AGI House Presents: World Models and Sim-to-Real Intelligence

Agent Skills, OpenClaw, World Models, & the Next Frontier

Last weekend, AGI House convened researchers and builders at the frontier of agent skills and world models, including Shlomi Fruchter (Research Director, Genie 3 @ Google DeepMind), David Luan (Fmr CEO, Adept), & many more.

The conversation kept returning to one tension: most world models predict pixels, not causes. Sharon Lee's Moonlake is making a different bet — physics enforced, not approximated.

A few things that stayed with us:

• RL is powerful enough — if the environments are right. The bottleneck isn't the algorithm. It's the data.

• Gaming is the trojan horse. Billions of hours of action-labeled 3D data. No lab can replicate it.

• The missing layer isn't the model. It's the curriculum that diagnoses failure and iterates automatically.

Also in the room: Sharon Lee (Co-founder, Moonlake AI), Kylan Gibbs (CEO, Inworld), Xiaoyin Qu (SkillBoss), Dhruv Batra (Yutori), Xiangyi Li (BenchFlow), Belinda Mo (Sundial), Umesh Khanna (xAI), and Amber Liu (Orchestra Research). Hosted with xAI, Sundial, Inworld, Moonlake, SkillBoss, and Yutori.

Read the post-event memo below:

Edelman Workshop @ AGI House

On March 18th, a select group of AI founders and executives spent an evening at AGI House with senior leaders from Edelman, the world's largest independently owned communications firm.

A few things that stuck:

• The best pitch is one you're not in the room for. If a stranger can repeat your story accurately the next morning, you've succeeded.

• Consistency beats volume. The founders who break through aren't necessarily the loudest, they're the ones who show up regularly with something genuine to say.

• AI changes the work, not the relationships. A well-placed call from someone a reporter actually knows still opens doors that no automated pitch can.

Please contact [email protected] to learn more.

Poker Nights @ AGI House

We've been playing a few more hands of poker recently (thanks MongoDB 🃏). Two weeks ago, we hosted Dex and fifty engineers and founders for an all-in poker night. The House always wins and we plan to keep hosting.

AGI Lunch Series: Ariyan Kabir, Founder @ GrayMatter Robotics

"In a Tesla Model 3 factory, the only thing you can build is a Tesla Model 3. What we're building toward is a fully reconfigurable autonomous factory — so that at a time of conflict, it can serve the national interest, and in peacetime, serve society."

Ariyan Kabir, founder of GreyMatter Robotics, stopped by for lunch last week. His company is building autonomous solutions for industrial manufacturing — aerospace, defense, specialty vehicles, submarines.

The insight that stuck: unlike most robotics applications, you can't simulate your way to precision manufacturing. The physics are too complex, too variable, too environment-dependent. The only path is real robots in real factories, building a data flywheel that no one else can replicate.

Job Board

We’re Hiring And So Are Our Portfolio Companies

▶ AGI House is building the infrastructure for AGI and is looking for exceptional people to join our core team.

We’re hiring - all full-time roles:

• AI Researcher to work on our social agents sim-2-real world model project

• AI Investor to invest in self-improving agents, world models, and physical AI - background at a top-tier firm or as an AI researcher preferred

If you’re interested, email [email protected]

Stay Updated with Us