Most products are finished the day they ship. Physical AI makes a different kind of object possible, one whose behavior keeps unfolding through use, shaped by the people and places around it. That idea sits at the center of Living Objects: Designing Products That Grow With Us, the talk our co-founder and Chief Design Officer Leonardo Giusti is giving at KIKK Festival on Oct. 23. At KIKK, Leo will share three objects developed by Archetype AI and the MIT Design Intelligence Lab: a lamp with memory, a radio that gives voice to the conditions around it, and a set of tags that connect everyday objects. Each one is designed to stay open-ended. The question he’ll leave the room with: what should be decided in advance, and what should be left for life to shape?

Physical systems are becoming increasingly complex, and it takes more and more specialized expertise to run them. At the same time, experienced workers are retiring, and hiring in physical industries like manufacturing is hard. The push for more energy, more GPUs, and more data centers only widens that gap. On Inside the Silicon Mind, our CEO @ipoupyrev says the answer is to pair AI with the workers and operators who run these systems. Operators bring the expertise, and agents turn data into insights. Watch the full episode: youtu.be/ZpOOpAOpc_Y?si=cvPD…

A new customer shouldn’t mean a new model. But that’s the trap for a lot of Physical AI and robotics startups. Each new site brings new machines, new sensors, and new operating conditions. With custom ML, that means labeling new data and building another model. Newton works differently. It’s one pretrained model that works across sensors, machines, and sites, and adapts from a handful of examples. In one manufacturing deployment, a single Newton model ran zero-shot across every motor type, from different sizes and manufacturers, and caught gearbox faults early. So your tenth customer runs on the same foundation as your first. Need a new capability? Add a Newton Agent on the same model. Start building on Newton: archetypeai.io/build-with-us

Physical AI can’t run like an LLM in a data center. There isn’t enough power for that. An alternative @ipoupyrev is proposing: train in the cloud, deploy to the edge, run online or fully offline. Decisions happen in real time, and your data stays yours.

Our Chief Scientist @_jaimelien_ joins the panel at Long Horizon in San Francisco on Oct. 20 to take on one question: “Will World Models Eat Physical AI?” “World model” is a broad term; some are built to generate or simulate the physical world. Newton is a physical world model that learns from real-world sensor data, so it’s built to understand what is actually happening on a machine or a factory floor. That’s the view Jaime brings to the debate. ➡️ In SF? Join us: archetypeai.io/resources/eve…

The most critical operational events often have very little historical data, making them difficult for traditional AI systems to understand. Our Rare Event Detection agent, powered by Newton, learns the unique patterns associated with known faults, failures, or operational events from just a handful of labeled examples, enabling teams to identify critical conditions before they become larger operational problems. Learn more: archetypeai.io/agents

Imagine asking a motor how it’s doing, and an AI agent giving you a response in natural language. That’s how @ipoupyrev described the type of world model we’re building on Inside the Silicon Mind. Sensors already measure things like temperature, rotation speed, vibration, current and flow. Physical AI is capable of reading those signals and tells you what’s going on inside the machine, in words you can act on. Ivan and the hosts get into how it all works. Have a listen: youtube.com/watch?v=ZpOOpAOp…

Most Physical AI today means one model per signal, per task, per site. Radio is the extreme case. Decoding, localization and transmitter ID each get a custom model, built and validated from scratch. Member of our research team, Jonah Casebeer and @_jaimelien_ , asked a simple question. Can one general model beat the specialized models built for RF signals? Newton beat models purpose-built for wireless on 18 of 20 tasks. The two exceptions were beam-selection tasks, where a model built for wireless channels came out ahead. Then we tested the same model on vibration, audio, EEG and seismic data. It led on 15 of 16 tasks. The radio-only models did not carry over. We are working toward one model that can read any sensor and any use case, without a new project for every one. The chart below shows win rates by task category. Newton is the blue line, at the outer edge in every category, radio and non-radio. Read the full research: archetypeai.io/blog/newton-r…

We have another open role! 🚀 We are #hiring someone with 10+ years of experience in B2B demand generation who has experience building a repeatable, marketing-sourced pipeline engine from founder-led or early-stage sales. If this sounds like you, reach out to our team here: careers.kula.ai/archetype-ai…

Our mission is to help humanity make sense of the real world. Making sense of it only matters if it leads to action. More than half of industrial downtime still traces back to manual handoffs. The next step for Physical AI is closing that loop, from perception to action to verification. On Oct 15, our CEO @ipoupyrev takes the stage at TechCrunch Disrupt in San Francisco. He'll break down what it takes for AI to handle the messiness of the real world. 🔗 techcrunch.com/events/techcr…

Across 63 wind turbines, Newton found nine failure modes no one had seen before. Anomaly Discovery Agent, powered by our world model, Newton, reads patterns across sensor streams and surfaces the issues no single signal stream can identify. Newton did this out of the box: no failure history to label first, and no model built turbine by turbine. archetypeai.io/agents

Most AI models are trained on text and images. We're building a model trained on millions of sensor measurements; this model, Newton, is created to understand the physical world. Over the next month, we're taking Newton on the road to meet the teams who keep turbines, pumps, and production lines running. In one wind deployment, Newton fused 40+ sensors across 63 turbines out of the box and surfaced 9 previously unknown failure patterns. That's what a world model trained on real sensor data can do, and we'd like to show you in person. The full schedule is in the graphic. If you'll be at any of these stops, message us to set up a demo.