The race to build AI that doesn't just chat but actually *does* science just got a new contender. A British lab founded by DeepMind alumni says its AI agent, Faraday, has outperformed systems from both Anthropic and OpenAI at one very specific, very demanding task: replicating scientific research. If the claim holds up, it could change how labs approach everything from drug discovery to materials science.
What is Faraday, the AI agent from DeepMind alumni?
Faraday is not another chatbot. It is an AI "teammate" designed to work alongside researchers. Its core skill is taking a scientific paper and reproducing the experiments or findings — a process known as replication. This is a notoriously difficult, time-consuming, and often frustrating part of science, which makes it a prime target for automation.
Why outperforming Anthropic and OpenAI on research replication matters
Anthropic and OpenAI are the giants of frontier AI. For a small, specialized lab to claim it beats them on a specific benchmark is significant. It suggests that the future of AI may not belong solely to the biggest general-purpose models, but to focused agents trained for deep, specialized work. For scientists, this could mean faster validation of new discoveries and fewer wasted hours in the lab.
The story behind Inherent: a British lab with a DeepMind pedigree
Inherent was founded by alumni of DeepMind, the London-based AI powerhouse. This pedigree matters. It means the team has deep experience in building advanced AI systems. Their decision to focus on scientific replication rather than general chat indicates a strategic bet that the next big AI breakthrough will come from automating the scientific method itself.
Who benefits most from an AI that can replicate research
The immediate beneficiaries are researchers in pharmaceuticals, biotechnology, chemistry, and academia. If Faraday can reliably reproduce results, it could accelerate peer review, help labs build on each other's work faster, and reduce the risk of publishing irreproducible findings. For young researchers, it could also mean less time spent on tedious validation and more time on creative hypothesis generation.
What Inherent says about Faraday's performance
According to the original story, Inherent claims Faraday outperformed Anthropic and OpenAI at the task of replicating research. The company frames this as a stepping stone for innovation. However, these results are based on Inherent's own benchmarks. Independent evaluation from third-party researchers or academic institutions has not yet been released, so the claim should be viewed with cautious optimism.
Why specialized AI agents may beat general-purpose models
The logic is simple: a general-purpose model like ChatGPT or Claude is a jack of all trades. Faraday, by contrast, is being optimized for one specific workflow — understanding and reproducing scientific methodology. This focus allows it to develop deeper reasoning patterns for that task, much like a specialist doctor versus a general practitioner. In complex fields, specialization often wins.
Confirmed facts vs what remains unclear about Inherent's claim
What is confirmed: Inherent exists, was founded by DeepMind alumni, and has released an AI agent called Faraday focused on research replication. What remains unclear: the exact benchmark methodology, the size of the performance gap, and whether Faraday's success in controlled tests will translate to messy, real-world lab conditions. Independent verification is still pending.
Why Inherent's approach could be a moat in the AI race
Inherent's potential moat lies in its data and focus. By concentrating exclusively on scientific replication, the company can build proprietary datasets of successful and failed experiments. This feedback loop is hard for general-purpose AI labs to replicate. If Inherent can build a trusted brand among scientists, it could become the default AI infrastructure for research validation.
Risks and balanced view: what if the claims are overstated?
There are legitimate concerns. First, self-reported benchmarks can be cherry-picked. Second, replicating a paper in a controlled AI test is different from handling the ambiguity of real lab equipment and biological variability. Third, if Faraday makes a mistake, the consequences in a lab could be costly. Scientists will need to verify its work carefully before trusting it fully.
The wider trend: AI moves from conversation to action
Faraday is part of a broader industry shift. AI is moving beyond generating text and images to performing tasks — booking flights, writing code, and now conducting scientific analysis. This trend toward "agentic AI" is being watched closely by investors and tech giants alike. If Inherent succeeds, it could validate the idea that the most valuable AI companies will be those that solve one problem extremely well.
What researchers and students should do now
For researchers, the practical step is to watch for independent evaluations of Faraday. For students in STEM fields, this is a signal to learn how to work with AI agents, not just chatbots. Understanding how to prompt, verify, and collaborate with specialized AI will become a core skill in the coming years. For now, treat Faraday as a promising tool under evaluation, not a proven replacement for human judgment.
Future outlook: what happens next for Inherent and Faraday
The next few months will be telling. If Inherent releases a public benchmark or invites academic partners to test Faraday, the claim will gain credibility. If other labs replicate Inherent's results, it could accelerate adoption. If not, it becomes another cautionary tale about AI hype. Either way, the focus on scientific replication is a smart bet on a real, painful problem.
Our Take
This story matters because it challenges the assumption that only the biggest AI labs can push the frontier. A small, focused team of DeepMind alumni is betting that specialization beats scale in at least one critical domain. Whether or not Faraday lives up to the hype, the direction is clear: the next wave of AI value will come from agents that do things, not just say things. For science, that could be transformative — if the claims survive scrutiny.
Frequently Asked Questions
What is Inherent's Faraday AI?
Faraday is an AI agent developed by Inherent, a British lab founded by DeepMind alumni. It is designed to replicate scientific research by reading papers and reproducing their experiments or findings, acting as a "teammate" for scientists.
How does Faraday compare to Anthropic and OpenAI?
According to Inherent's own claims, Faraday outperformed systems from Anthropic and OpenAI at the specific task of replicating scientific research. These results have not yet been independently verified by third parties.
Why is AI that replicates research important?
Replication is a critical but slow part of the scientific process. An AI that can do it quickly could accelerate peer review, help validate new discoveries faster, and reduce the risk of publishing irreproducible results, ultimately speeding up innovation.
Should scientists trust Faraday's results?
Not yet without verification. The claims are based on Inherent's internal benchmarks. Scientists should treat Faraday as a promising tool under evaluation and independently verify its outputs before relying on them in critical research workflows.
