Anthropic's first real brain drain is happening, and it's heading straight for the lab bench.
The Signal
Former Anthropic researchers are raising $175 million at a $1 billion valuation for Mirendil, a startup building AI models for scientific R&D in biology and materials science. Andreessen Horowitz and Kleiner Perkins are in talks to co-lead.
What makes this interesting isn't the money or the valuation. It's the source. Anthropic has been the exception in AI talent hemorrhaging. While OpenAI, Google, and Meta have been leaking researchers to new ventures for years, Anthropic held its people. That's changing now, and the timing matters.
These founders aren't chasing general intelligence or better chatbots. They're building what the industry is calling "neolabs," specialized AI for hard science. Think agent systems that can propose experiments, analyze protein structures, or discover new materials. This is Web4 infrastructure for knowledge work that's currently stuck in PhD labs running 18-month cycles.
The billion-dollar pre-revenue valuation tells you what VCs believe: AI agents won't just automate existing workflows, they'll compress decades of scientific iteration into months. Materials science and drug discovery are multi-trillion dollar problems constrained by human research bandwidth. If agents can genuinely accelerate that cycle, the economics justify the bet.
The Implication
Watch for more "vertical AI labs" in fields where research cycles are long and expensive. The pattern: take foundation model expertise, narrow it to a specific domain, sell directly to industries desperate to move faster. If you're in scientific R&D, your job isn't disappearing, it's getting a turbocharged research partner. Start thinking about what questions you'd ask if experiments were 10x cheaper.
Source: The Information