Everything fact-checked. Everything verified. Every decision Olga owns is marked.
Packet dated · May 21, 2026 Show date · Friday May 22 · 11am CT
✓ News content verified (Research agent + Grok)✓ Funding deployed✓ Predictions log updated⚡ Callback approval needed⚡ News order approval needed⚡ Deep dive approval needed⚡ Live-flow HTML pending
The Week's Big Shift
This is the week predictions hit reality. Olga called the Musk verdict four days before the jury. Chris's "AI doesn't need a UI" call landed at Google I/O. And Olga's accounting books just became an agent surface.
The biggest week for AI headlines was the smallest week for AI funding. OpenAI filed for a $1 trillion IPO. Anthropic is negotiating $900 billion. Google rebuilt Search around agents. And while all that was happening, the rest of the AI market raised $2.3 billion. Down 76 percent from last week.
Section 01
The Callback
Opens the show · Before news · Scripted + graphic
⭐ Olga called it · Four days later
Episode 41. Four days later. The jury agreed.
Olga · Ep41 · ~23:30
"I don't think Elon Musk is going to win. There's too much money at stake. OpenAI will pay some fines, paid in the form of something to a nonprofit arm."
HIT (core call): May 18 jury took under 2 hours to unanimously dismiss all of Musk's claims. OpenAI, Altman, Brockman, Microsoft all cleared.
MISS (sub-call): No fines. No nonprofit relief. OpenAI paid nothing.
Sources verified May 21 (Grok + research agent): NPR · NBC News
Approve this Callback as the show opener?
Yes — open with the Musk hit. Strongest signature beat of 2026. Your prediction. Hard date. Four days later. The honest tracking of the sub-miss makes it even stronger.
No — use the Google UI-bifurcation Callback instead. Ep39 Chris + Olga UI predictions landed at Google I/O. Demoted to Beat 1 in current order. Pull it up as Callback?
Different angle entirely. Tell me what you're seeing.
Section 02
News Briefing · Week of May 21
Hard cap · 14-min block · After the Callback
SEGMENT 01
The Musk Verdict Unblocked OpenAI's IPO
3 min · Lead
Ride directly from the Callback into the IPO story. They're one chain. The dismissal cleared the structural obstacle. OpenAI is filing this week.
Target: $1 trillion valuation. $60 billion raise. September 2026 debut. Goldman Sachs and Morgan Stanley leading. The biggest tech IPO in years.
And Chris's Ep36 call is now on the clock: "OpenAI IPO gets chopped at the knees post-IPO. $60 then the next day it's $7. Early retail investors destroyed." September is when we find out.
If you're building on OpenAI, this is the moment OpenAI stops being a startup and starts being a public company answering to shareholders every 90 days. Roadmap discipline up. Pricing pressure up. Behavior change incoming.
SEGMENT 02
Anthropic Ascendant · $900B Valuation In Negotiations
2.5 min
Anthropic is negotiating a $30 billion round at a $900 billion valuation. If it closes there, Anthropic passes OpenAI (~$852B last valuation) for the first time.
Revenue run rate: $30 billion+ annualized, up from $9 billion at year-end 2025. Hit #1 on CNBC's Disruptor 50 this week.
The "Sam shifts doctrine to win" Callback has hit four times. But here's the twist: Anthropic now has the revenue AND the valuation, while OpenAI has the distribution (Dell, Singapore, IPO momentum).
If you're betting on a horse, the race just got tighter. Anthropic = revenue. OpenAI = distribution. Which one matters more for your business is the question. We'll argue it on air.
SEGMENT 03
Google I/O · The Agent Layer Ships
3 min
Cover as ONE arc, not 4 separate stories. Google's I/O wasn't a model race — it was an architecture pivot.
Gemini Spark: 24/7 cloud agent that works while your phone is locked. Gemini Intelligence: screen-aware automation across apps. Gemini Omni Flash: multi-turn conversational video editing (free on YouTube Shorts). Gmail Live: voice-powered inbox queries. Antigravity 2.0: agentic dev platform. AI Mode at 1 BILLION monthly active users.
This week's structural echo: Chris's Ep39 "AI doesn't need a UI" and Olga's "UI bifurcation" calls landed in Google's biggest product. Three weeks after the prediction.
The agent layer is at OS level now. If you sell anything that touches Android (apps, content, retail), your customer's phone is about to do work for them without opening your app. Plan accordingly.
SEGMENT 04
Wow Moments · OpenAI Solves An 80-Year-Old Math Problem · Mythos Cracks macOS In Five Days
3 min
OpenAI's general-purpose reasoning model disproved the Erdős planar unit distance conjecture (open since 1946). Cross-domain proof connecting geometry to algebraic number fields. Independently verified by Fields Medalist Tim Gowers and Princeton mathematicians. This was a general-purpose model. Not a specialized math model. That's the wow.
