Ex-Anthropic Researchers Raise $200M Seed to Build AI That Automates AI Research
Summary: A San Francisco startup built on the idea that AI can do what AI researchers do — design experiments, iterate on engineering, and improve models over time — just raised $200 million before shipping a product.
Key Facts
- Founders: Behnam Neyshabur and Harsh Mehta left Anthropic in December 2025, months after joining; Mirendil announced its $200M seed on June 25, 2026.
- Investors: Andreessen Horowitz and Kleiner Perkins co-led; Nvidia participated. Post-money valuation: $1 billion.
- Team: ~25 researchers and engineers recruited from Anthropic, xAI, Google DeepMind, and OpenAI — including Shayan Salehian (early xAI) and Tara Rezaei (MIT).
- Mission: Build AI systems that automate the frontier R&D loop — experiment design, hypothesis iteration, model improvement — making it accessible to universities and labs beyond the top-5 AI giants.
Why It Matters
Frontier labs like Anthropic and Google DeepMind are already pursuing internal AI-for-AI-research programs. Mirendil's bet is that this capability should be an external platform, not a proprietary moat — democratizing research automation for institutions that can't afford armies of PhD researchers. The $200M seed signals investors believe the "meta-research" market is real and imminent, even before a product ships. Watch for this to accelerate the already-heated poaching war between labs.
Read More
- Ex-Anthropic researchers raise $200M to build AI that creates better AI — Tech Funding News
- Mirendil secures $200M seed round led by a16z and Nvidia — Crypto Briefing