Experience assessing the quality of AI-driven research processes, including identifying false positives, missed vulnerabilities, unstable results, and reproducibility gaps, as well as sound judgment on agent architecture trade-offs: when constrained, orchestrated pipelines deliver reproducible results and when open-ended tool-using agents are worth the nondeterminism. 5+ years of hands-on vulnerability research, offensive security, application security, or penetration testing, with a demonstrated track record of choosing targets, forming and revising hypotheses, and establishing exploitability and impact in complex software or production systems.