A former Anthropic employee has started a company and secured $200 million in funding.
Recently, two prominent figures who left Anthropic—Behnam Neyshabur and Harsh Mehta—officially announced the founding of a startup named Mirendil. Their vision aligns closely with the previously announced venture by Tian Yuan Dong: creating AI capable of self-evolution to accelerate human scientific progress. Both projects follow the technical approach of “recursive self-improvement.”
The company’s founding team consists of 20 researchers and engineers from leading institutions such as Anthropic, xAI, Google DeepMind, and OpenAI, united by their passion for science and drive to build technologies that advance scientific progress. The company name, “Mirendil,” is derived from The Lord of the Rings and means “Friend of Treasures” in Elvish.
The $200 million in seed funding was raised from venture capital firms Andreessen Horowitz, Kleiner Perkins, and NVIDIA. Post-funding, the company reached a $1 billion valuation, making it one of the highest-valued AI startups in recent years at the seed stage.
Origin: An unexpected email,
A seven-year "partnership"
The story of the two founders begins in 2019.
At that time, Mehta was just a regular researcher at Google, while Neyshabur had recently joined Google and had already gained some recognition in academia for his deep research into the fundamental question of “why AI models work”—in Mehta’s own words, “he was already something of a rising star in the field.” Summoning his courage, Mehta sent him an unsolicited email, and thus began their connection.
They were passionate about using AI to accelerate scientific research from an early stage, but at the time, limitations in model capabilities meant this idea could only remain theoretical. It wasn’t until late 2024, when both joined Anthropic, and shortly after the release of Claude Opus 4.5 in December 2025, that they decided to leave and start their own company.
The release of Claude Opus 4.5 significantly enhanced the ability of AI agents to handle complex tasks, and perhaps this breakthrough led them to believe the time was ripe.
Mission: Not "AI for Science",
Instead, it's "AI for AI for Science"
Mirendil’s positioning may sound a bit convoluted, but Neyshabur put it simply: “What we do is enable scientists to build their own AI with the help of AI, rather than merely using AI to assist science.”
In other words, their goal is to build a tool platform that enables research teams across various vertical fields—such as medicine and materials science—to independently train and iterate on their own specialized AI models without relying on generic models provided by large corporations. One example they cite is helping researchers develop models to predict the risk of Alzheimer’s disease.
Behind this lies a more controversial technical path—recursive self-improvement, which involves enabling AI to continuously optimize its own code and capabilities. Neyshabur explicitly stated that this is the shortest path to “AI-accelerated science,” and he also believes it can be safely advanced under human supervision. “I don’t accept the argument that it’s impossible—it’s just a very difficult problem.”
Competition: The moat of major players
It is precisely Mirendil's opportunity.
Mirendil can receive this money, and the reasoning behind it is quite clear.
Currently, major AI companies, including Anthropic, are increasingly using AI to accelerate their own research. Anthropic has disclosed that, as of May this year, over 80% of its internal code has been written by Claude. However, at the same time, these companies explicitly restrict external developers from using their models to train competing products in their user agreements.
In the view of a16z investor Matt Bornstein, this “for internal use only, not for lending” strategy is simply a normal response from large corporations as “rational economic actors.” But precisely because of this, a structural gap has emerged in the market that requires an independent company to step in and fill it.
In addition to the two co-founders, Mirendil’s core team includes Shayan Salehian, an early member of Elon Musk’s xAI, and Tara Rezaei, a graduate of MIT. The company currently has approximately 20 technical staff and is headquartered in downtown San Francisco.
Over the coming months, Mirendil plans to release its first models and products to gather early feedback from users. Neyshabur’s vision is: “We hope that in the future, thousands of laboratories will exist worldwide, each tackling the most important challenges of our time. We want to be the force that empowers them.”
Reference link:
https://x.com/bneyshabur/status/2069860934148079800
https://www.wsj.com/tech/ai/anthropic-veterans-startup-seeks-to-help-scientists-develop-their-own-ai-09e2f3e5?mod=author_content_page_1_pos_1
This article is from the WeChat public account "Machine Heart," authored by the Machine Heart editorial team.