Agentic systems are increasingly expected to improve after deployment, yet single-entity self-evolution often remains bounded by static learning contexts, such as fixed tasks and feedback. A new survey from Hugging Face introduces the concept of co-evolution, a multi-component form of self-evolution in which multiple agents and their environment impose adaptive pressure on one another.
The survey organizes existing research into a progressive three-stage taxonomy that traces how systems gradually shed human-engineered constraints. The first stage, Agent–Agent Co-Evolution, examines how agents adapt through dynamic peers, including adversarial, collaborative, and organizational adaptation. The second stage, Agent–Environment Co-Evolution, extends this loop to adaptive tasks, feedback, and interaction spaces that change alongside the agents.
The final stage, Meta Co-Evolution, explores the possibility of making the evolution mechanism itself evolvable, pushing toward systems that can self-direct their own improvement beyond human-designed paths. The authors also discuss open challenges, including evaluating such systems, scaling them across multiple components, and ensuring safety and controllability as evolutionary processes become increasingly autonomous.
This survey aims to provide a unified foundation for building robust and open-ended agentic systems that can improve beyond fixed human-designed trajectories.