In partnership with
Good morning. It’s Monday, April 20th.
There was a “new groundbreaking model” model going around on X yesterday called MOG-1 that claimed to beat every LLM benchmark test. Turns out it was a hoax, designed to capture signups as it was trained on only the benchmark test questions it claimed to ace.
If you hear about it, don’t be tempted to sign up.
Claude 4.7 Opus is currently the premier LLM for reasoning and coding, and yesterday Grok 4.3 was showing a huge upgrade by being able to produce full Microsoft Word docs, PDFs, PowerPoints, and Excel spreadsheets on command.
Rumors continue about a massive OpenAI model release this week. We’ll let you now when it drops!
-Jeff
AI Breakfast
You read. We listen. Let us know what you think by replying to this email.
Accio Work: Your Business, On Autopilot
Meet Accio Work, the agentic workspace designed to run your business operations end to end. From sourcing products and negotiating with suppliers to managing your store and launching marketing campaigns, Accio Work handles the execution so you don’t have to.
Powered by verified capabilities and deep integrations with business tools, it doesn’t just generate ideas, it takes action. Backed by Alibaba.com’s global supplier network and over 1B products, it seamlessly connects strategy to execution.
Stay in control while everything runs on autopilot.
Claude Design and Word add-in move Anthropic into the workflow layer
Anthropic launched Claude Design and a native Claude for Word add-in, marking a move into the workflow layer. The design tool generates prototypes and marketing assets via layout controls and conversation, while the Word integration enables AI-driven redlining through native tracked changes. Simultaneously, the company is preparing a public security tab for broader code-scanning and repository analysis. These releases coincide with a massive financial update; annualized revenue reportedly surged to $30 billion, tripling in months, with gross margins hitting +40%.
The "Claude Mythos" security edge faces fresh scrutiny as small open models replicate its vulnerability-finding performance in new benchmarks. CEO Dario Amodei remains committed to aggressive compute expansion, telling the FT "there is no end to the rainbow" for scaling, even as he warns that 50% of entry-level office roles could disappear within five years.
Amodei is mending ties in D.C. after he met with U.S. Treasury Secretary Scott Bessent and White House Chief of Staff Susie Wiles, despite a Pentagon designation labeling the company a supply-chain risk. Read more.
OpenAI consolidation triggers executive exodus and the end of the Sora app
OpenAI is undergoing an aggressive structural change to centralize its product lineup. CPO Kevin Weil, Sora lead Bill Peebles, and B2B head Srinivas Narayanan are out as the company consolidates its tech stack. The experimental OpenAI for Science unit and the Prism tool are being folded into a unified Codex coding environment, while the standalone Sora app has been scrapped due to heavy compute bottlenecks. To shore up its enterprise strategy and public messaging, OpenAI recently acqui-hired fintech team Hiro and media startup TBPN.
Sam Altman’s World project is scaling its biometric "proof of human" layer across the web. The iris-scanning identity system is moving into Tinder, Zoom, and DocuSign to verify users against AI bots. A new Concert Kit is also hitting the market, allowing Ticketmaster and AXS to reserve ticket pools exclusively for verified humans. Read more.
Google launches A2UI 0.9 to let AI agents build dynamic interfaces from existing code
Google has introduced A2UI 0.9, a generative UI standard that lets AI agents build dynamic interfaces using an app's existing components. The framework includes a shared web core and a React renderer, with support for Flutter and Angular.
By using a Python-based Agent SDK, agents can perform bidirectional data syncing and handle client-defined functions, allowing them to "speak UI" across mobile and desktop without the security risks of arbitrary code execution.
Google Labs launched Flow Music, a standalone platform that turns natural-language prompts into fully produced songs and playlists. Formerly known as ProducerAI, the system handles end-to-end audio production and remixing.
Google is also reportedly in talks with Marvell Technology to co-develop two specialized AI chips: a memory processing unit to pair with existing Tensor Processing Units and a next-generation TPU optimized for inference to rival Nvidia. Read more.
Watch: A2UIAgent gives AG2 agents a face, using Google’s protocol to build interactive, two-way interfaces dynamically
Meta prepares 8,000 layoffs to fund massive AI infrastructure expansion
Meta is reportedly readying a massive restructuring to fund its AI infrastructure, planning to cut about 8,000 employees, roughly 10% of its workforce, on May 20. CEO Mark Zuckerberg is reallocating capital toward a $135 billion hardware push, flattening organizations to increase reliance on AI-assisted labor and a new "Applied AI" unit for autonomous agents.
A new study highlights "VisionClaw," an always-on agentic system for Ray-Ban Meta glasses. By combining continuous perception with OpenClaw and Gemini Live, the setup achieved up to 37% faster task completion in everyday activities like shopping and scheduling. While the "opportunistic delegation" lowered cognitive load, the system struggled with fine-grained accuracy, hitting only 58% on note-taking tasks while raising fresh concerns over persistent data privacy. Read more.
xAI poised to enter the agentic coding race this week
xAI is entering the agentic coding market next week with the launch of Grok Build and Grok CLI. Grok 4.3 is already in early access for Heavy users, featuring improved frontend performance and a architecture designed to power these new tools. The suite will likely support local CLI execution and remote workflows via "Grok Computer," an Electron-based desktop app featuring third-party connectors and a multi-agent "Arena" mode.
Recent system instructions for Grok 4.3 reveal a native "Skills" framework that allows the model to create and deploy hidden automated workflows. This skill-based automation is expected to be a primary lever for interacting with X data, moving Grok beyond simple chat toward specialized, autonomous task execution directly within the platform's ecosystem. Read more.
Lanes orchestrates parallel AI coding agents in isolated macOS sessions to plan and ship complex projects faster.
Buildra converts prompts into production-ready full-stack apps with live previews, GitHub sync, and one-click Vercel deployment.
Verdent 2.0 acts as an AI technical cofounder that plans, executes, and delivers end-to-end product progress autonomously.
Avina automates lead discovery and personalized outreach by tracking web signals to engage high-value customers on autopilot.
LTVX.ai uses machine learning to predict lifetime value and autonomously deploy revenue-boosting monetization and retention strategies.
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.
