British lab Inherent says its Faraday agent outperforms rivals at replicating scientific research, positioning AI-run science as the next competitive front.
Inherent, founded by former DeepMind researchers, released Faraday, an AI agent it claims beats both Anthropic's and OpenAI's models at replicating published scientific papers. The company is framing this as evidence that specialized agents built for scientific reasoning can outperform general-purpose frontier models on specific high-value tasks.
This lands the same week Nvidia published research arguing the agent harness matters more than the underlying model, reinforcing a pattern: smaller teams are finding real advantages by optimizing narrowly rather than competing on raw scale.
If narrow, well-engineered agents can beat frontier labs on specific tasks, the competitive moat shifts from who has the biggest model to who builds the best-fitted system for a domain. That opens real opportunity for startups and enterprise teams that can't out-spend OpenAI but can out-engineer it on a use case.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →