Inherent unveils Faraday, a compact AI research agent

A London artificial intelligence lab founded by former Google DeepMind researchers has introduced an agent called Faraday. The startup, known as Inherent, recently emerged from stealth with a $50 million seed funding round and a small team based in King's Cross.

Replicating scientific findings

Faraday's primary task is to independently reproduce the findings of published scientific papers without knowing the final answers in advance. According to cofounder and chief scientist Edward Hughes, this replicates a standard training exercise used by human PhD students.

While frontier artificial intelligence systems often rely on massive scale, Inherent's agent operates using a smaller model called Qwen 3.6, which contains 27 billion parameters. Parameters act as a proxy for a model's size and training costs. Despite its smaller size, the startup reports that Faraday outperformed larger systems developed by Anthropic and OpenAI on this replication task.

Building research taste

Beyond simple accuracy, Inherent wants its agents to demonstrate research taste, which is an instinct for selecting valuable experiments and designing them effectively.

To encourage this behavior, the team uses reinforcement learning — a training method that rewards an artificial intelligence system for positive outcomes rather than following fixed rules. For coding tasks, the agent utilizes OpenAI's GPT-5.5 Codex rather than a proprietary tool, mirroring how human scientists rely on existing software.

Inherent plans to expand its team from a dozen employees to between 20 and 25 workers by the end of the year, focusing on the development of broader world models.