Mirendil, a San Francisco-based startup founded in early 2026 by former Anthropic researchers Harsh Mehta and Behnam Neyshabur, officially launched its autonomous AI research platform on June 25, securing a 200 million dollar seed round that values the company at approximately 1 billion dollars. The investment, led by a16z and Kleiner Perkins with participation from NVIDIA, underscores strong institutional confidence in a venture born from internal experimentation at one of the industry’s leading AI labs. The founding team’s technical pedigree directly informed Mirendil’s architecture. While at Anthropic, Mehta developed an internal system that automated early stages of AI development, eventually scaling it into a core research capability. Neyshabur, a co-inventor of the SAM optimizer and former leader of Google’s Minerva and Gemini math initiatives, previously co-led Anthropic’s Discovery team, which focused on building autonomous AI scientists. The twenty-person engineering and research team also includes Shayan Salehian from xAI and Tara Rezaei from OpenAI, collectively bringing expertise from Anthropic, Google DeepMind, xAI, and OpenAI. Mirendil’s core product is a self-optimizing system designed to automate the full cycle of AI research. The platform independently formulates experimental hypotheses, generates code, allocates computational resources, executes training runs, evaluates checkpoints, and iterates on subsequent trials. Co-founder Neyshabur describes the architecture as self-accelerating AI, engineered to rapidly deepen specialization within targeted research domains. By abstracting away the infrastructure complexity typically required to train frontier models, the system enables domain scientists to run advanced experiments without maintaining large-scale machine learning operations. Venture capitalists have framed the platform as a catalyst for vibe research, a term denoting the democratization of high-end AI development. a16z noted that the current landscape forces non-AI specialists to either secure billions in compute capital or rely on external labs, creating a structural bottleneck. Mirendil addresses this by providing a managed research environment where users can continuously refine models and workflows. The firm views the company as filling a critical infrastructure gap, positioning it alongside the open-source model layer, which a16z acknowledged is currently dominated by Chinese developers but remains a secondary strategic concern. The startup’s rapid capital raise and lean twenty-person headcount reflect a broader industry shift toward operationalizing AI research. By productizing autonomous experimentation, Mirendil aims to compress development cycles, reduce infrastructure costs, and extend frontier AI capabilities across disciplines ranging from drug discovery to materials science. As computational barriers continue to dictate innovation velocity, the platform’s ability to scale specialized research loops may fundamentally alter how scientific AI is developed and deployed.