← Monday, August 24, 2026
AI Frontier · Aug 22
ModerateNew AI teammate redoes science tests on its own and beats bigger rivals
A London startup named Inherent built an AI agent called Faraday. It can read a science paper and redo the tests on its own, with no answers given first. In one contest it did this better than AI from Anthropic and OpenAI. It won even though it runs on a much smaller model. The team taught it to pick which tests are worth trying.
Why it matters to GCs
This shows AI agents are getting good at hard work with little hand-holding. The same kind of agent now shows up in construction apps to draft RFIs and check plans. Knowing what agents can do helps you decide how much to trust them on the job.
TechCrunch·Via POLR AI Hub