Imagine a construction site where robots work seamlessly together, digging, hauling, and building, all orchestrated by the power of words. That future is closer than you think. Researchers have developed DART-LLM, a system that uses large language models (LLMs) like the ones powering ChatGPT to coordinate teams of robots through complex, multi-step tasks. Unlike previous attempts at robot collaboration, DART-LLM understands the dependencies between actions. It knows that the dump truck shouldn't leave the excavation site before the excavator finishes loading it. This dependency awareness is key to efficient teamwork. DART-LLM breaks down high-level instructions (like “clear the obstacle, then dig soil”) into a sequence of smaller, robot-specific commands. It then manages these sub-tasks, ensuring the correct order of operations while allowing independent tasks to run in parallel. The system was tested with simulated construction robots in various scenarios, from simple inspections to complex, coordinated excavations. Remarkably, even smaller LLMs achieved impressive results, outperforming larger models in some cases due to their superior instruction-following abilities. The real-world tests further validated DART-LLM's potential, showcasing its ability to control real robots with remarkable efficiency. While still in its early stages, DART-LLM suggests a future where human language can directly control complex robotic systems, opening doors to more automated and efficient construction, disaster relief, and even space exploration. The biggest challenge ahead lies in scaling up the system for larger teams and more dynamic environments. However, the groundwork laid by DART-LLM represents a major step forward in bridging the gap between human language and robotic action.