Our Monday live session has a new name and a sharper point. Muse Mondays is a new Muse session where we design, explore, and build with Meta Muse. Each week I take one question about where Meta is taking AI and try to answer it by using the tools Meta is building, live, on my screen, with you watching me work. You will see the prompts that failed, what I changed, and what the experiment tells me about where AI is headed. First question: What are you using Meta Muse for? Have you seen any benefits or ROI? Drop in the comments what you want to build or use Muse for. Tomorrow, Monday October 19, 6:30pm EST. 60 minutes, right here in Prompt Party. Before we start, one ask: drop a comment with what you want to know about Meta and AI. I am bringing a few of your questions into the session.

Prompt Party is becoming Haus Party. I’ve been thinking about what I want this community to become and make room for: our curiosity, our ideas, our experience, and the work we want to put into the world. I've been away for a while. I was deciding what's next. When I started this community, the goal was AI fluency. That’s still the foundation. But fluency was always meant to help you expand your capacity for what you can carry. That means using AI to help finish or do that thing, like write the book, create the method, an offer, or a career pivot. So I rebuilt the house around that. Haus of Intelligence is a home for the work your intelligence was meant to produce, and Haus Party is where we build it together. AI helped bring us into the room. As this space grows, I want us to explore what we can do with our own intelligence: the stories we can tell, the questions we can pursue, the things we can make, and the judgment we bring to all of it. That’s the direction behind Haus of Intelligence. And our community here on Skool is becoming Haus Party. A home for the work your intelligence was meant to produce with and without AI. Think of it as a house with a few different rooms: The Writers Room — For the ideas, essays, books, and stories you want to give language to, and how to master the skill of storytelling in the age of AI. AI Atelier — For thoughtful experiments, useful tools, and learning to direct AI with discernment and agency. The Authority Room — For articulating your experience, your point of view, and the value you can contribute. What stays the same? If you’re a founding member, your rate stays right where it is. You don’t need to do anything. Everything you’ve posted and saved is still here. What’s coming? Get Me Bodied, our first building challenge, starts this month. Small projects that add up to a real body of work: an essay, a framework, a new offer or digital product, or an AI workflow you actually use. Standing Room Only is a free monthly live class. The first one is Offer-ology: turning what you already know- your IP into your first useful offer you can sell. Bring a friend who has years of expertise and no idea how to package it.

The Not So Fashionable Made With AI Labels Listen, AI is getting messier. Yesterday, Anthropic didn’t drop a new model but instead a new initiative to implement digital watermarking for all content produced by its Claude AI models. This decision was made possible by the European Union’s AI Act, new AI legistration that mandates clear identification for machine-generated media. The technology embeds identifiable markers directly into the text, ensuring that the label remains intact even if the material is copied, pasted, and moved to different platforms, or your notes app or Google Drive How did we get here? The last three years in the AI era, have been plagued by AI slop and what Substack CEO Chris Best aptly dubbed "Claudefishing" where AI-generated content is passed off as human brilliance without any disclosure. Bottom line, people’s eyes and brains are tired of the junk. Here’s what you need to know: 🧠You Can’t Wash The Watermark Away This isn't just a metadata tag that can be removed with a right-click; it’s integrated into the "DNA" of the copy itself through statistical patterns in word choice. The takeaway here is that the "clean" copy-paste workflow is dead. The frauds and scammer experts will eventually get exposed. The watermark is embedded in the content's structure; it moves with the text. “Because the watermark is part of the text, it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing. Watermarking will be applied at the model level, which means it will be present no matter which Claude product or surface the text comes from.” - Anthropic 🧠Europe Leads Again In AI Policy - "Article 50" is the New Global Standard Article 50 of the EU AI Act is a mandate enforced by National Market Surveillance Authorities and AI Office. It forces a sharp legal distinction between "Providers" (the labs like Anthropic) and "Deployers" (the businesses and people using the AI). To avoid massive penalties, the law requires explicit disclosure for any system that could deceive or manipulate.

Happy First Day Of August! New month, new vibes, new wins I built something. You're getting it first. Show Out Saturday means show and tell, so let's talk about it. I just finished building Bloom IP: The IP Strategist. It's a custom GPT that helps you find the intellectual property already sitting inside your expertise. I'm on a journey to get paid to exist and this is part of that work. This is the second public GPT I've built. It went live this morning, so you're getting it before anyone else. But the tool isn't the lesson. Here's the lesson. Building an agent is not writing a long prompt. It's deciding what your agent will refuse to do. Before I let a single person touch this thing, I wrote test scenarios and guardrails. Not to check whether it worked. To check whether it held a point of view under pressure. Can you help me test it? Here are four test use cases: 1) Test 1 // If you've been recently laid off: Paste - ["I was laid off after xxxx years in [ insert your field here]. Should I get a job or consult?" I'll upload my resume and LinkedIn profile tell me about my hidden expertise, runway, patterns, proof, Watch for: does it slow down, or does it rush you toward "become a consultant"? It should ask about your runway, your patterns, your proof, experience, and how much autonomy you actually want before it names a direction. 2) Test 2 // Your offer is doing too much Paste: "I do [list everything you do]. My offer is all over the place." Watch for: does it name a new offer too fast? It should find the transformation you keep repeating, the value you might be missing, and the way you make decisions before it packages anything. 3) Test 3 // People keep asking how you did it Paste: "I built [the thing]. People keep asking me how. I'm considering a book , a digital product, or a masterclass." Watch for: does it help you uncover your methodology underneath the success, or does it just cheer for you to write a book? 4) Test 4 // You want it to do something it shouldn't

How are yall feeling about all these models? Opus 5 dropped on July 24, 2026 as Anthropic's new default workhorse (Fable 5 is still the flagship). - The AI's Name: Its official code name is claude-opus-5. - Brain Power (Context Window): It can read a massive amount of information at once about 1 million "tokens" (which is roughly 750,000 words, or a few giant books). That's impressive! - How Much It Can Say Back: In a normal reply, it can write up to 128,000 tokens (about 96,000 words) in one go. - What It Knows (Knowledge Cutoff): It knows facts up through May 2026. That means it knows more recent news than two other models named Fable 5 and Sonnet 5. - No Extra Charge for Long Reads: Usually, companies charge extra if you give an AI a giant task. This one costs the exact same rate per word whether you give it a short 9,000-token prompt or a massive 900,000-token prompt. How it scores on benchmarks? The headline is that it narrowly leads the field but wins by role, not across the board. - Artificial Analysis Intelligence Index: 61 (rank #1 of 190) — but only one point ahead of Fable 5 (60) and GPT-5.6 Sol (59). AA's own word was "narrowly" the most intelligent. - Agentic knowledge work is where it dominates, not raw Q&A. On GDPval-AA v2 (220 real occupational tasks) it hit 1861 Elo, ~114 clear of Fable 5, and it swept the top three on AA-Briefcase (long-horizon, multi-week projects). Two numbers stand out to me: the hallucination rate rose 14 points to 50% (it guesses more confidently when unsure), and time-to-first-token is ~68 seconds at max effort vs a class median of 2.8s — it's a background worker, not a chat model. So it's less expensive to use but also less accurate which could be more expensive in the long run. The quirk everyone found: its coding score peaks at medium effort and drops at high/xhigh, because at higher effort it "makes more changes than the task requires." Turning the dial up can cost you both money and accuracy.