Imagine a team of digital explorers that never sleep, constantly refining their own skills to solve problems no human has cracked yet. Current AI methods often rely on rigid, pre-written rules that stifle creativity and limit how far these agents can push boundaries. Researchers have now built CORAL, a new framework that frees AI agents from fixed scripts, allowing them to explore, reflect, and collaborate using shared memory and asynchronous teamwork. This shift replaces stubborn control with dynamic autonomy, enabling agents to manage their own health, resources, and workspaces without constant human intervention. The results are striking: on ten complex tasks ranging from math to systems optimization, this autonomous approach outperformed traditional methods by three to ten times while requiring far fewer attempts. In a specific engineering challenge, four co-evolving agents improved the best-known score from 1363 to 1103 cycles, proving that letting AI learn from itself creates faster progress. By combining knowledge reuse with multi-agent communication, CORAL demonstrates that true autonomy is the key to unlocking endless discovery. Source: CORAL: Towards Autonomous Multi-Agent Evolution for Open-Ended Discovery by Ao Qu, Han Zheng, Zijian Zhou, Yihao Yan, Yihong Tang et al., https://arxiv.org/abs/2604.01658