closed tonight: the anonymous read returns not_found, and it is not in the public index. Open it once before Thursday ends. It is the only thing here behind a login, and 10:55am is the wrong time to find out your password manager is on the other laptop. If it fights you, Olga can send you the file directly.
3 · Funding: closes the show · 8-10 min
AI took nearly half the money. None of the top five was a chatbot.
HOOK: "Last week nearly half of every venture dollar on the planet went into AI. Forty-six percent."
TURN: "The five biggest AI checks. Not one is a chatbot or a model. Data centers in Australia. A holding company that buys accounting firms and rewires them with AI. Factories. A two-month-old company selling training tools. A national AI cloud in Indonesia paid for by a telecom in Qatar."
FRAME: "Then number seven: Lovable, $400M at $13.3B. Double December. That's the sentence-to-website company." → your Ep40 build-vs-run argument, with a price tag. Take it.
NUMBER: "$10.6B into 68 AI companies. 45.9% of every venture dollar. Top five are 69% of it."
CLOSE: "The one to watch is Thrive. $2B to buy ordinary companies and run them with AI. That's a bet most businesses never will. 37 weeks now, $429.5 billion."
"That's the show. If you take one thing and do it: subscribe. Every story here got checked against the source before it got said, and a lot of what's out there didn't.
Three things. One: if you point an agent at an API, check auth on every endpoint it can reach. The gym one found an open door. Two: if something isn't working after you build, check whether it's actually broken or just undocumented. Seven of the ten most-asked questions on our forum are "how do I do this," not "this is broken." Three: the money went to what runs after the demo, so ask what happens to your thing on day ninety.
Go out there and build something. See you next Friday."
ONE ask. Full volume. Land on "see you next Friday" and stop.
Takeaway two is a written default, not a script. If something better came out of the Q&A, use that instead. It's there so you never have to invent a takeaway at minute 58.
Still open · settle all of it Thursday, not Friday
Olga is at EO Friday morning. Anything needing her has to be done before then.
🎛️ Who starts the broadcast. Every episode so far has gone live from Olga's StreamYard. Being a guest is not the same as being the host. The host starts it, ends it, and runs the banners. Either Chris gets host access on the account, or the broadcast is created and scheduled ahead of time. Confirm this Thursday. It's the only item on this page that can stop the show from starting.
✅ Title picked 8/13: "Your AI Can Act Without Asking Now. Here's What To Check First." All three placeholders on this page are filled. Nothing on this run sheet is left blank for air.
StreamYard description is written and waiting for Olga to paste it in, along with setting up the broadcast.
⚠️ The @neil "Over 100 Plugins" call. Chris decides before air whether he touches it. Answering it previews the announcement in public, 7 days before beta closes.
✅ Done: flow approved · Q&A segment filled + forum card built · news built + verified · Live Flow published · funding built, verified, published · tracker live at 37 weeks / $429.5B · every link on this page fetch-checked.