DeepMind Spinoff’s AI Agent Just Replicated Research Better Than OpenAI and Anthropic

DeepMind alumni have long been at the vanguard of AI, but a new British lab is turning heads with a claim that could redefine the field. Inherent, founded by former DeepMind researchers, has unveiled Faraday, an AI agent it says outperforms Anthropic’s and OpenAI’s offerings in replicating scientific research. The achievement isn’t just a parlor trick—it’s a potential game-changer for how AI integrates into the scientific process.

Faraday’s claim to fame is its ability to take a scientific paper, replicate its experiments, and produce comparable results with minimal human input. According to Inherent, this puts Faraday ahead of Anthropic’s latest models and even OpenAI’s reasoning engines in tasks involving reproducibility. The company argues that this capability could accelerate research cycles by automating the tedious work of validation and replication, freeing scientists to focus on breakthroughs rather than verification.

But before we declare Faraday the new sheriff in town, let’s pump the brakes. Replicating research is a critical function, but it’s not the same as generating novel insights. Faraday’s performance is impressive in a vacuum, but the real test is whether it can uncover new findings or streamline collaborations in ways that move science forward. So far, Inherent hasn’t released peer-reviewed data or independent benchmarks to substantiate its claims—only internal tests and cherry-picked examples. That’s a red flag in an industry where vendor benchmarks are notorious for overselling capabilities.

The broader implication here is about the role of AI agents in science. If Faraday (or its competitors) can reliably replicate and validate research, it could democratize access to high-quality science, reduce fraud, and even help non-experts verify claims. But the flip side is the risk of over-reliance on AI-generated replication, which might stifle creative problem-solving or introduce systematic biases if the underlying models are flawed. After all, an AI can’t replicate what it doesn’t understand—and understanding is where much of the magic (and mess) of science happens.

Inherent’s team, with its DeepMind pedigree, brings credibility to the table. But credibility doesn’t equal capability. For Faraday to truly stand out, it needs to prove it can do more than mimic past research—it needs to push the boundaries of what’s possible. Until then, it’s a fascinating experiment, not a revolution. And in the world of AI agents, that’s still a pretty big deal.

The real race isn’t just about replicating research; it’s about who can use AI to expand it. Faraday might be leading the pack today, but the finish line is a moving target.

Photo: Franck V. / Unsplash (https://unsplash.com/@possessedphotography)