Caitlin Kalinowski, former head of robotics and consumer hardware at OpenAI, argues that keyboard-based artificial intelligence is reaching saturation and the technology industry must pivot to the physical world. In an episode of Lanny’s Podcast, Kalinowski discusses how this transition from software to robotics demands new manufacturing capabilities, supply chain resilience, and safety protocols—transforming corporate strategy into a matter of national security.
Caitlin Kalinowski, former head of robotics at OpenAI / Photo credit: Caitlin Kalinowski
Software Progress Slows as AI Focuses on Physical Objects
Kalinowski contends that as AI labs build better models, the value of text generation diminishes. “What you can do behind a keyboard with AI is going to saturate,” she argues. “The next frontier is the physical world: robotics, manufacturing, and industrialization.”
To compete in this new era, companies must build physical sensors, operate factories, and deploy robots in real-world environments rather than rely on digital-only applications.
Virtual Reality Enabled Robot Spatial Awareness
According to Kalinowski, VR technology laid the groundwork for robotics by solving spatial orientation challenges. “VR helped us understand how to orient things in space and connect a simulated world to the real world,” she explains. “We figured out SLAM (simultaneous localization and mapping), depth sensors, and how humans perceive visual data. Now robotics is using all of it.”
She notes that this tracking technology is universal and now serves as the foundation for autonomous vehicles, drones, and manufacturing systems.
Smart Glasses Face Production and Social Barriers
Transitioning from digital code to physical wearables introduces immediate challenges. Kalinowski identifies two key barriers:
- Usability in public: Controlling devices quietly while moving requires new interaction methods that do not exist at scale.
- Social acceptance: Covering a user’s face disrupts normal human conversation, making mainstream adoption difficult.
Regarding Meta’s Orion smart glasses, Kalinowski explains: “The Orion smart glasses are a bit ahead of their time because they’re using waveguides and microLEDs that are not quite ready for mass production. The yields just aren’t there. The cost is still high.”
She adds that VR encountered the same social barrier; once a device covers the face, consumer adoption becomes an uphill battle.
Global Supply Chains Expose Fragmentation
While consumer hardware faces social obstacles, industrial robotics reveals fragile supply networks. Scaling production is the primary challenge—even with reliable designs, companies immediately encounter supply bottlenecks.
Kalinowski outlines the layered supply chain: “Start with raw materials and magnets… then process them, integrate them into actuators, and integrate those actuators into robots. Each layer of the supply chain is outsourced to China, Japan, and Korea. To have a safe supply chain, we need independence in these layers.”
Consumer electronics and military weapons rely on the same global supply chains, leaving the U.S. vulnerable to disruptions.
Rising Memory Costs Create Financial Strain
The hardware arms race is driving up component costs, forcing difficult operational decisions across the industry:
- Robots, phones, and data centers all require computer memory.
- AI companies purchase the majority of available memory, driving prices upward.
- Large businesses can absorb higher costs, while smaller teams building consumer gadgets face lower profits or product delays.
Kalinowski advises: “I have been advising startups and companies to pre-buy memory to ride out price spikes. If a key component, like memory or silicon, is constrained, there’s not much you can do. You either pay or you have already pre-bought enough.”
When a single component like RAM becomes unavailable, it forces redesign of the entire product’s internal architecture. To survive supply chain shocks, Kalinowski argues companies must bring manufacturing in-house, allowing rapid design pivots when components vanish—similar to how Tesla navigated the global silicon shortage.
Robot Safety Requires Strict Protocols
Securing supply chains is secondary to public safety. Engineers must prioritize making robots safe and predictable rather than producing impressive demonstrations.
True human-robot collaboration remains distant because most industrial machines still require strict exclusion zones. Kalinowski notes: “You can get a Chinese robot, but the booklet says, ‘No human can be within three feet of this robot.’ There aren’t very many robots strong enough to do meaningful work that don’t have that warning right now.”
Defense Partnerships Demand Clear Ethical Guardrails
Deploying autonomous robots requires public trust. Merging AI with defense contracts demands explicit ethical boundaries; without them, a company’s reputation and engineering teams fracture.
Reflecting on OpenAI’s Department of Defense partnership, Kalinowski criticizes the rushed decision-making and lack of defined guardrails. She ultimately walked away to avoid future unpredictability, hoping her exit made it “easier for others to talk about their boundaries.”
Kalinowski emphasizes that preventing internal conflict requires absolute clarity from leadership. The cultural gap between AI researchers and hardware engineers creates serious miscommunication risks. High-stakes military contracts demand shared mission alignment to ensure unified direction.
Counter-Argument: Software May Not Be Saturating
While Kalinowski’s thesis on the physical world is compelling, software is not obviously reaching a ceiling. Gartner forecasts worldwide AI spending to hit US$2.52 trillion in 2026, while supply-chain software with agentic AI is expected to grow from less than US$2 billion in 2025 to US$53 billion by 2030. This suggests the next wave may not be a clean pivot from software to hardware, but a hybrid cycle where software agents increasingly run factories, logistics systems, and industrial workflows behind physical AI.
Supply Chain Constraints Beyond In-House Manufacturing
The supply-chain argument faces a harder constraint than simply bringing manufacturing in-house. According to Reuters reporting from May 2026, China still refines over 90% of the world’s rare earths, while RSIS noted that China’s 2025 controls targeted selected rare-earth magnets and separation technologies. Vertical integration may help companies respond faster, but it cannot fully solve upstream dependence on materials, processing know-how, and export licensing that sit outside any single company’s factory walls.
Robotics Safety Standards Are Evolving
Kalinowski’s concern about robot safety and defense use is becoming less of a blank space than the argument implies. ISO updated its 10218-1:2025 robotics safety requirements, while the U.S. ANSI/A3 R15.06-2025 revision formally recognized cybersecurity vulnerabilities as physical safety hazards, according to the Association for Advancing Automation.
OpenAI also stated in 2026 that its Department of Defense agreement includes explicit red lines and layered guardrails. This does not remove the ethical tension, but it suggests the industry is starting to formalize rules for physical AI rather than moving into robotics and defense with no safety architecture at all.
