Ex-Anthropic researchers' startup Mirendil raises $1B to let AI agents do an AI lab's work

Mirendil, a startup founded by former Anthropic researchers, has raised $1 billion. According to MLQ.ai's report, the company's goal is to have AI agents take on work that AI labs currently do.

The weight of that claim comes from moving the target of automation from tasks to the research process itself. Getting agents to write code and run evaluations is no longer novel; the hard part is having them propose hypotheses, design experiments, judge whether results are trustworthy, and replan after failure.

The report leans on the team's pedigree as part of its narrative: the founders came from Anthropic. That background carries a specific market meaning right now, since investors read it as evidence that the team understands the trade-offs between large-scale training, evaluation and safety constraints rather than merely knowing how to wire up an agent framework.

A $1 billion raise signals that capital is betting on more than another agent application; it is betting on the more aggressive premise that research itself can be automated. Bets at that layer pay back over years and depend heavily on compute supply and talent density.

The gaps in the report are just as notable. There is no investor list, no valuation, no indication of what stage the product has reached, and no evidence of a working demonstration. The framing that agents will do an AI lab's work is the company's claim, not a validated capability.

What actually needs watching is verifiability: whether Mirendil can produce reproducible research output, how outsiders should judge the quality and novelty of AI-produced research, and whether the model beats hiring a human research team once compute costs are counted.