نسخة أولية وصول مفتوح
SquidAgent: Parallelize Wisely, Coordinate Efficiently
LLM-based agents solve complex multi-step tasks, but sequential execution incurs substantial latency. In principle, parallelizing work across multiple agents should yield near-linear speedups. Yet existing parallel multi-agent systems often run slower than a single-agent baseline. We attribute this gap to two hidden co …