Hiring is bilateral: employers assess fit while candidates present and defend evidence of qualifications. Conventional résumé screening is usually a static, one‑call judgment on a résumé‑job pair. This work proposes a two‑agent alternative where an employer‑side agent and a candidate‑side agent represent the two parties, exchange evidence, and update their judgments before deciding who proceeds. We evaluate the approach on 600 constructed résumé‑job pairs using GPT‑5.5 and Claude Opus 4.7. The two‑agent screening increases advance rates—from 33.3% to 39.3% for GPT‑5.5 and from 34.0% to 35.5% for Opus 4.7. Across three runs on a common 191‑pair borderline pool, pass‑instance rates rise from 4.5% to 26.2% (GPT‑5.5) and from 6.5% to 16.1% (Opus 4.7). The improvement is not a uniform relaxation: the two‑agent method also rejects applications that would pass in a one‑call setting, changing decisions in both directions. At comparable pass volumes the two procedures advance different applications, and no single‑call threshold consistently recovers all applications selected by the two‑agent process. When discovery‑selected cases are re‑executed in fresh runs, selections unique to the two‑agent method recur less often than shared selections—clearly under GPT‑5.5 and less certain under Opus 4.7—while a separate one‑call follow‑up shows no comparable decline. The study shows that as both sides adopt agents, the screening procedure itself, not just the underlying model, shapes who reaches human review and how reliably that access recurs.
Review