Inherent: DeepMind Alumni's AI 'Teammate' Outperforms Giants in Research Replication

Inherent, an AI laboratory based in London and founded by former Google DeepMind employees, has recently introduced its AI agent, Faraday. This new system has shown remarkable capabilities, surpassing the performance of more extensively developed AI models from industry leaders like Anthropic and OpenAI in the complex task of replicating scientific research. This achievement is particularly noteworthy given that Faraday operates with a considerably smaller model size and fewer computational parameters, indicating a highly efficient and innovative approach to artificial intelligence development.

Emerging from stealth mode with a significant seed funding round of $50 million, Inherent's Faraday project represents a pivotal step in their ambitious goal: to create AI that not only verifies existing scientific findings but also actively contributes to the discovery of new knowledge. The ability of Faraday to independently reproduce the results of published scientific papers, without prior knowledge of the outcomes, is a testament to its advanced design. Edward Hughes, a co-founder and chief scientist at Inherent, emphasized that while outperforming other frontier AI agents was a welcome outcome, the true significance lies in the unique methodology employed in Faraday's construction. This process mirrors the foundational training of human scientists, where replicating experiments is a crucial initial step in understanding and validating scientific principles.

A key aspect that distinguishes Faraday is its efficiency. Despite being compared against models such as Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5, which are substantially larger in scale and possess billions more parameters, Faraday leverages a comparatively compact model, Qwen 3.6, with just 27 billion parameters. This efficiency translates into lower training costs and computational demands. Furthermore, Inherent's criteria for Faraday's success extended beyond mere accuracy; they aimed for the AI to exhibit "research taste"—an intuitive understanding of which experiments hold the most promise and how to optimally design them. This nuanced capability suggests a move towards AI that can make more sophisticated, human-like judgments in research.

The development of such 'taste' in AI is primarily achieved through reinforcement learning, a training paradigm that rewards desired behaviors and outcomes rather than explicitly programming rules. This reward-based approach, rather than solely focusing on the mechanics of scientific conduct, is intended to enable Faraday to generalize its learning across various scientific domains, thereby bringing Inherent closer to its long-term vision of creating versatile AI scientist agents. This strategic focus also dictates what Inherent chooses not to develop. Instead of building proprietary coding tools, Faraday integrates existing advanced software like OpenAI's GPT-5.5 Codex, mirroring how human researchers utilize available tools rather than developing everything from scratch, fostering a collaborative and efficient research ecosystem.

The company culture at Inherent also mirrors this collaborative philosophy. All twelve employees work in person at their King’s Cross office in London, a district that has become a prominent AI hub thanks in part to Google DeepMind’s influence. This co-location fosters a dynamic and interactive work environment, which Hughes believes is crucial for innovation. Hughes also openly advocates for the abolition of "garden leave" in the UK, a practice that restricts departing employees from joining rival companies for a period. He argues that this hampers talent mobility and gives US startups an unfair advantage in acquiring top researchers. Despite such challenges, Hughes, along with three other co-founders, successfully navigated these constraints to establish Inherent, which plans to expand its team significantly by the end of the year. With the current shifts at DeepMind, Inherent's growth trajectory and focus on pioneering AI research make it an attractive destination for AI professionals seeking new opportunities.

Inherent's innovative Faraday AI agent marks a significant advancement in the field of artificial intelligence, particularly in its ability to efficiently replicate complex scientific research. By focusing on a lean yet effective model and integrating principles of 'research taste' through reinforcement learning, the company, founded by former DeepMind talents, is setting a new standard for AI's role in scientific discovery. This approach not only challenges the conventional wisdom that larger models are inherently superior but also paves the way for AI to act as a genuine teammate in accelerating scientific progress.