Beyond text generation: large language models as autonomous agents for orchestrating synthetic biology cycles

Synthetic biology, which aims to design and construct novel biological systems, faces inherent challenges in navigating high-dimensional design spaces and optimizing complex experimental cycles. Large language models (LLMs) are emerging as semantic engines capable of parsing the deep logic of biological systems, making them particularly suitable for processing complex biological sequence data. In this review, we first systematically show how different LLM architectures are uniquely suited to specific biological tasks. We also highlight that LLMs shift the paradigm from manual feature creation to semantic-driven approaches, compared to traditional machine learning. Through a critical examination of transformative applications in novel biological sequence design, context-aware metabolic engineering, and intelligent automation laboratories, we chart both the unprecedented accelerations and the profound new challenges, including computational cost, interpretability, and ethical governance, that this paradigm introduces. We position LLMs not merely as predictive tools, but as autonomous agents that orchestrate the synthetic biology design-build-test-learn cycle. The emergence of a collaborative intelligence, where human expertise is dynamically partnered with multiple agents to navigate the vast design space of biology, steers the field toward breakthroughs in sustainable manufacturing, medicine, and beyond.