Presented by
Good morning, AI enthusiasts. Google paid $2.7 billion less than two years ago to bring Noam Shazeer home from Character AI. He's now joining OpenAI.
Shazeer co-authored the 2017 Transformer paper that started the modern AI boom and has since led Gemini's charge against ChatGPT. Has Google just handed its fiercest competitor its most valuable researcher?
In today's recap:
Gemini co-lead Noam Shazeer defects to OpenAI
Midjourney's ultrasound scanner aims to beat MRI
Build and deploy an AI agent with Vercel eve
Pinterest launches AI shopping app, Ask Pinterest
4 new AI tools, prompts, and more
OPENAI & GOOGLE
Gemini co-lead Noam Shazeer joins rival OpenAI
The Information
Recaply: Google's Gemini co-lead Noam Shazeer just announced he will leave the company to join OpenAI, a major talent blow for Google's AI division less than two years after it paid $2.7B to bring him back.
Key details:
Shazeer serves as Google's VP of engineering and Gemini co-lead, credited with narrowing the performance gap between Gemini and OpenAI's ChatGPT in recent model generations.
Google reportedly paid $2.7B to bring Shazeer back from Character.AI in 2024, with OpenAI's valuation now exceeding $300B ahead of its planned IPO.
Shazeer joined Google in 2000 and co-authored the seminal 2017 "Attention is All You Need" paper that catalyzed the modern AI boom, according to Reuters.
Reuters reports the timing of his departure from Google was not immediately clear, though he confirmed the move on X to his followers.
Why it matters: Google paid $2.7 billion less than two years ago to bring Shazeer back, and now he's heading straight to OpenAI. It's not just a talent loss. It's a statement about where the best researchers want to be. Gemini has spent the past year closing the gap on ChatGPT, and losing its co-lead mid-race is exactly the kind of setback Google can't afford.
PRESENTED BY MINTLIFY
The docs platform Anthropic, HubSpot, and Coinbase trust
Your documentation is a product decision. The companies building the most-used developer tools in the world chose Mintlify because great docs reduce support load, accelerate adoption, and convert evaluators into customers. Backed by a16z and Salesforce Ventures, Mintlify powers 20,000+ companies with AI-native documentation that keeps pace with your product — without pulling engineers off the roadmap to maintain it.
MIDJOURNEY
Midjourney pivots from image gen to medical scanning
Midjourney
Recaply: The AI image generator Midjourney just unveiled Midjourney Medical, a full-body ultrasound scanner that moves patients through a pool of sensors. It's a sharp pivot from making AI images to building medical AI hardware.
Key details:
Patients step into a pool and descend through a ring of sensors, with the full-body scan completing in under 60 seconds from start to finish.
The machine uses 500,000 sensors and generates terabytes of data per second. Midjourney says it aims to outperform MRI machines in diagnostic capability.
The company says the project comes from asking what AI can do for human health, a separate effort from its image generation business.
The scanner isn't available yet, with no commercial launch date set. Midjourney calls this a long-term vision.
Why it matters: Midjourney made AI art something anyone could use. Now it's pointing that same instinct at medical imaging. Body scans are expensive, hard to get, and slow to improve. If Midjourney can apply the same push for accessibility to ultrasound that it brought to image generation, the upside for healthcare is real. But going from a creative tool to a medical device is a long road, and the company still needs to prove this works at scale.
GUIDES
Build and deploy an AI agent with Vercel eve
Recaply: In this tutorial, you will learn how to build and run an AI agent using Vercel's eve framework, deploying it with scheduled actions, custom tools, and built-in approvals without managing any infrastructure.
Step-by-step:
Install eve by running
npm install eve. Then create anagent/folder with three items inside:agent.ts,instructions.md, and atools/folder.Open
agent.tsand define your agent usingdefineAgent(). Set the model you want, such as"anthropic/claude-opus-4.8". This is where the agent gets its core configuration.Write
instructions.mdin plain English. Tell the agent who it is and what it should do. Then add tool files to thetools/folder. Each tool is a TypeScript function the agent can call.Add a
schedules/folder for recurring tasks. Connect a channel like Slack by adding a channel file. This is where the agent sends its output.Deploy to Vercel by running
vercel deploy. Open the eve dashboard to watch agent runs and configure approval gates for any actions that need a human check before they execute.
Pro tip: Add a subagents/ folder to hand off specialized tasks to separate agents. Your main agent stays focused while subagents handle things like data pulls or long research runs.
TOGETHER WITH WISPR FLOW
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
TOOLS
Trending AI Tools
⚡ Nitrosend - AI-native email platform that sends campaigns from Codex, ChatGPT, Claude, or any MCP agent with one prompt, no dashboard needed
🎨 Claude Design - Anthropic's design tool with on-canvas editing, design system imports, and two-way sync with Claude Code, in beta on paid plans
🔍 Ask Pinterest - Pinterest's experimental AI shopping app for conversational product discovery, pulling from personal Pins and Boards for personalized recommendations
🤖 Fable 5 - Anthropic's most capable Claude model, currently under US export controls with access restricted to users inside the United States
NEWS
What Matters in AI Right Now?
Vercel introduced eve, an open-source framework for building AI agents with durable execution, sandboxed compute, and approval workflows built in. The company describes it as "like Next.js, for agents" and runs its own internal agents on the framework.
Anthropic updated Claude Design with on-canvas editing, design system imports, and direct sync with Claude Code, allowing teams to hand designs off to build or start in code and pull projects into the design editor. The beta is now available on all paid plans.
OpenAI rolled out scheduled tasks in ChatGPT, letting users automate recurring work, set reminders, and ask ChatGPT to monitor information and send proactive notifications. The feature includes a dedicated task management hub for paid subscribers.
Nitrosend launched an AI-native email platform that removes the dashboard entirely, letting users trigger email campaigns from Codex, ChatGPT, Claude, or any MCP agent with a single prompt. The free-to-start tool builds and sends sequences directly from AI agents.
Pinterest launched Ask Pinterest, an experimental app for conversational shopping and product discovery powered by its internal Taste Graph data. The web app supports multistep queries, retains context across sessions, and connects to personal Pins and Boards when signed in.
Poland invested $11M in ElevenLabs through its state development fund Vinci, acquiring a minority stake in the $11B AI voice company and establishing Warsaw as ElevenLabs's EU headquarters and R&D center.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
