Every great technology wave creates a new kind of entrepreneur. We believe Physical AI will create data entrepreneurs. That’s why we built @DeepReach_AI - a platform where entrepreneurs build local data businesses for Physical AI. Today we’re excited to launch on @ycombinator 🚀 YC launch link: ycombinator.com/launches/SiT… #PhysicalAI #Robotics #YC

DeepReach DR-H01.5 has landed in the US! 🇺🇸 The world’s most stable egocentric stereo capture device. Rugged metal body, built for real-world data capture. Connecting Physical AI to the real world. deepreach.ai/devices #PhysicalAI #Robotics #EgocentricData #RealWorldData #DeepReachAI

Adcock understands this. In 2025 he walked away from OpenAI and called LLMs "the smallest piece of the puzzle." Now he's building Hark, an AI model and assistant company, alongside Figure. It's the same playbook: separate companies, one ecosystem. The rule is simple: the game belongs to the LLMs. Unless you have enough capital to build a second ecosystem and fight for it. Musk has one. Adcock is building one. Almost no one else can. But there's still a gap. Driving data teaches a model how the world moves. Text teaches it what things are. Neither teaches a hand how to fold a towel. Physical interaction data is still scarce. So the endgame isn't LLMs replacing robotics. It's a merger. Whoever owns the world model and whoever owns physical data will end up on the same side of the table. #PhysicalAI #Robotics #Humanoids #WorldModels #LLM #EmbodiedAI #Tesla #FigureAI #Optimus #AI

Join me at the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)! invt.io/1txbcd8pobv

lol

This is a massive signal for Physical AI. Figure just committed to spending $1B+ on data and compute over the next 12 months, with a clear thesis: scaling general-purpose robotics requires massive amounts of diverse, real-world human data. This is exactly the problem we built DeepReach AI to solve. Instead of building another centralized data farm, we’re building a global data business network — enabling local entrepreneurs across 20 countries to capture diverse human skills, environments, and physical interactions from the real world. Figure is one company. The bigger question is: how many robotics foundation model companies will be making billion-dollar-scale investments in real-world data next? The scaling law for Physical AI is becoming clear. And someone needs to build the global supply infrastructure behind it. That’s what we’re building at DeepReach AI. #PhysicalAI #Robotics #EmbodiedAI #RealWorldData #RobotLearning #AI

This is fun, I lost $5 on this bet.

This is way better than use in the war.

We shouldn’t only think about robots as products people buy. Robots can become economic participants —creating entertainment, content, services, and entirely new markets.

We just won a multi-million dollar contract. Our product is watching people work. Three months ago people asked why anyone would pay for videos of a bakery. Now you know. Learn my startup store here: ycombinator.com/launches/SiT… #PhysicalAI #Robotics #YCombinator

Attended Ken Goldberg’s ICRA keynote. His “two cultures” framing (after C.P. Snow) is the cleanest articulation I’ve seen of where robotics sits in 2026. Model-based (classical control, Kalman/PID/motion planning—modular, verifiable) vs. model-free (VLA: pixels in, actions out, end-to-end). He believes in the bitter lesson, but reframes the question from whether scale wins to when. Three constraints he laid out: (1) Data gap is structural. Robot state spaces hit 50+ DoF (bimanual humanoids); training data trails LLMs by ~5 orders of magnitude. And data ≠ everything—Waymo > Tesla on operational metrics with far less of it. (2) VLA “pseudo-generality.” Open X-Embodiment’s 160K tasks are mostly pick-and-place under the hood. Brittleness is the tell: 100% on LIBERO → ~17% when the target shifts 2cm. OOD = failure. (3) The synthesis: GAP. Agents auto-generate decomposable, verifiable ROS-style computation graphs, self-improve via execute→evaluate→revise, then distill to lightweight deployable graphs. 95–98% on cable insertion, pot-on-stove, robust to variation—well past VLA brittleness on the same benchmarks. The concept I find most useful: “variational automation.” Between fixed automation and full generality—same task, varying initial conditions, structurally similar. The regime where the data flywheel is actually economically viable. Takeaway: VLA hasn’t reached robotics’ ChatGPT moment. The near-term path is agent-generated verifiable structure—classical reliability × large-model generality—not pure scale.

Why Filtering Makes the Dataset: EgoView vs EgoVerse vs EgoDex deepreach.ai/blogs/egoview-v…

Most egocentric datasets train robots on 5s tabletop tasks. Real work doesn’t look like that. We built EgoView, industrial egocentric data with ~20s+ long-horizon, multi-step workflows + strict filtering for training-grade quality. If you’re building physical AI, this matters ↓ deepreach.ai/blogs/egoview-v… #PhysicalAI #Robotics #RobotLearning #AIData #TrainingData #EmbodiedAI #DeepTech #AIInfrastructure