In the high-stakes arena of artificial intelligence, where the industry standard has long been "bigger is better," a London-based startup is quietly rewriting the rules. Inherent, a boutique AI lab founded by alumni of Google DeepMind, has emerged from stealth with a bold claim: its new AI agent, Faraday, has outperformed the world’s most powerful frontier models at complex scientific tasks, all while operating on a fraction of the computing power.
While tech giants like OpenAI and Anthropic engage in an arms race defined by trillion-parameter models and multi-billion-dollar data centers, Inherent is betting on "research taste"—an intangible quality of intuition and scientific methodology that it believes will define the next generation of AI.
The Core Innovation: Faraday and the Art of Replication
At the heart of Inherent’s recent breakthrough is Faraday, an AI agent designed to independently replicate the findings of published scientific papers. On the surface, this might appear to be a straightforward verification task. However, in the realm of academic research, the ability to replicate findings without prior access to the results is the hallmark of a skilled researcher.
"Many PhD students actually start by doing this," says Edward Hughes, co-founder and chief scientist at Inherent. "It is a standard training exercise for human scientists to prove they understand the methodology and the rigor required to reach a specific scientific conclusion."
What sets Faraday apart is not just its success in these replications, but the efficiency with which it achieves them. Measured against industry titans like Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, Faraday runs on Qwen 3.6—a model with just 27 billion parameters. To put this into perspective, frontier-scale systems often boast parameters in the hundreds of billions or even trillions. By achieving superior results with a "tiny" model, Inherent is challenging the prevailing belief that intelligence must be tethered to sheer scale.
Chronology: From Stealth to Scientific Breakthrough
The trajectory of Inherent has been swift, marked by a quiet incubation period that shielded the team from the hype-cycle pressures of the broader AI market.
- The Inception: Founded by a quartet of DeepMind alumni—including Edward Hughes, Louis Kirsch, Kaloyan Aleksiev, and Tantum Collins—Inherent was built with the explicit "north star" of creating an AI capable of independent scientific discovery.
- The Funding Milestone: In May 2026, the company emerged from stealth, announcing a $50 million seed round. This infusion of capital provided the runway necessary to move from theoretical research to the development of actionable agents.
- The "Faraday" Launch: Only weeks after its public debut, the company unveiled Faraday. The agent was stress-tested against frontier models in blind trials, where it was tasked with reconstructing scientific experiments from scratch.
- Future Roadmap: With a current headcount of roughly a dozen employees, Inherent plans to scale to 25 staff members by the end of 2026. This expansion comes at a pivotal time for the London AI ecosystem, which is currently experiencing a period of talent mobility following shifts in leadership and strategy at major local labs.
The Methodology: Reinforcement Learning as "Research Taste"
The most compelling aspect of Inherent’s approach is its rejection of rote memorization in favor of "reinforcement learning." Rather than feeding the model a massive library of existing papers and asking it to summarize them, the team rewards the agent for the process of scientific inquiry.
"We are imbuing our agents with taste," Hughes explains. "We want the agent to develop an instinct for what experiments are worth running and how to design them well."
This philosophy of "research taste" involves teaching the model to evaluate the utility of an experiment before executing it. To support this, Inherent has made a strategic decision not to build its own coding tools. Instead, Faraday leverages OpenAI’s GPT-5.5 Codex, mirroring the way a professional scientist might use a standardized software suite or programming language to conduct their work. By outsourcing the commodity of coding, Inherent keeps its focus on the higher-level cognitive task: scientific reasoning.
The goal is to move away from "chatbots" that simply parrot information and toward a teammate that proactively asks: "I got curious about this phenomenon, so I went off and performed these experiments. Here are the results—what do you think?"
Industry Implications: A Shift Toward Lean AI
Inherent’s breakthrough carries significant implications for the future of AI investment and development.
1. The Death of the "Scale-Only" Paradigm
For years, the industry has assumed that the path to Artificial General Intelligence (AGI) is paved by larger GPUs and more extensive datasets. Inherent’s success with a 27-billion-parameter model suggests that architectural efficiency and training methodology may be more critical than brute force. If an agent can demonstrate high-level reasoning with a fraction of the compute, the economic barrier to entry for high-performance AI drops significantly.
2. The London Hub Effect
The company’s decision to plant its flag in London’s King’s Cross is a deliberate play on regional density. The area has become a global epicenter for AI, largely due to the early presence of Google DeepMind. By positioning themselves at the heart of this ecosystem, Inherent is positioning itself as the primary beneficiary of the "DeepMind diaspora"—researchers who are looking for smaller, more agile environments to conduct high-risk, high-reward science.
3. Policy and the "Garden Leave" Barrier
The growth of Inherent is also shedding light on the structural challenges facing British startups. Edward Hughes has been a vocal critic of "garden leave"—a U.K.-specific labor practice that mandates long periods of professional inactivity for employees leaving established firms to join competitors.
Hughes argues that these restrictions hinder innovation by preventing the fluid exchange of ideas and talent. His personal experience navigating these barriers has made him an advocate for a more dynamic labor market, one that mirrors the "at-will" employment environment that has historically fueled the rapid iteration cycles of Silicon Valley.
Looking Ahead: The Quest for the AI Scientist
The ultimate ambition of Inherent Labs is not to win benchmarks, but to automate the scientific method. If an agent can independently conceive, execute, and verify scientific hypotheses, the implications for drug discovery, material science, and climate modeling are profound.
However, the team remains grounded. They are not attempting to replace human scientists; they are attempting to augment them. By handling the repetitive, labor-intensive aspects of experimental design and replication, Faraday acts as a force multiplier for human researchers.
"We’re always guided by that north star," Hughes says. "We want to build an AI scientist that can stand alongside human researchers and challenge our assumptions, not just confirm them."
As Inherent continues to scale, the industry will be watching closely. If they can successfully translate "research taste" into a repeatable, scalable agentic architecture, they may well prove that the next leap in artificial intelligence won’t come from a bigger server farm—but from a smaller, smarter, and more curious piece of software.
Quick Summary of Key Facts
- Company: Inherent (London-based AI lab).
- Founding Team: Comprised of former Google DeepMind alumni.
- Key Product: Faraday, an AI agent focused on independent scientific replication.
- Technical Edge: Achieved top-tier results using a 27-billion-parameter model (Qwen 3.6), challenging the "bigger is better" status quo.
- Strategic Focus: Using reinforcement learning to instill "research taste" and intuition in agents.
- Funding: Raised $50 million in a seed round as of May 2026.
- Operational Stance: Focused on high-density, in-person collaboration in London’s King’s Cross hub.
