According to @DeepLearningAI, at AI Dev 25 x NYC Groq’s Head of Developer Relations showed that compound AI systems can build deep-research agents with a single API call where agents select tools, reason over results, and iterate until reaching an answer. According to @DeepLearningAI, the talk highlighted that latency is the real bottleneck in agentic workflows and that Groq’s LPU architecture is designed to keep the loop fast enough for real applications.