Drug discovery demands continuous scientific decision‑making across an immense research space. We introduce PharmAgents2, an autonomous system that merges knowledge‑enhanced decision making with organized large‑scale exploration for end‑to‑end drug discovery.\ \ Paper2Skill extracts transferable scientific decision knowledge from public sources and, together with expert‑crafted skills, supports research decisions.\ \ Self‑Scaling Discovery expands exploration via parallel agents that inherit research context and decision‑making capabilities across targets, hypotheses, and discovery strategies.\ \ We release PharmAgents‑Bench v0.1, comprising 121 questions across nine decision scenarios and three end‑to‑end tasks, evaluated with expert‑defined process rubrics. PharmAgents2 outperforms baselines in most decision categories and achieves the highest process scores on all three tasks.\ \ In three real‑world studies, autonomous investigations produced findings later confirmed by human experiments. For LpxH, the system hypothesized a previously unreported allosteric site, screened compounds, and identified a novel inhibitor scaffold with experimentally verified activity and binding. For JAK1, it prioritized a covalent inhibition strategy and nominated active compounds. Starting from a known bioactive molecule, Self‑Scaling Discovery explored candidate targets across the proteome, discovered a new binding target, and experimentally validated the interaction.\ \ PharmAgents2 represents a prototype of a virtual pharmaceutical company, marking a significant step toward autonomous end‑to‑end drug discovery.\ \ Review: This work tightly integrates knowledge extraction, decision models, and massive parallel exploration, provides a reproducible benchmark, and lays a solid foundation for fully autonomous drug development.