🔥 Two Founders Return to OpenAI

Plus Google reads your entire digital life and Bezos backs robots

January 16, 2026

Reading time: 5 minutes

🗞️In this edition

  • $2 Billion Startup Loses Two Co-Founders to OpenAI

  • Google Gemini Gets Access to Your Entire Digital Life

  • Bezos and Nvidia Back Robot Startup With $1.4 Billion

  • In other AI news –

    • OpenAI Hires AWS's Top AI Sales Leader

    • Lego Bricks Now Have Sensors and Speakers Inside

    • US Approves Nvidia AI Chip Sales to China

  • 4 must-try AI tools

Some moves reveal more than they're meant to.

A startup losing two co-founders back to the company they all left together. An AI getting access to your entire digital history. And a robotics company raising over a billion dollars from names that don't typically agree on investments.

Each story shows something about control. About what companies think they need to win. About the calculations being made behind the announcements.

The pattern isn't what you'd expect. The company spending the most isn't necessarily the one with the strongest position.

This edition looks at what's actually happening when you read past the press releases.

What's happening:

Thinking Machines Lab lost two co-founders this week. Both are returning to OpenAI months after leaving to start the company with Mira Murati.

Barret Zoph, the CTO and co-founder, is gone. So is co-founder Luke Metz. Another researcher, Sam Schoenholz, also left for OpenAI.

Murati announced Zoph's departure Wednesday. An hour later, OpenAI's applications CEO posted that all three were rejoining. She said it had been "in the works for several weeks."

Thinking Machines raised $2 billion in July at a $12 billion valuation. This is the second co-founder they've lost. Andrew Tulloch left for Meta in October.

Why this is important:

This shows how hard it is to compete with OpenAI when you're made of former OpenAI people. The gravity pull is strong.

Losing your CTO and a co-founder less than a year after raising $2 billion signals deeper problems than normal talent movement.

OpenAI clearly wants these people back. They were planning this for weeks. That's targeted recruitment, not coincidence.

Murati left OpenAI in September 2024. By January 2026, two of her three co-founders are back at OpenAI. That's barely a year.

Comments from the editor:

Watch the announcement timing. Murati posts about one person leaving. An hour later, OpenAI reveals it's actually three. That's not how coordinated exits work.

Here's the brutal part. OpenAI said this was "in the works for several weeks." That means Murati's co-founders were planning their escape while still building the company with her.

This isn't normal startup churn. This is losing people back to the exact company you all left together. That suggests either OpenAI made an offer they couldn't refuse, or Thinking Machines wasn't what they thought it would be.

Raising $2 billion buys you office space and GPUs. It doesn't buy you loyalty when your founding team wants out.

What's happening:

Google launched Personal Intelligence for Gemini on Wednesday. The feature connects Gmail, Photos, YouTube, and Search history to personalize AI responses.

Personal Intelligence is off by default. Users with AI Pro or AI Ultra subscriptions can turn it on and choose which apps to connect.

Gemini can now reason across your data. If you ask for weekend plans, it pulls from your past trips, photos you took, and things you searched for to make suggestions.

Josh Woodward, Gemini VP, gave an example. He needed tire specs while standing in line. Gemini found his car model from Gmail, checked Photos for road trip conditions, and suggested specific tires.

Google says the data isn't used to train AI models. They filter personal details before using prompts to improve the system. You can disconnect apps or delete history anytime.

Why this is important:

This shows Google betting that personalization beats generic AI responses. They're using the massive data advantage they already have from running Gmail, Photos, and Search.

The opt-in approach matters. After backlash over Gmail AI scanning, Google learned to make data access explicit. But most users don't read settings carefully.

For competitors like OpenAI, this is a problem. ChatGPT has memories from conversations. Gemini has your entire digital life across Google services. That's hard to compete with.

The privacy claims feel optimistic. Google says they don't train on your inbox. But they're still reading it to answer questions, which requires the same access.

This makes Gemini much stickier. Once it knows your car, preferences, and history, switching to another AI means starting over. That's exactly what Google wants.

What's happening:

Skild AI raised $1.4 billion on Tuesday, valuing the Pittsburgh robotics startup at over $14 billion. That's triple its valuation from just seven months ago.

SoftBank led the round. Nvidia, Jeff Bezos, Samsung, LG, and Salesforce also invested. Existing backers like Sequoia and Lightspeed doubled down.

Skild AI builds the "Skild Brain," software that can operate any robot without being trained on that specific robot first. Humanoids, quadrupeds, warehouse arms, they all use the same brain.

The company hit $30 million in revenue within months of launching in 2025. It's deploying the tech for security, facility inspection, delivery, and warehouse automation.

The challenge Skild solved is that robots don't have an internet of training data like language models do. So they train the Skild Brain by watching human videos and running simulations with 100,000 different robot forms.

Why this is important:

This shows investors betting that one universal robot brain beats specialized software for each robot type. That's a massive assumption that hasn't been proven at scale.

The $14 billion valuation after barely a year of revenue is staggering. It reflects hype around "physical AI" but also real traction with enterprise customers.

For companies buying robots, this matters. If Skild Brain works, you don't need custom software for every machine. That changes deployment economics drastically.

SoftBank leading is notable. They just bought ABB's robotics business for $5.4 billion. Skild AI fits their strategy of controlling both hardware and software in robotics.

The real test is whether learning from human videos actually translates to robots performing reliably in unpredictable real-world conditions. Simulations are one thing. Factories and hospitals are another.

  1. Rizemail - An AI-powered email summarization tool that helps users get to the core of their unread newsletters and long email threads

  2. Kastro Chat - An AI-powered chatbot platform that allows businesses to create their own chatbots without any coding knowledge

  3. Verbalate - A video translation and lip sync software designed to help businesses reach a global audience

  4. Taranify - A platform that uses AI technology to provide mood-based recommendations for music, Netflix shows, and books

The decisions keep getting more aggressive.

Poaching back your own co-founders. Connecting AI to every app and photo. Betting billions on technology that hasn't proven itself at scale yet.

None of these are cautious moves. They're bets that the window for positioning is closing fast, and whoever moves hardest now wins later.

What matters is whether the people making these decisions have it right. Or whether they're overextending based on assumptions that won't hold.

If you have any feedback for us, please reply and let us know how we did. We're always looking to improve.

How did we like this version?

Login or Subscribe to participate in polls.

Interested in featuring your services with us? Email us at [email protected]