Instagram | Sponsor this newsletter

We start today with where to find and how to actually use the new Claude models.

And back-to-school season just got a plot twist nobody saw coming. America's biggest school district is going full prohibition on AI, right as new research drops explaining exactly why. Meanwhile, Google just signed the internet's biggest creator to sell you on Gemini, and OpenAI casually announced a model that can hack systems without a human telling it how.

Buckle up. It's a heavy one.

In today's edition:

Anthropic quietly dropped Claude Fable 5.1 and Claude Mythos 5.1 this month, its most advanced models for coding, research, and long-running knowledge work. Here's how to get hands-on it.

💡 Pro tip: Fable 5.1 defaults to High effort in Claude Code and Medium in Cowork and Claude.ai. If you're burning through usage fast, drop the effort level manually since Anthropic's own data shows Low and Medium effort often matches Fable 5's old performance at a fraction of the cost.

Fable vs Mythos vs everything else: they're the exact same underlying model. Fable 5.1 is the generally available version with safety layers on cybersecurity and biology. Mythos 5.1 strips those specific restrictions for vetted cyberdefenders and life scientists through Anthropic's trusted access programs. It's not a smarter model, just a less restricted one. Against the competition, Anthropic's own benchmarks show Fable 5.1 pulling ahead of both Claude Opus 5 and GPT-5.6 on agentic coding and long-horizon reasoning tasks, while costing up to 45 percent less than its predecessor for heavy agentic workloads.

Want the deeper playbook on actually operationalizing frontier models like this inside your workflow, not just the how-to? That's exactly what The Operator Brief is for. Subscribe for the in-depth version of how you actually implement this stuff.

Small teams don't have time to stitch together five tools and hope it works.

Apollo gives you everything you need to find leads, reach them, and close deals — all in one place:

Stop juggling tools and start building pipeline that scales.

With Apollo, the AI revenue engine powering 4M+ users.

Sign up for free

New York City is about to run the biggest AI experiment in American education, and the experiment is "don't."

Starting next week, Mayor Mamdani and Schools Chancellor Kamar Samuels are rolling out a ban on generative AI for every student from pre-K through 8th grade. That's roughly 600,000 kids. High schoolers get access, but only under "restricted" rules.

Here's what's actually changing:

The timing isn't random. A new Al Jazeera breakdown of a 30-month study tracking 27,000 students in China found that after kids started using generative AI for homework, their scores rose 18 percent and the time spent per assignment dropped by nearly a third. Sounds great, until you look at what happened next: when the same students sat closed-book exams without AI, their scores fell 20 percent within six months. Researchers are calling it "metacognitive laziness," the erosion of the actual thinking that homework was supposed to build in the first place.

Here's the uncomfortable part.

The research doesn't actually say AI is the problem. It says how kids use AI is the problem. A Vodafone Foundation survey of 7,000 European students found 56 percent use AI just to look things up, and 45 percent use it to get concepts explained, both genuinely useful. Only 31 percent used it to hand in complete finished answers. NYC's blanket ban treats all of that the same. Meanwhile China, the exact country whose students showed the exam decline, isn't banning AI in schools. It's mandating eight hours of AI instruction a year. The UAE, Saudi Arabia, and Qatar are integrating it into curricula too. NYC is choosing prohibition while the rest of the world is choosing (better or worse) curriculum design.

🔮 Prediction: This ban will not survive intact past its one year review, but it will reshape the national conversation in the meantime. Expect a wave of copycat announcements from other big districts over the next two months, mostly symbolic, since none of them have NYC's enforcement capacity or its 600,000 student scale to actually test.

The real consequence lands somewhere else entirely: wealthier families will simply route around the ban with private tutoring services and personal devices at home, while lower-income students lose access to the exact tool their better-resourced peers keep using outside school hours. That's the opposite of what a moratorium is supposed to protect against. By next fall, I'd bet NYC quietly reintroduces AI for grades 6 through 8 under a "structured use" framework almost identical to what the UAE and Qatar are already running, essentially conceding that the OECD's actual finding was correct all along: it's not the tool, it's whether anyone taught kids how to use it. The district that banned AI first will likely be the one writing the curriculum for using it responsibly, second.

Reply and tell me: is an outright ban the right call, or does this just delay a problem that structured AI literacy could actually solve?

Google didn't just land a sponsorship. It landed the internet's single biggest creator as a walking, talking Gemini demo.

Google announced a multi-year partnership with Jimmy Donaldson, better known as MrBeast, extending their existing YouTube relationship into Gemini and Google Health.

The details:

Here's the uncomfortable part.

MrBeast content already runs on product placement as a business model. Folding Gemini directly into the stunts themselves blurs entertainment and advertisement past the point most viewers will even register it as a sponsorship. And this lands one week after we covered Gemini's search history personalization rollout, the "tell me about myself" feature that pulls straight from your search history. Google is simultaneously pushing AI deeper into how much it knows about you, and paying the most watched creator on earth to make that feel fun and aspirational rather than unsettling. That's not an accident.

🔮 Prediction: This is Google borrowing Apple's old playbook: pair the product with a beloved face until the association becomes automatic, except compressed into creator economy speed instead of decade-long brand building. Expect OpenAI and Meta to respond with their own creator partnerships within the next two quarters, likely targeting creators with strong Gen Z reach specifically, since that's the demographic least likely to have formed AI habits around a single assistant yet.

The bigger shift is structural: MrBeast's audience effectively becomes an unpaid focus group for consumer AI features. Whatever works in his videos (faster wayfinding, real-time hazard spotting, and logistics planning) gets fast-tracked into Gemini's mainline consumer app because Google now has hard data on what makes millions of viewers actually want to open it. Watch for a backlash cycle too. The first time a Gemini suggestion visibly steers a MrBeast stunt in a way that feels scripted for the sponsor rather than the story, expect creator ethics discourse to follow within days.

Reply and tell me: is this a smart distribution play, or the moment AI marketing officially jumped the shark?

Advertise to 200k engineers and CTOs choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

🏗️ Investor Gavin Baker says the AI data center boom is quietly the best thing to happen to blue-collar Americans in decades, arguing electricians and HVAC techs are out-earning college grads.

🏖️ San Diego's tourism authority just launched an AI chatbot that builds your entire trip itinerary, though it recommends paying members before anyone else.

🔓 OpenAI's upcoming Astra model just crossed a line nobody's crossed before; the company says it can find and exploit unknown security flaws with zero human guidance.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI