Joseph Krause is co-founder and CEO of Radical AI, which combines AI with robotic laboratories to discover and test materials. He argues that physical experiments and later manufacturing—not model predictions alone—determine whether a candidate material can reach an industrial product. — Latent Space — The Self-Driving Lab.

Part 1: The Bottlenecks of Materials Science
- On Materials as Technology's Foundation: Krause argues that progress in fields such as aerospace, energy, and electronics can depend on developing materials with properties that existing ones cannot provide. — Upstarts — Joseph Krause Interview.
- On the Cost of Materials R&D: Krause says a new material system can require more than a decade and over $100 million to develop, helping explain why firms often favor incremental improvements. — Open Source CEO — Joseph Krause Interview.
- On a Fragmented Pipeline: Krause describes materials research as split among academic discovery, small-scale testing, and large companies optimizing established systems; data and know-how are often lost between those stages. — Latent Space — The Self-Driving Lab.
- On Materials Beyond a Formula: Krause contrasts some molecules that can be represented by strings with structural alloys, whose performance also depends on supply chains, microstructure, and processing methods. — Latent Space — The Self-Driving Lab.
- On Exploring Untried Alloys: High-entropy alloys span many element combinations; Krause says Radical AI's system has tested combinations that materials scientists had not pursued because they expected them to fail. — Latent Space — The Self-Driving Lab.
- On Supply-Chain Constraints: Krause says customers have asked Radical AI to redesign an aerospace alloy without hafnium because its supply and cost have become concerns; he presents this as a materials-design problem as well as a sourcing problem. — Latent Space — The Self-Driving Lab.
- On Customer Problems Before Blue-Sky Research: Krause says Radical AI first targets material problems customers already need solved, using that work to prove its discovery system before pursuing broader enabling technologies. — Startup Project — Joseph Krause Interview.
- On the Experimental Bottleneck: Krause argues that materials discovery is constrained more by the rate and quality of physical experiments than by compute; a closed lab feedback loop is therefore central to his approach. — Latent Space — The Self-Driving Lab.
Part 2: Autonomous Labs and the Data Advantage
- On Serial and Parallel Research: Krause contrasts a scientist's sequential reading, synthesis, testing, and interpretation loop with an AI system that can compare many papers and experimental images in parallel; the physical experiments still take time. — Latent Space — The Self-Driving Lab.
- On Lost Experimental Knowledge: Krause says failed or unpublished experiments often remain in notebooks or scientists' heads, leaving later researchers without the negative data needed to learn from past attempts. — Startup Project — Joseph Krause Interview.
- On Capturing Negative Results: Radical AI records outcomes from its synthesis, characterization, and early property tests—including negative results—so later campaigns can use those observations to revise hypotheses. — Latent Space — The Self-Driving Lab.
- On Physical Ground Truth: Krause argues that a model's proposed composition is not enough: the candidate must be made, characterized, and tested before the team can know whether it is useful. — Latent Space — The Self-Driving Lab.
- On Lab-Instrument Interfaces: Krause says instrument vendors have often lacked usable APIs or raw-data access, forcing Radical AI to build software and custom hardware to connect tools into an autonomous workflow. — Latent Space — The Self-Driving Lab.
- On Encoding Composition for Optimization: In the Materialism interview, Krause explains that the team uses language-model embeddings to represent composition descriptions for Bayesian optimization, rather than relying on a raw phrase such as a titanium percentage. — Materialism — Closing the Discovery Loop.
- On Autonomy Beyond Individual Instruments: Krause describes a self-driving lab as a coordinated system that plans a research campaign, runs synthesis and characterization, collects results, and uses them to shape the next campaign—not merely a faster machine. — Latent Space — The Self-Driving Lab.
- On Teaching Scientific Intuition: Radical AI asks materials scientists to annotate microscopy images and other results so its AI system can learn the judgments experts make when examining an experiment. — Latent Space — The Self-Driving Lab.
- On Specialized Scientific Agents: Krause expects an AI scientist to combine specialist tools and agents—for literature, images, experiments, and other tasks—rather than be one monolithic model. — Latent Space — The Self-Driving Lab.
- On Testing Neglected Combinations: Krause says Radical AI's system has tested alloy families absent from the literature because human scientists assumed some element combinations would not cast or perform well. — Latent Space — The Self-Driving Lab.
- On Moving Scientists to Higher-Order Work: Krause says automating manual lab steps and connecting data can free scientists to define target properties and interpret results rather than format files or repeat routine measurements. — Investment Reports — Joseph Krause Interview.
- On Measured Experimental Throughput: In a 2026 interview, Krause reported 300 novel hypersonics-related material compositions generated in 16 weeks; he compared that with 3,425 unique alloys studied in the field over more than 20 years. These are company-reported figures, not independent validation. — Investment Reports — Joseph Krause Interview.
Part 3: From Lab Discovery to Manufacturing Scale
- On Processing as Part of a Material: Krause says composition alone does not determine an alloy's useful properties; casting, annealing, microstructure, and later processing can change its performance. — Latent Space — The Self-Driving Lab.
