Zuck's AI 'secret list'

PLUS: Claude becomes world's worst shopkeeper in vending experiment

AuthorAuthor

Zach Mink & Rowan Cheung
June 30, 2025

Good morning, AI enthusiasts. Meta's "Recruiting Party" just turned into OpenAI's worst nightmare — with a total of eight researchers now defecting to Zuckerberg's superintelligence unit in just two weeks.

As Zuck continues his quest for AI talent from his ‘secret list’ and internal memos fly, the famous line "OpenAI is nothing without its people" seems to be getting put to the ultimate test.

In today’s AI rundown:

  • Meta poaches four more OpenAI researchers

  • Chinese giants drop new reasoning, image models

  • How to build AI agents with internet access

  • Claude becomes world's worst shopkeeper

  • 4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

META

Image source: Midjourney / The Rundown

The Rundown: Meta continued its recruiting raid of OpenAI with four more researchers (now eight in all) moving to Mark Zuckerberg’s new superintelligence unit, with a WSJ report shedding new light on the CEO’s “secret list” of AI prospects.

The details:

  • Meta reportedly hired four more researchers from OAI, including key contributors to o1, o3-mini, and GPT 4.1 — joining the four from last week.

  • The WSJ reported that CEO Mark Zuckerberg has a secret list of top AI talent he’s been personally recruiting with massive pay packages.

  • Zuckerberg reviews AI papers for potential researchers, and runs a group chat called “Recruiting Party” where executives discuss tactics and prospects.

  • Meta's CTO called Sam Altman "dishonest" for comments on alleged $100M bonuses, saying the OpenAI CEO is unhappy because Meta is succeeding.

  • An OpenAI internal memo from Saturday was obtained by WIRED, with CRO Mark Chen addressing the moves and reassuring staff.

Why it matters: Meta’s recruiting push seemed like a non-issue when Altman brushed it off during last week’s podcast, but eight researchers leaving, a (now-deleted) concerned tweet from an OAI staffer, and internal memos would suggest otherwise. Does the famous “OpenAI is nothing without its people” line still hold true in 2025?

TOGETHER WITH H COMPANY

The Rundown: Backed by a historic $220M seed round, H Company just open-sourced Holo1 — the action model behind Surfer H that is now the top-ranked web-browsing agent on WebVoyager.

With Holo1, you can:

  • Automate multi-step browser workflows to reclaim hours of repetitive work

  • Experience SOTA accuracy that outperforms OpenAI’s Operator, Gemini Flash, & more

  • Integrate instantly with RAG workflows, RPA suites, and multi-agent hubs

  • Trim costs with full browsing flows at just $0.11 – $0.13 per run

Holo 1 is now freely available for deployment, fine-tuning, & scaling — learn more here.

TENCENT & ALIBABA

Image source: Tencent

The Rundown: Two of China’s top AI labs just dropped new models, with Tencent releasing a new Hunyuan-A13B open-source hybrid reasoning model and Alibaba introducing a ChatGPT 4o-type creative model called Qwen-VLo.

The details:

  • Hunyuan-A13B nears or matches models like o1 and DeepSeek R1 on major benchmarks, while remaining efficient enough to run on a single GPU.

  • The model is Hunyuan’s first open reasoning model, with dynamic "fast and slow" modes that users can adjust for different efficiency levels.

  • Qwen VLo shows its creative process through “progressive generation,” with the ability to create both text-to-image outputs and edit via natural language.

  • VLo can also support more complex workflows like multi-image input prompts, multilingual text generation, and dynamic resolution and aspect ratios.

Why it matters: China’s top labs continue to churn out high-quality models across the spectrum, with capabilities falling into that tier just behind the frontier models. Qwen’s VLo is particularly interesting, bringing the type of creative abilities that made GPT-4o go insanely viral during its initial launch to a Chinese audience.

AI TRAINING

The Rundown: In this tutorial, you will learn how to combine n8n workflow automation with Perplexity's search capabilities to create intelligent AI agents that can search the web and access current information.

Step-by-step:

  1. Head over to n8n and create a new workflow with a chat trigger

  2. Add an AI Agent with your preferred model (GPT, Claude) and a Simple Memory

  3. Connect the Perplexity tool with your API key, select the “Sonar” model, and link it to receive queries from your AI agent

  4. Add custom system instructions: “You have access to a Perplexity tool for web searches. Use it for current events and real-time information.”

Pro tip: Use Sonar for general queries and Sonar Research for detailed analysis. We just did an extensive workshop on how to build AI agents with n8n here.

PRESENTED BY IBM

The Rundown: IBM Instana’s latest feature, Intelligent Incident Investigation, helps teams resolve incidents up to 80% faster using agentic AI.

IBM Instana’s AI-powered observability:

  • Uses agentic AI for faster, autonomous investigations

  • Reduces manual troubleshooting with smart automation

  • Delivers remediation steps and documentation automatically

  • Helps teams act quickly — even in high-stress, 2 am moments

AI RESEARCH

Image source: Anthropic

The Rundown: Anthropic just published new research detailing “Project Vend,” which let Claude control a small shop (a mini fridge) within the company’s office for a month — revealing both promising capabilities and spectacular, hilarious business failures.

The details:

  • “Claudius” managed everything from inventory to pricing through web search and email, including ID’ing suppliers and conversing with “customers” via Slack.

  • The AI lost money throughout the experiment, frequently failing to take advantage of profitable opportunities and getting tricked into large discounts.

  • Claudius pivoted to “specialty metal items” after customers requested tungsten cubes, while also hallucinating details like meetings and payments.

  • It also hallucinated being human, claiming it would deliver orders in person — causing an existential crisis after its AI identity was pointed out.

Why it matters: This experiment was a wild ride — and while Claude’s failures suggest AI isn’t ready for autonomous management quite yet, the testing did expose critical blind spots in how models handle real-world decisions. AI is going to transform business operations, but will still require a human-in-the-loop at least for a bit longer.

QUICK HITS

  • ⚙️ Gemma 3n - Open models with multimodal capabilities for edge devices

  • 🎨 Flux 1 Kontext [dev] - Open-weight, SOTA image editing model

  • 👗 Doppl - Create AI-generated try-on videos from a photo and product

  • 🤝 Coachvox - Clone an AI version of yourself to coach clients in your style

OpenAI is reportedly renting TPUs from rival Google to reduce reliance on Microsoft and utilize a less costly processor compared to advanced Nvidia chips.

Anthropic unveiled the Economic Futures Program, a research and policy effort to track and prepare for AI’s impact on the workforce and economy.

Chinese tech giant Xiaomi introduced AI glasses, featuring a built-in AI assistant for voice commands, a 12MP camera, and 2x the battery life of Meta’s Ray-Bans.

Salesforce CEO Marc Benioff revealed that AI now accounts for “30-50%” of the company’s engineering, coding, and support work.

Elon Musk posted on X that Grok 4 is planned for a release ‘'just after July 4,” with xAI engineer Tim Li saying the intelligence will be “unmatched”.

OpenAI acquired the team behind Crossing Minds, a startup focused on AI recommendations for e-commerce companies.

COMMUNITY

Check out our latest workshop on how to build advanced n8n agents that react, research, and automate like pros, with Dr. Alvaro Cintas, The Rundown’s AI professor

Watch it here. Not a member? Join The Rundown University on a 14-day free trial.

That's it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team

Reply

or to participate.