S12 Bonus: The Human In the Loop Bottleneck: Moving Beyond Static Models to Unleash Autonomous, Recursive Self-Improving AI with Kunal Bhatia, Co-Founder & CEO of Hexo Labs

Episode Summary

Executive Summary: Kunal Bhatia, co-founder and CEO of Hexo Labs, explains the company’s pivot from AI consulting to research on self-improving AI—systems that can build and improve AI themselves. The episode emphasizes Hexo’s research-first MVP (a published paper), partnerships with top labs, compute as the main bottleneck, and Kunal’s belief that AI will drive an epoch-level societal shift that will upend industries and reward asymmetric thinking.

Main Topics: Hexo Labs’ origin story and pivot to self-improving AI (Priority: 5/5): Hexo began as a consulting services company building custom AI for clients, but demand outpaced delivery. That bottleneck led Kunal and his co-founder to ask whether AI could build AI, turning the company toward research on self-improving systems. What self-improving AI means (Priority: 5/5): Kunal frames self-improving AI as the endgame of the AI trend: systems that can improve themselves and eventually solve a wide range of hard problems, effectively becoming technology that builds all other technology. Research-first MVP and real-world validation (Priority: 4/5): Instead of a conventional product, Hexo’s MVP is its first research paper, SIA (Self-Improving AI). The company is testing its systems with researchers at major institutions on frontier problems in energy, quantum tech, particle physics, biomedicine, and AI. Compute and energy as core bottlenecks (Priority: 5/5): Kunal argues that the biggest constraint on self-improving AI is compute access. Hexo is working with hyperscaler partners and labs focused on energy and compute because these are the same bottlenecks AI systems face. Team building, hiring, and founder fit (Priority: 4/5): Hexo hires for technical depth but also mission alignment, grit, and a willingness to take bets. Kunal says the team uses short trial projects to assess whether candidates are a mutual fit. Industry disruption and the coming societal epoch shift (Priority: 5/5): Kunal believes AI will evolve into a few dominant superintelligence labs that can wipe out entire industries via agents. He argues the change is bigger than a platform shift—it is an epoch shift akin to civilization-level transitions. Advice for founders in the AI era (Priority: 5/5): He warns that traditional startup playbooks are becoming obsolete. Founders should think asymmetrically, assume AI-native competitors can replace point solutions, and build for a world where exponential progress accelerates dramatically.

Key Arguments: Hexo’s move from consulting to research came from a practical bottleneck: customer demand exceeded delivery capacity, prompting the question of whether AI could build AI. Self-improving AI is positioned as the logical destination of AI development—software that iteratively improves itself and can attack hard problems across domains. A research paper can function as an MVP when the company is fundamentally a research lab rather than a standard enterprise software business. The most important problems for self-improving AI are the ones that are also AI’s own bottlenecks, especially compute and energy. Partnerships with top labs and universities serve both validation and problem selection, while potentially creating valuable new IP. Scaling a self-improving AI lab is constrained less by front-end product-market scaling and more by backend compute access. Hiring for mission, grit, and long-term conviction matters more than pedigree alone in an early-stage research startup. Early founder mistakes included over-listening to outside experts and under-trusting their own instincts. Traditional startup frameworks and linear planning are insufficient in an AI-driven world because change is exponential and can invalidate point solutions quickly. Founders should think in terms of asymmetric bets because a major AI lab or agent can disrupt an entire company or industry. Kunal believes the AI transition will reshape society on a scale comparable to past civilizational shifts from hunter-gatherer to agricultural to industrial societies.

Data Points: Founding timeline: Over 4 years - Kunal says he and his co-founder have been together for more than four years. Team community: 800+ researchers - Hexo runs the Frontier Research Club in the Bay Area and has a pipeline of researchers. Trial hiring period: 2-3 weeks - The company uses short projects to test mutual fit with candidates. Experience in AI: 12 years - Kunal says he has worked in AI for 12 years. Company stage: 3rd company in AI - Kunal says Hexo is his third AI company. Child age: 3-year-old daughter - Part of Kunal’s personal background in the introduction. Research partner institutions: Stanford, Lawrence Livermore, UC Santa Barbara, University of Oxford - Institutions Hexo collaborates with on frontier problems. Problem domains: Quantum technologies, energy, particle physics, AI research, biomedical research - Areas where Hexo’s agents are being applied. Compute spend for small experiments: Thousands of dollars of GPU funds - Kunal describes the compute cost of running even small experiments. Societal transition timeline: Hunter-gatherer to agriculture: millennia; agriculture to industrial: centuries; industrial to superintelligence: decades - Kunal’s analogy for the scale of the coming AI shift. Progress acceleration claim: 100 years of progress in 5-10 years - Kunal argues AI will compress the pace of advancement dramatically. Industry disruption example: Claude Design and Figma shares dropped - He cites Anthropic’s agent as evidence of AI impacting an entire industry.

Pivotal Quotes: "We’re trying to build the end game of the whole AI trend itself, right? Where is AI going to? It’s basically going to a point where AI starts building itself." — Kunal Bhatia: Describing Hexo Labs’ mission and vision for self-improving AI. "The next most important thing for a self-improving system is to solve problems that are its own bottleneck." — Kunal Bhatia: Explaining why Hexo focuses on energy and compute as research targets. "This isn’t just a platform or technology shift... We’re talking about a societal epoch shift." — Kunal Bhatia: His view of AI’s broader impact on society and industry.

Implications: Listeners should expect AI to move from tool to autonomous builder, making compute, energy, and research partnerships strategic advantages. The episode suggests founders must abandon linear playbooks, think exponentially, and prepare for industry consolidation around a few superintelligence labs.