- On the Discovery-to-Manufacturing Gap: Krause says a promising small-sample result must still survive qualification and scale-up; in his June 2026 interview, Radical AI had not yet integrated manufacturing data into its loop. — Latent Space — The Self-Driving Lab.
- On Concurrent Engineering: Krause wants material design to advance alongside product design, connecting early experimental data with later integration and yield decisions rather than treating scale-up as a separate afterthought. — SemiWiki — Joseph Krause CEO Interview.
- On Learning From Production: Krause argues that manufacturing-stage measurements and know-how would make Radical AI's discovery loop more valuable, while acknowledging in 2026 that the company had not yet built that end-to-end manufacturing system. — Latent Space — The Self-Driving Lab.
- On Processing Know-How: Krause says large materials companies often protect processing methods as trade secrets; a chemical composition alone may not capture the know-how needed to make it at industrial scale. — Startup Project — Joseph Krause Interview.
- On Defining a Real Discovery: For Krause, a material candidate is not a finished discovery simply because a model predicts it or a lab makes a sample; the longer test is whether it can be manufactured and used in a real product. — Latent Space — The Self-Driving Lab.
- On Third-Party Testing: Krause said Radical AI sent promising hypersonic-alloy samples to Purdue Applied Research Institute for independent tests before attempting production scale-up; the interview reports the testing step, not a published independent result. — Investment Reports — Joseph Krause Interview.
Part 4: Go-to-Market Strategy and Commercialization
- On a Staged Full-Stack Ambition: Krause's long-term goal is a materials company that discovers, makes, and sells new materials; in 2026, he described selling access to the self-driving lab and AI software as the first go-to-market path, while manufacturing scale-up remained ahead. — SemiWiki — Joseph Krause CEO Interview.
- On Choosing a Customer-Led Beachhead: Krause says Radical AI starts with material problems customers already face, such as high-temperature alloys for hypersonics, before pursuing more speculative enabling materials. — Startup Project — Joseph Krause Interview.
- On Public and Private Partners: Krause reports both a U.S. Air Force materials contract and private-sector relationships across aerospace, energy, and semiconductors; he presents these as different channels for learning what applications need. — Investment Reports — Joseph Krause Interview.
- On Public Investment in Frontier Labs: Krause expects government and national laboratories to help bridge the materials-discovery gap by investing in self-driving labs and public-private partnerships. — Latent Space — The Self-Driving Lab.
- On Models Versus Experimental Data: Krause says Radical AI shares some models because he expects models to become more widely available; he sees the proprietary experimental dataset and lab execution as the company's edge. — Latent Space — The Self-Driving Lab.
- On Strategic Investors: Krause says investors including RTX Ventures, NVIDIA, Eni, and Universal Display have helped Radical AI understand future industry requirements, not just provide capital. — Investment Reports — Joseph Krause Interview.
- On Customer Trials: Krause says letting prospective partners run their own research on Radical AI's lab is a practical way to show the platform's capabilities and learn from real material problems. — Investment Reports — Joseph Krause Interview.
Part 5: Leadership, Culture, and Building a Deep Tech Startup
- On a Concrete Vision: Krause says a founder should make the company's vision concrete while staying flexible about the details of how to reach it. — Open Source CEO — Joseph Krause Interview.
- On Resource Pressure: Krause argues that abundant capital can make a startup comfortable, while constraints can sharpen urgency and execution. — Upstarts — Joseph Krause Interview.
- On Competitors: Krause says he pays less attention to competitors than to building the best technology and executing on the company's own work. — Open Source CEO — Joseph Krause Interview.
- On Staying the Course: Krause advises founders not to abandon a deep-tech plan at the first obstacle, while still listening carefully to customers and adapting execution. — Upstarts — Joseph Krause Interview.
- On Interdisciplinary Teams: Krause describes the difficulty of bringing materials scientists, AI engineers, and robotics specialists together around one working laboratory system. — Upstarts — Joseph Krause Interview.
- On Military Discipline: Krause credits his military experience with teaching him discipline and a habit of continuing to learn under demanding conditions. — Open Source CEO — Joseph Krause Interview.
- On Moving from Research to Products: Krause says he left an academic research path and joined AlleyCorp to learn how scientific work becomes a commercial company. — The Startup Project — Joseph Krause Transcript.
- On Decision Velocity: Krause describes acting when a decision is about 51% likely to be right, then learning from the result instead of waiting for certainty. — Open Source CEO — Joseph Krause Interview.
- On First Principles: Krause says the team asks why, challenges assumptions, and evaluates work against the company's mission regardless of seniority. — Open Source CEO — Joseph Krause Interview.
- On Mission-Driven Hiring: Krause says he seeks people who see the company mission as meaningful long-term work, given the ambiguity and difficulty of deep-tech development. — Open Source CEO — Joseph Krause Interview.
- On Preserving Specialization: Krause argues that a multidisciplinary lab needs specialists to contribute their own deep expertise while collaborating across disciplines. — Latent Space — The Self-Driving Lab.