Drug discovery is traditionally a slow and costly process. This study introduces a modular, multi-agent AI framework that automates early-stage discovery—from target identification to optimized hit generation. Integrating LLM-driven literature mining, generative chemistry, and predictive modeling, the system rapidly designs drug-like molecules across multiple Alzheimer’s disease candidate targets. Results show a 3–10× acceleration and a cost reduction of up to 40%. However, data quality remains critical, as poor datasets limit predictive reliability. The work highlights the power of human-in-the-loop AI and was featured at the 2025 Open Conference of AI Agents for Science. Produced by Dr. Jake Chen.