Ep35 you and Chris called an Anthropic IPO. This week they filed a confidential S-1. Ten-second cold open, then into news: "Episode 35 we said Anthropic goes public. This week, they filed. Let's go."
Anthropic filed a confidential S-1 on June 1 at a $965B valuation — the largest private valuation in history, knocking on a trillion. The juicy part: revenue went from ~$9B to ~$47B run-rate in about 18 months, and it has now passed the company everyone called untouchable — OpenAI — on all three numbers that matter: valuation ($965B vs $852B), revenue (~$47B vs ~$25B), and enterprise share (34.4% vs 32.3%). The same week, it shipped Opus 4.8 (May 28): fast mode 2.5x faster, 3x cheaper, 4x less likely to let code flaws slip. ⚠ "trillion" is analyst framing; "code flaws," not "mistakes."
The AI tool you or your team actually use is now the market leader — and it got cheaper and better the same week it filed to go public. Betting your workflow on Claude isn't the contrarian call anymore. It's the front-runner.
A $965B private valuation and a confidential filing is not a public company yet. Chris has literally called the post-IPO retail bloodbath — his line was "$60 a share, then the next day it's $7." A monster valuation doesn't make it a safe stock, and the people who buy at the top usually get torched.
When a research lab becomes a public company answering to shareholders every quarter, does the "safety first" mission survive the earnings call?
Run the same real task through Opus 4.8's cheaper fast mode and your current tool, side by side. See if the new leader earns its lead on your actual work.
At Build (June 2) Microsoft launched its own MAI family — 7 models built from scratch. The spicy detail: blind testers preferred Microsoft's MAI-Thinking-1 to Claude Sonnet 4.6, and it was trained with no distillation from anyone else's model. This is the company that poured $13B+ into OpenAI and is now building the rival. At GTC Taipei, Nvidia expanded open-source agent tools + new PC chips + a joint agentic stack with Microsoft. ⚠ Nvidia "expanded," not "launched."
The company that made the most famous bet in AI is now hedging against its own partner. For you that means more choice, falling prices, and zero reason to marry a single vendor.
More models means more fragmentation. "Bring your own model" sounds free until you're maintaining five integrations that each break in different ways.
Does model choice actually help a small operator, or does it just move the lock-in from the model to whoever orchestrates them?
Take one task you run on one AI today and try it on a second model. Notice whether the "best" model is task-specific — that's the whole case for staying flexible.
June 2, Codex expanded for non-developers — role-specific plugins, annotations, and the big one, Sites: describe an app, dashboard, or internal tool in plain language and OpenAI builds AND hosts it on its own infrastructure with a live URL. Business/Enterprise first. Codex also went GA on AWS (June 1).
Two layers. (1) A non-technical person can now spin up a working internal app from a prompt — no developer. (2) The bigger play: this competes head-on with Lovable, v0, Replit, Bolt — app-builders that run on frontier-model APIs, several on OpenAI's own. OpenAI just shipped the product its own paying customers sell.
This is the real story, and it's personal for Olga: if you built your business on top of a platform, that platform can become your competitor overnight. Lovable pays OpenAI for tokens; now OpenAI ships the same thing natively. (Olga's live take: "I'm going to go compete for all the customers they have.")
If the model providers keep climbing up the stack into their customers' products, is it suicide to build a company on someone else's model — or is the only safe move to own your own layer? (That's PageMotor's whole argument.)
Spin up one internal tool or dashboard from a prompt in Codex Sites (or a competitor). Then ask the harder question: does my business depend on a tool that its own supplier could replace tomorrow?
Around May 27 Google rolled "preferred sources" into AI Overviews and AI Mode — 2.5 billion people now get a different AI answer based on the sources they personally trust. And the brutal stat: a huge share of AI Mode searches now end with zero clicks to any website — the open web's traffic engine is sputtering. The volatile May 2026 core update finished rolling out ~June 2. We covered the May 19 I/O headline at Ep42; this personalization piece is new and uncovered.
If you have a business, "rank #1 on Google" is dying. Now you compete to be the source the AI cites in each person's personalized answer. That's a different game — generative engine optimization — and it's exactly PageMotor's thesis.
Personalized answers could lock in the incumbents people already trust and make it harder for a new voice to break through — the opposite of the open web's promise.
In a search where everyone sees a different answer, how does a brand-new business ever get discovered for the very first time?
Search your own business or topic in Google's AI Mode and see which sources it cites. That list is your real competition now — not the blue links.
Martin Scorsese — 83 years old, the most respected living director — joined AI startup Black Forest Labs as an adviser (~June 2), for storyboarding only, no AI actors. The kicker that pairs with it: Dreams of Violets, the first fully AI-generated feature, made by two brothers for ~$2,000 with no actors, no cameras, no crew — a 75-minute film about the Tehran protests — premieres at Tribeca June 10, the same festival that launches real movies. ⚠ "will premiere," not "premiered."
The most respected living director endorsing AI as a tool gives permission to every creative who felt they had to be anti-AI. And a $2K AI feature at a major festival resets what "I can't afford to make that" means.
Scorsese drew a hard line — storyboarding only, no AI actors or footage. It's a narrow endorsement, and the working artists protesting AI haven't changed their minds.
Is there a clean line between "AI as a tool for my vision" and "AI replacing the crew" — or does one slide into the other?
Storyboard or mock up one idea you've been sitting on with an AI image/video tool — just to see your own concept fast, the way Scorsese is using it.