We believe that Multi-Agent Systems are the next step of useful AI and are essential to close the gap to artificial general intelligence. Unlike single-agent systems, they can use several specialized models for appropriate tasks, parallelize tasks to execute them quickly, and run for longer durations of time to solve more complex tasks.

However, current systems are handcrafted for specific tasks. This is a bottleneck for large-scale use of multi-agent systems. So we propose an automated framework for designing multi-agent systems. Such a framework would decide at runtime, what agents would be needed and and design workflows around them. Moreover, it would also be able to modify the design as and when needed based on both internal and external feedback.