Anthropic's Mythos cracked macOS in five days under Project Glasswing — controlled access to ~40 orgs including Apple, Google, Microsoft. Two vulnerabilities chained into a kernel privilege escalation on macOS 26.4.1 / Apple M5. Mythos is not publicly released.
Honest prediction tracking: Chris's Ep36/37 "Mythos releases by end of May" was logged as a MISS in Ep40 (Pentagon exclusion). That MISS still stands for public release. But the model exists and is operationally devastating in controlled hands. The call on air: "Chris was wrong about public release. He was not wrong about Mythos being a thing."
One model just did original math. Another just walked through Apple's kernel in a week. Capability isn't where the ceiling was last year. Plan for what AI can do for YOUR business assuming it's already 2x what's publicly demoed.
SEGMENT 05
Cultural Punchline · 6.7 Million People Trashed A Real Monet
1.5 min · Closer
An artist posted a real Claude Monet painting (Water Lilies, 1915) claiming it was AI-generated. 6.7 million people piled on with anti-AI critique. It was a genuine masterpiece.
The "Claude" in the headline is Claude MONET the painter — not Anthropic's Claude. End the news segment laughing.
Anti-AI bias is now so reflexive that people trash 100-year-old masterpieces when they think a machine made them. If you market with AI-assisted assets, brace for irrational backlash. Transparency is the strategy, not the threat.
Approve this news order? Five segments, ~13 min total.
Yes — ship as ordered. Musk/IPO lead → Anthropic → Google I/O agent layer → Wow moments (Math + Mythos) → Monet punchline closer.
Reshuffle. Tell me the new order.
Cut a segment. Drop one to make room for something. Which one and what's replacing it?
Mythos honesty call — on air or off?
Call it on air. "Chris was wrong about public release. He was not wrong about Mythos being a thing." Honest tracking strengthens credibility.
Off air. Keep the miss in the predictions log only. Don't surface live.
Section 03
Funding Report · Week 25 · The Drop
10-12 min on air · Deployed and live
While OpenAI was filing for a trillion and Anthropic was negotiating $900B, the rest of the AI market raised $2.3 billion in a week. Down 76% from $9.45B last week. The biggest week for AI headlines was the smallest week for AI funding.
✓ Deployed · Cloudflare
Week 25 Visual Report
4 hero stats · AI-only regional pie · top 5 companies · 5 trends · the one-stat show segment. Designed for live show display.
"While OpenAI was filing for a $1 trillion IPO and Anthropic was negotiating $900 billion, the rest of the AI market raised $2.3 billion in a week — down 76% from $9.45 billion last week. The biggest week for AI headlines was the smallest week for AI funding."
Top 5 companies — keep SendCutSend or swap in Multiverse + Commure?
Keep SendCutSend. Borderline AI (robotics-tagged), but it's the only sub-$200M industrial-robotics scale-up. Pairs with the China robotics seed-wave theme.
Drop SendCutSend. Replace with Multiverse ($70M, UK workforce upskilling) and Commure ($70M, healthcare AI). More clearly AI-native.
Section 04
Deep Dive · QuickBooks + Claude
12-13 min · Olga solo + Chris reacting · Story-driven
The huge-rule headline: "Your accounting software is lying to you in five small ways. Here's how I found mine." Leads with viewer value. Olga's painful story is the proof. Never "watch what I built."
BEAT 01
Cold open · The hook
60s
"I spent a day building the wrong version of something that had an easy version the whole time. By the end of it, my QuickBooks had told me five things that weren't true. And the AI I connected to it found every one of them."
BEAT 02
The wrong path first · Developer API rabbit hole
90s · No screen
Tried QuickBooks developer API access. Killed it: security, privacy, and Intuit's pricing tier built for app companies ($300–$4,500/mo), not single businesses. The lesson: for one business pulling its own data, the connector beats a custom app on every axis. Chris reaction: "naked vibe coding trap."
BEAT 03
The pivot · Cowork connector + Claude Code
90s · On screen
On screen: open Claude Code, run /mcp, show 30+ QuickBooks tools. The big idea: Your QuickBooks just became an agent surface — the same shift Google announced for Search this week, except Intuit shipped it six weeks ago and nobody noticed.
BEAT 04
A vendor outage is not your setup
60s
Intuit's reporting backend down for hours. Ruled out everything else first (email, permissions, Claude plan, dev tier). The lesson: don't fight the broken endpoint. A/R aging was dead; invoice + P&L + company info worked. Built the workaround around what worked.
BEAT 05
The guardrail
60s
P&L wouldn't run until QuickBooks "industry" field was set. Safety guardrail blocked Claude from writing it. Olga set it by hand. The lesson: guardrails blocking a write is the system working. Ask if it's a guardrail before assuming it's broken.
