DeepMind alumni-led Inherent bets on smaller, ‘tasteful’ AI agents

While industry giants chase scale, Inherent is prioritizing a concept cofounder Edward Hughes calls “research taste”—an algorithmic instinct for designing experiments and determining which avenues of inquiry are worth pursuing. By employing reinforcement learning, the team aims to build systems that function as proactive scientific partners rather than mere answer-generators. To achieve this, Faraday utilizes existing tools like GPT-5.5 Codex, mirroring how human researchers leverage established software rather than building every component from scratch.

Operating out of a King’s Cross office, the dozen-strong team is positioning itself as a hub for talent leaving the broader DeepMind ecosystem. Hughes, who navigated the U.K.’s restrictive “garden leave” policies to launch the $50 million-backed venture, is now scaling toward a headcount of 25. As the startup expands its focus into world models, it aims to prove that smaller, highly specialized architectures can reliably push the boundaries of scientific knowledge.