Meta's major AI restructure

PLUS: Automate performance reviews and PIPs with ChatGPT

AuthorAuthor

Zach Mink & Rowan Cheung
August 22, 2025

Good morning, AI enthusiasts. Meta’s summer AI hiring spree looks to be complete, and the high-priced new team is already making big organizational moves.

With a massive restructure underway that includes dissolving entire teams and a mysterious new “TBD Labs”, Meta’s quest for superintelligence is starting with major changes from the ground up.

P.S. Our next live workshop is today at 4 PM EST with The Rundown’s AI Educator, Nate Grahek. Learn how to use AI as an operating system for the future of work: RSVP here.

In today’s AI rundown:

  • Meta’s massive AI restructure

  • Google analyzes Gemini’s environmental footprint

  • Automate performance reviews and PIPs

  • Musk: Grok 5 has ‘a shot at being true AGI’

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

META

Image source: Getty Images / Reuters

The Rundown: Meta is undergoing a massive restructure of its AI teams, dissolving its AGI Foundations division and reorganizing operations into four units under Alexandr Wang — with the company also imposing a hiring freeze after a major poaching spree.

The details:

  • Wang sent a memo to employees outlining new teams for research, training, products, and infrastructure, with most division heads reporting directly to him.

  • The company froze hiring across its AI division last week, now requiring Wang’s personal approval for any exceptions to the mandate.

  • The AGI Foundations team is being scattered across departments, with Meta also creating a ‘TBD Lab’ to explore “omni” models and frontier AI research.

  • Wang revealed that Chief Scientist Yann LeCun will now report to him as well, describing FAIR as the “innovation engine for MSL” in the new structure.

Why it matters: Meta’s summer of hiring looks to be officially over, with the focus now turning to building a new internal structure under the direction of Alexandr Wang. It’s clear that the high-profile new team wants to move fast — what isn’t clear is how the changes will sit with the broader AI and FAIR teams that now feel lost in the shuffle.

TOGETHER WITH CLOUDERA

The Rundown: Join industry leaders at EVOLVE25 NYC on Sept. 25th to explore how to transform AI potential into measurable business outcomes. Cloudera will showcase how leading enterprises unify, manage, and leverage their data to power intelligent transformation at scale.

At EVOLVE25, you'll experience:

  • A deep dive into trends for data, analytics, and AI with industry analysts

  • Customer-led sessions on enabling secure data architectures and analytics

  • Hands-on labs on implementing secure AI from development to inference

  • Exclusive announcements from Cloudera’s leadership

GOOGLE

Image source: Google

The Rundown: Google released a new blog detailing the environmental footprint of its Gemini chatbot, claiming the model consumes the equivalent of five drops of water per query — though researchers argue it left out most of the actual water usage.

The details:

  • The published findings claim each Gemini text request uses energy equal to watching TV for nine seconds and creates minimal carbon emissions.

  • Google said Gemini became 33x more energy efficient and cut carbon output by 44x over the past year, all while the models became more capable.

  • The paper found that A Gemini query consumes 0.24 Wh of energy, slightly lower than the 0.34 Wh average that Sam Altman revealed for ChatGPT.

  • Researchers criticized the study for ignoring water consumed by power plants that generate power for data centers, which represents the majority of usage.

Why it matters: While Google’s efforts to provide more transparency around AI’s environmental impact (a key issue for AI detractors) are positive, not everyone agrees with the company’s process, which may be painting an artificially rosy outlook. An industry-wide third-party standard may be needed to truly understand the full picture.

AI TRAINING

The Rundown: In this tutorial, you will learn how to connect ChatGPT to your Google Drive knowledge base and then use agent mode to automatically transform messy employee notes into structured performance reviews and improvement plans.

Step-by-step:

  1. Create a Google Doc called "Employee Review" with rough notes for each employee (name, role, KPIs, feedback)

  2. In ChatGPT, click "+" → "Connected apps" and enable Google Drive access

  3. Prompt with system instruction + templates: "Analyze my 'Employee Review' doc and create performance reviews and 30/60/90 PIPs using these formats: [paste templates with SMART goals structure]"

  4. Enable Agent Mode and prompt: "Update the Employee Review doc with the structured sections for each person"

  5. Schedule monthly automation: "Run third Saturday at 9 AM, update Employee Review doc from latest notes, regenerate reviews and PIPs"

Pro tip: Keep adding rough notes weekly — the scheduled agent consolidates and formats monthly. If the agent can't edit initially, grant permissions and re-run. This beats manual updates by preventing skipped review cycles.

PRESENTED BY HYPRNOTE

The Rundown: Hyprnote is the privacy-first AI notepad that records conversations locally and converts quick notes into clear, share-ready summaries. No audio or text leaves your stack, giving regulated teams complete control.

Hyprnote’s features include:

  • Open-source, self-hostable

  • 100% local or on-prem deployments

  • The ability to bring your own AI models

XAI AND ELON MUSK

Image source: Elon Musk (@ElonMusk on X)

The Rundown: Elon Musk had a busy day of AI commentary on X, revealing new information about Grok 5, making bold claims about xAI’s ‘Imagine’ generator, and speaking on AI and declining birthrates in a series of posts and replies on the platform.

The details:

  • Musk posted that xAI’s Grok 5 model will begin training in September, saying he believes the model “has a shot at being true AGI”.

  • He also said Grok Imagine will be better than Google’s VEO 3 video generation model “in every respect, with no exceptions”.

  • Musk also commented on the declining birthrate, saying AI will actually increase birth rates and will be “programmed that way”.

Why it matters: AGI is a benchmark without a very clear definition, which will make the first official declaration of it all the more interesting. With OpenAI being the other major lab dancing around the notion of its models officially reaching the bar soon, the term could end up being the topic of the next inevitable feud between Altman and Musk.

QUICK HITS

  • ⚙️ Claude Code - Anthropic’s agentic coding tool

  • 📊 Copilot in Excel - Bring AI to your spreadsheet formulas

  • 🔎 AI Mode - New agentic capabilities for search in Google

  • 🌎 Game Worlds - Runway’s Beta release for generating text-based games

Google is expanding access to its AI Mode for conversational search, making it globally available, alongside new agentic abilities for handling restaurant reservations.

Cohere released Command A Reasoning, a new enterprise reasoning model that outperforms similar rivals like gpt-oss and DeepSeek R1 on agentic benchmarks.

Runway introduced Game Worlds in beta, a new tool to build, explore, and play text-based games generated in real-time on the platform.

ByteDance released Seed-OSS, a new family of open-source reasoning models with long-context (500k+ tokens) capabilities and strong performance on benchmarks.

Google and the U.S. General Services Administration announced a new agreement to offer Gemini to the government at just $0.50c per agency to push federal adoption.

Chinese firms are moving away from Nvidia’s H20 and seeking domestic options after being insulted by comments from U.S. Commerce Secretary Howard Lutnick.

COMMUNITY

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Bill B. in Seattle, WA:

"As a Chartered Financial Analyst (CFA) and Fractional CFO for multiple organizations, I’ve integrated AI into my financial review process to deliver faster, sharper insights. By securely uploading client financial statements into ChatGPT and prompting for targeted analysis and actionable recommendations, I’ve cut review time by up to 90% — turning what once took days into just hours.”

How are you using AI in your work? Tell us here.

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, Shubham, and Jennifer — the humans behind The Rundown

Reply

or to participate.