DeepMind Alumni Startup Inherent Releases AI Agent to Replicate Scientific Studies

Inherent, a London-based artificial intelligence lab established by Google DeepMind veterans, announced that its new AI agent, Faraday, successfully outperformed larger frontier models from OpenAI and Anthropic in independently replicating scientific research papers.

The startup, which recently exited stealth after securing $50 million in seed financing, accomplished the task using a comparatively compact 27-billion-parameter model named Qwen 3.6. In benchmarking comparisons against Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, Faraday achieved superior results when validating published scientific studies without pre-programmed answers, as first reported by TechCrunch AI.

Co-founder and chief scientist Edward Hughes noted that reproducing academic papers mirrors the early training of human researchers, who frequently replicate existing studies during their doctoral programs. While Inherent's broader objective is building software that generates original scientific breakthroughs, Hughes explained that paper replication serves as an essential preliminary benchmark.

To help the agent develop an instinct for experimental design—referred to by the company as "research taste"—Inherent employed reinforcement learning, rewarding the model for effective outcomes rather than hardcoding procedural instructions. Instead of constructing proprietary software utilities, Faraday utilizes OpenAI's GPT-5.5 Codex to write code, mirroring how human researchers rely on existing digital tools.

The company currently operates with a team of 12 employees working in person from an office in London's King’s Cross district. Inherent plans to double its staff to roughly 20 to 25 people by the end of the year, tapping into the local pool of machine learning talent.

Hughes also criticized the United Kingdom's widespread use of "garden leave," a legal provision that prevents departing corporate researchers from immediately joining competing ventures or starting new businesses. He argued that the practice creates hiring friction for European startups compared to U.S. competitors, where departing engineers face fewer post-employment restrictions.