Good morning. It’s Friday, March 27th.
After attending Denver’s AI Tinkerer’s meetup on Wednesday, it’s clear that serious builders are all in on Claude Opus for developing quick coding projects. Even self-proclaimed “non-technical” people had working software that they were already getting customers for. Truly a game-changer.
I’d encourage all of you to explore coding with Claude even if it’s your first time.
A good place to start for building apps is the Vibecode App developed by my friend Riley Brown, and you can use a promo code “Vibes” for 20% your subscription.
Share with me what you’ve built and I’ll include it in the newsletter!
-Jeff
AI Breakfast
On this day in tech history: 50 years ago today, A 20-year-old Bill Gates gave the opening address at the First Annual World Altair Computer Convention in Albuquerque, New Mexico.
You read. We listen. Let us know what you think by replying to this email.
Most coverage tells you what happened. Fintech Takes is the free newsletter that tells you why it matters. Each week, I break down the trends, deals, and regulatory shifts shaping the industry — minus the spin. Clear analysis, smart context, and a little humor so you actually enjoy reading it. Subscribe free.
Anthropic’s Claude Mythos: The Most Powerful AI They’ve Ever Built — And Why They’re Taking It Slow
Yesterday a deleted Anthropic blog post leaked (but was quickly scraped by an X user), revealing Claude Mythos: a brand-new model the company calls “by far the most powerful AI we’ve ever developed.”
It’s a whole new tier above Claude Opus 4.6: bigger, smarter, and scoring dramatically higher on coding, academic reasoning, and especially cybersecurity benchmarks.
They picked the name “Mythos” to represent the deep connections between knowledge and ideas. But here’s the part that raised eyebrows: Mythos is already far ahead of every other model when it comes to finding and exploiting vulnerabilities. Anthropic openly warns it points to an “impending wave” of AI-driven cyberattacks that could outpace defenders.That’s why they’re being extra cautious.
The post indicated that instead of a full launch, they’re starting with a small group of cybersecurity organizations for early access. The goal was to give defenders a head start hardening their systems before things get serious.
The model is also apparently huge and expensive to run, so a wider release is still months away while they work on efficiency. If the post is true, Anthropic built something that could shift the entire cybersecurity landscape. And for now, they’re keeping it under tight wraps.
Meta launches TRIBE v2 and new Ray-Ban AI glasses
Meta AI has released TRIBE v2, a tri-modal model integrating video, audio, and language to predict human brain activity across naturalistic and experimental tasks. Trained on 1,000+ hours of fMRI from 720 subjects, it outperforms linear encoding models, replicates classical neuroscience results in silico, and maps multi-sensory integration. Complementary systems, v-Sonar, v-LCM, DINOv3, and dyadic motion datasets, advance understanding of audiovisual processing and neural language development.
Meta is also preparing two new Ray-Ban AI glasses, Scriber and Blazer, with FCC filings confirming dual sizing, portable charging cases, and production-ready units.
The company is cutting 700 jobs, shifting $135 billion to AI data centers, launching Meta Small Business for 250 million SMBs, and placing CEO Mark Zuckerberg on the President’s Council on Science and Technology. Read more.
Claude 4.5 auto-fixes PRs and masters ‘vibe physics’
Claude Code now offers cloud-based auto-fix for pull requests, resolving CI failures, addressing review comments, and optionally auto-merging changes via GitHub. Mobile access extends to Figma, Canva, and Amplitude. A new engineering blog outlines Claude Code auto mode, detailing a prompt-injection probe and two-stage classifier designed to reduce permission fatigue.
Claude 4.5 tackles a two-week quantum field theory project in “Vibe Physics,” producing a full technical paper across 110 drafts and 36 million tokens with AI-driven drafting and minimal human oversight. Anthropic also introduced abstractive red-teaming, efficiently detecting biases, unsafe advice, and hallucinations in Claude Sonnet 4, GPT-4.1 Mini, and other models.
Anthropic is considering an IPO as early as October, potentially raising over $60 billion, and won a federal injunction blocking the Pentagon’s Claude ban, citing likely First Amendment retaliation. Read more.
OpenAI adds plugin ecosystem to Codex as new ‘Spud’ model pretraining completes
OpenAI has added plugin support to Codex, letting developers package skills, app integrations, and MCP configurations into installable units. These plugins now integrate with Slack, Figma, Notion, and Gmail, can be shared across repos or personal marketplaces, and include metadata for apps, skills, and assets. A built-in plugin-creator handles scaffolding, versioning, and authentication, while an official directory and self-publishing are coming.
OpenAI also completed pretraining its “Spud” model, rumored to power its desktop superapp with ChatGPT, Codex, and Atlas, and released a GPT-5.4 prompting playbook for structured UX/UI design with React, Tailwind, and automated checks. Read more.
Omma turns simple prompts into 3D scenes, apps, and websites instantly using parallel AI agents.
Edgee compresses Claude Code usage, extending limits and preserving context so long workflows don’t get interrupted mid-task.
Slicer copies any website element into ready-to-use AI prompts or code for instant reuse across design and development tools.
Agentplace enables rapid creation and deployment of task-specific AI agents integrated with existing tools and enterprise workflows.
Magine orchestrates vision-enabled AI agents for autonomous web navigation, perception, and real-time task execution without human input.
Thank you for reading today’s edition.
Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.
Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!
Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months.