BEAT 06 · THE MEAT
The five lies your QuickBooks is telling you
4 min · On screen
"Connecting the tool was the easy 20%. The real value was discovering my own data was lying to me in five small ways."
Lie #1 · Suspense account contamination. A/R looked like $63K. Real 90+ days outstanding was ~$95K — masked down to ~$41K by unapplied payments back to 2013.
Lie #2 · Cash vs accrual basis. April profit ~$9,573 on accrual. Books kept on cash. The connector's number was directional, not real.
Lie #3 · Bill.com doesn't sync. A client payment sat in the bank for a week while QB still showed the invoice open. The dashboard trusts the bank, not QB.
Lie #4 · Duplicate software spend. $8K–$10K/year of two tools doing one job. AI didn't fix it. AI exposed it.
Lie #5 · The Monday-cron lie. The connector lives in the AI session. A 7am cron cannot call it. One trigger, not zero. Name the limit.
BEAT 07
The real win + close · Both dashboards on screen
2.5 min · On screen
"The engineering win wasn't connecting QuickBooks. It was discovering QuickBooks alone is not the source of truth — then building a system that reconciles instead of trusting one feed."
Close: "Your accounting software is lying to you in five small ways. Here's how I found mine. Try this on your own books this weekend and tell me what you find."
Demo file · Local only
Monthly Health Snapshot
R12 supposed vs invoiced vs collected. 5 of 12 months behind pace. Feb was −$22K from slipped placements.
Yes — 7-beat structure, Olga solo + Chris reacting, ~13 min. Cold open → wrong path → pivot → outage → guardrail → five lies (the meat) → real win + close.
Reorder. Open with the cash-report Falcon-chase moment for more energy?
Different scope. Tell me what's missing or what to cut.
Deploy both demo HTMLs to Cloudflare so you have clean tabs on show day?
Yes — deploy both. URLs would be practical-ai-ep42-monthly-health.pages.dev and practical-ai-ep42-weekly-cash.pages.dev. Add HUB cards. Show-day reliability over local files.
No — keep local. Run from disk during the show.
Run one live Claude query on air during the deep dive?
Yes — one prompt. "Pull my P&L for April and show me cash vs accrual. Then list my open invoices over 60 days." Hits two of the five lies in one query.
No live query. Narrate from the dashboards only. Lower risk if Intuit backend hiccups live.
Section 05
Predictions Log · Resolutions This Week
Verified by Grok
✓ HIT · Core call · Same week
Olga · Ep41 · ~23:30
"I don't think Elon Musk is going to win. There's too much money at stake."
Jury dismissed all claims May 18 in under 2 hours. Becomes the Ep42 Callback opener.
✗ MISS · Sub-call
Olga · Ep41 · ~23:30
"OpenAI will pay some fines, paid in the form of something to a nonprofit arm."
OpenAI paid nothing. No damages, no nonprofit relief. Full dismissal.
✗ STILL A MISS · Public release
Chris · Ep36/37
"Mythos releases to public last 2 weeks of May, if not before."
Still controlled access only (Project Glasswing, ~40 orgs). No public release as of May 21. Honest tracking: the capability is real, the public-release call was wrong.
⚡ STRUCTURAL ECHO · Same announcement
Chris + Olga · Ep39
Chris: "AI doesn't need a UI." Olga: "Managers and knowledge workers stop using software UIs entirely."
Google rebuilt Search around agents at I/O May 19. Both predictions landed in one announcement, three weeks later.
Section 06
Knox Tasks · Pending
What Knox owes us before show day
Ep40 promo title correction — send Anastasia the new locked title "What Can You Do With Your Website When You No Longer Have To Edit It?" replacing yesterday's send. Filed at top of INBOX-KNOX. Confirm-DONE proof required.
Callback graphic for Ep42 — split panel: Olga's Ep41 verbatim quote LEFT, NPR jury-verdict headline RIGHT. Caption: "Episode 41. Four days later. Called it." Owe by Friday morning.
News writeup status — automated cron or manual? Answer due in INBOX-ATHENA with proof (run logs, schedule expression, script path).
Transcript format — standing rule locked: running-text .txt + raw VTT in raw/. Same heartbeat he pulls.
Section 07
What's Still Pending
Owed · After approvals above
live-flow-ep42.html
The on-screen show-day script. Masthead + theme banner + conversational beats. What Olga reads live.
Owed · Step 4 of workflow
Agenda + Title + StreamYard + X Promo
Names the episode LAST. Builds after deep dive + news are locked. Per PRE-SHOW-WORKFLOW.md.
Owed · After full approval
.docx exports
News flash, funding report, deep dive script — all to .docx for NotebookLM.