Join us November 18 at OpenAI in San Francisco for our final Startup Showcase of the year. Apply to attend.
For an engineer choosing what to work on next, “hard problem” isn’t enough. A more useful question is what becomes possible if you solve one: pesticides protecting crops while sparing pollinators, robots learning to cook dinner, AI agents working through the night, or larger satellite fleets staying connected to Earth. The four founders below are building toward those possibilities.
Philipp Wu, Co-Founder of XDOF
Try explaining how to tie a shoelace without using your hands. At some point, it’s easier to say, “Here, let me show you.” AI labs such as Google DeepMind are now adapting their models to control robots that can tackle many different tasks. But the language and images that helped those models understand the world don’t tell them everything about acting in it. They also need experience of movements and their consequences.
At Berkeley, Philipp helped create GELLO, a small, 3D-printed arm that lets a person guide a larger robot through a task. XDOF now supplies robotics teams with physical demonstrations and tests their models on real machines. With research collaborators, it has released more than 130,000 demonstrations across nearly 200 tasks. That shared foundation could let more teams try new approaches to robot intelligence without first building an operation to collect experience and test what their robots learned.
Peter Voell, Founder of Mount Thor
The race for AI is becoming a race for compute. Cloud giants and a new generation of specialized providers are securing GPUs, building data centers and competing for the power to run them. Peter Voell left OpenAI to make a bet on an overlooked major computing ecosystem: Apple.
Apple’s high-memory Macs can run large AI models, and Mount Thor is developing an inference service to make that computing power available in the cloud. Peter isn’t trying to beat NVIDIA on peak performance. He’s betting Apple Silicon could run background AI tasks at lower cost when an immediate response isn’t necessary.
Customers already use its Mac environments to develop computer-use models and deploy long-running agents. The inference platform would also supply the compute running the models themselves. The challenge is turning that hardware into a dependable, economical service other AI companies can build on.
Bridgit Mendler, Co-Founder of Northwood Space
During the pandemic, Bridgit Mendler and her husband, Griffin Cleverly, built antennas from Home Depot supplies at a New Hampshire lake house to receive weather-satellite data. Today, their company Northwood is building the network that lets satellite operators make those connections reliably, at a much larger scale.
Commercial satellite fleets are growing, but the ground infrastructure they rely on has been slow to manufacture and deploy. SpaceX is focused on the sky, not the ground. More satellites can mean more frequent images of a changing planet.. provided operators can get the data back to Earth.
Northwood builds antennas, deploys sites and runs them as a shared network. Its Portal system can communicate with multiple satellites simultaneously, and a new Torrance factory is designed to produce hundreds of antennas a year. Bridgit’s goal is for a space company to add a mission without waiting for its ground connections to catch up.
Navvye Anand, Co-Founder of Bindwell
An AI system that learns from physical experiments could change how we discover medicines, protect crops and make useful materials. That’s the promise of bringing AI into the wet lab: each experiment can test a proposed molecule and help train a model to find better ones. Navvye Anand and Tyler Rose are building Bindwell around that possibility.
They’re starting with pesticides. Global agricultural production was worth $5.2 trillion in 2024; even small improvements in protecting harvests could mean more food from the land we already farm. Bindwell’s first target is fall armyworm, a caterpillar that can devastate maize harvests. The goal is to stop the pest while minimizing harm to people, pollinators and other beneficial species.
The founders built their own lab to test candidate pesticides on insects and use the results to improve their AI. Bindwell says it can now measure the early success rate of a model’s proposed compounds in these lab tests in about 72 hours. The ambition is for what it learns about one molecule or organism to help it tackle the next—a system that becomes more capable of discovery as it experiments.
Come meet Philipp, Peter, Bridgit, and Navvye—and the rest of the lineup. Apply to attend the November 18 showcase.