Building Self-Accelerating AI with Mirendil (41 min)
ai-driven-innovation-economy
ai-in-workforce-disruption
ai-monetization-strategies
- Release date: 2026-06-24
- Listen on Spotify: Open episode
- Episode description:
Matt Bornstein speaks with Mirendil cofounders Behnam Neyshabur and Harsh Mehta about their vision for building self-accelerating AI. After leading research efforts at Google and Anthropic, the founders started Mirendil around a simple question: what happens when AI systems can meaningfully contribute to their own development? Rather than focusing solely on AI as a tool for productivity, they argue that the most important application may be accelerating scientific and technological progress itself. The conversation explores AI research, scaling laws, automated engineering, scientific discovery, and the challenges of building systems that can improve over time. They discuss the future of AI-assisted research, why they believe scientific progress remains bottlenecked by intelligence, and how more capable AI systems could help unlock advances across medicine, engineering, and the natural sciences. Resources: Follow Behnam Neyshabur on X: https://x.com/bneyshabur?lang=en Follow Harsh Mehta on X: https://x.com/HarshMeh1a Follow Matt Bornstein on X: https://x.com/BornsteinMatt Check out everything a16z is doing with artificial intelligence here, including articles, projects, and more podcasts. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.
Summary
- 🚀 Self-Accelerating AI for Science: Mirendil targets AI that conducts its own research and engineering to remove intelligence bottlenecks in solving grand scientific challenges.
- 🏢 Startup Advantage for Disruptive Tech: Founders left Anthropic to build a company optimized end-to-end for sharing self-improving AI rather than hoarding capabilities.
- 📈 Systems Over Single Models: Progress comes from ecosystems of specialized agents plus humans that recursively improve themselves with minimal oversight.
- 🌍 Democratizing AI Capabilities: Goal is to let every lab or company own optimized AI infra and data instead of depending on a few frontier providers.
- 🔬 Positive AI Future Focus: Directing powerful AI at longstanding problems like disease cures creates jobs and shared prosperity rather than just automation.
Insights
- How can self-accelerating AI systems shift from narrow coding assistance to broadly accelerating scientific discovery across domains?
- Time: 4:37 – 5:48
- Answer: The founders argue that AI must develop deep domain expertise and recursive improvement loops, enabling it to tackle superhuman problems like curing diseases rather than just automating routine tasks. This matters because current frontier labs may be misaligned with open scientific acceleration due to their business models.
- Why might startups be better positioned than large labs to responsibly deploy disruptive self-improving AI technologies?
- Time: 15:04 – 16:15
- Answer: Existing companies face misaligned incentives around sharing powerful AI research tools, while new entities can redesign culture, business models, and safety approaches from the ground up to prioritize broad access and scientific progress.
- What does it take to scale AI systems (not just models) so that adding agents and humans yields superlinear productivity gains?