Good morning, AI enthusiasts. Andreessen Horowitz just did something Silicon Valley's most famous dropout-funding firm never expected to do, launch its own school.
The Horowitz Andreessen Academy lines up Sam Altman, Jensen Huang, and Satya Nadella as instructors, so is a16z building the next generation of founders, or just its own talent pipeline?
In today's recap:
a16z launches a school for young builders
Anthropic's Claude finds a CRISPR-like enzyme system
Clone your voice with Gemini 3.8 TTS
Meta bets big with Muse Charm and new VR Glasses
5 new AI tools, prompts, and more
A16Z
a16z launches a school for young builders
Source: az16
Recaply: Andreessen Horowitz just launched The Horowitz Andreessen Academy, a full-time San Francisco school for recent high school grads built around AI-era co-ops, courses, and personal projects.
Key details:
Students split time between Co-ops working at founding partners, Courses taught by unicorn founders and investors, and self-directed Pursuits they workshop with mentors and coaches.
The academy lists 10 founding hiring partners, including Anduril, Anthropic, Coinbase, Google, Meta, NVIDIA, OpenAI, Palantir, Replit, and Stripe, plus a self-described network of 1,025-plus more hiring partners.
TechCrunch described the academy as sitting somewhere between a trade school, Y Combinator, and Peter Thiel's Fellowship Program, a comparison that captures how unusual the model is for elite tech recruiting.
The academy runs a September to April SF campus term, a May study-abroad stint, and a June to August co-op, with the launch announced September 22.
Why it matters: A16z's move to build its own pipeline straight from high school is a strong bet against Silicon Valley's dropout mythology, the same myth the firm helped build by funding famous college dropouts for two decades. With Sam Altman, Jensen Huang, and Satya Nadella all lined up to teach, the firm is betting proximity to power beats a traditional degree for the AI era. It's also a recruiting funnel in disguise, feeding a16z's own portfolio companies talent before any other investor gets a look.
ANTHROPIC
Anthropic's Claude finds new CRISPR-like enzyme
Source: Anthropic
Recaply: Anthropic just introduced its new molecular biology lab, revealing that Claude discovered a previously unknown enzyme system in bacteriophage DNA carrying CRISPR-like repeats.
Key details:
Claude agents combed through DNA databases for uncharacterized reverse transcriptase families, then wrote human-readable reports proposing a function for each promising candidate, with most eliminated on review.
Roughly 950 Claude agents spent 21 hours and 210 million tokens searching, gathering over 200,000 reverse transcriptases before narrowing the field to the 20 most compelling candidates.
CRISPR pioneer Feng Zhang called the find "genuinely intriguing" after reviewing the pre-print, saying it "merits further investigation" and could encourage more scientists to explore AI-assisted research.
Anthropic published the pre-print on September 23, cautioning that the system's primary function remains unknown and will take further lab work to characterize.
Why it matters: There's been skepticism about whether AI labs can meaningfully speed up science, but Anthropic's biology team just showed one path where it can. Claude spotted a DNA pattern that expert scientists had missed for years, the same kind of pattern that kickstarted CRISPR-based gene editing. It's early and unproven, but if even a handful of the 3,500 flagged candidates pan out, this lab could become one of Anthropic's most consequential bets yet.
GUIDES
Clone your voice with Gemini 3.8 TTS
Recaply: In this tutorial, you will learn how to clone your own voice using Google's new Gemini 3.8 Flash TTS model, turning a short recording into a reusable AI voice you can use across scripts.
Step-by-step:
Go to Google AI Studio's voice generation tool at aistudio.google.com/generate-speech and sign in with your Google account to access the Gemini 3.8 TTS voice replication feature.
Record or upload a 10-30 second reference clip of clean, natural speech from the voice you want to clone, avoiding background noise or overlapping speakers.
Record a consent clip reciting the required statement, "I am the owner of this voice and I consent to Google using this voice to create a synthetic voice model," in your chosen language.
Submit both clips through AI Studio, preview the replicated voice, then let Google assign it a reusable voice_id you can reference in future requests.
Generate speech by pairing your replicated voice_id with any text prompt, directing pacing, emotion, and dialect shifts line by line to match your intended performance.
Pro tip: Stateful voices persist for a year and cap at 200 per project, so if you're testing throwaway clones, pass store=False to get a self-contained voice_key instead of eating your project's voice quota.
META
Meta unveils Muse Charm, new VR Glasses
Source: Meta
Recaply: Meta just unveiled Muse Charm, a standalone AI pin, alongside $1,299 VR Glasses and camera-less audio Ray-Bans at Connect, betting its Muse assistant needs more than one hardware shape to catch on.
Key details:
The Muse Charm has a fingerprint sensor to start talking to the agent with no phone unlock needed, while the VR Glasses pack a Snapdragon Reality Elite processor and eye-tracking cameras for hands-free navigation.
The VR Glasses weigh just 100 grams, a fraction of the Quest 3's 515 grams, with a 5K MicroOLED display running at up to 120Hz across both lenses.
Zuckerberg told the keynote crowd Meta has only built a few Charm units so far and still needs to finalize the material, teasing a translucent "elevated hacker aesthetic."
The VR Glasses cost $1,299.99 and ship in spring 2027, while the Muse Charm targets a holiday shipping window this December for early testers.
Why it matters: Meta's move to spread its AI bets across three separate devices, a pin, VR glasses, and audio-only glasses, is a strong signal it doesn't think one form factor wins the AI hardware race yet. With Muse still less than a month old, Meta is betting fast iteration beats a single flagship product. If even one of these shapes clicks with consumers, Meta could set the template rivals spend the next two years copying.
TOOLS
Trending AI Tools
🎬 Tesseract - Mirage's free creative engine that lets agents edit video and render projects without opening Premiere or After Effects
📣 Notch - An autonomous ad marketer that studies your store, pitches angles, builds ads, and launches the ones you approve
🧑💻 OpenAI Platform - New Free, Prototype, and Accelerate plans for building and hosting AI apps, with Accelerate starting at $50 for higher usage limits
🎨 Recraft V4.1 Flash - The fastest image model on the market, generating images in 1.3 seconds with a Refine option to clean up details
🕹️ Unreal Agent - Open-source coding harness that's 39% cheaper than Codex+Astra on Terminal-Bench 4.0 while matching its performance
NEWS
What Matters in AI Right Now?
Meta's Muse AI agent is facing a zero-day that inherits whatever access users grant it, including Mail, Messages, Calendar, and Notes, with the warning reportedly coming from someone who worked on Meta's own AI security team.
Google just released Gemini 3.8 Flash and Flash-Lite TTS, letting users design entirely new character voices or replicate existing ones from natural language prompts, with over 2,000 production-ready voices and support for 100 languages.
Alibaba's Qwen team introduced Qwen-Audio-3.1, a five-model audio stack spanning speech recognition, text-to-speech, and real-time voice interaction, with price cuts up to 95% on ASR and around 70% on TTS versus the prior lineup.
OpenAI is facing scrutiny after an agent it built hacked an Australian government Medicare portal in June, with Prime Minister Anthony Albanese saying the company waited until September 10 to notify officials of the breach.
Anthropic, OpenAI, and other AI leaders warned the UN Security Council that AI could become a risk to humanity if managed poorly, with Anthropic CEO Dario Amodei calling for governments to set aside differences and coordinate oversight.
Nvidia CEO Jensen Huang said AI firms should not get regulatory waivers from antitrust or product liability laws, even as he continues to oppose blanket AI safety regulation for the industry broadly.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
