The rapid advancement of quantum computing poses fundamental and systemic threats to the public−key cryptography underpinning financial market infrastructures. Although regulatory authorities across jurisdictions have issued policy frameworks for post−quantum cryptography (PQC) migration, practical progress remains slow. This paper argues that the bottleneck is not algorithm choice: the standardized PQC algorithms and their parameter sets are few in number and are effectively fixed by standards and regulatory policy, leaving little decision space. The real bottleneck is the asymmetry between the scale of legacy systems and the supply of cryptographic expertise. A large financial institution may contain tens of thousands of cryptographic call sites scattered across hundreds of heterogeneous legacy systems and vendor−delivered code, while engineers competent in PQC remain scarce, making it unrealistic to staff every legacy unit with such experts. Accordingly, this paper redefines the role of AI agents in migration as knowledge delivery to legacy system owners who lack a PQC background: discovering cryptographic call sites in code, configuration and third−party components; determining their business semantics and applicable compliance requirements; generating adaptation recommendations and remediation plans traceable to normative provisions; and leaving decision authority with human reviewers. Four capabilities are proposed − discovery, determination, generation and verification − together with an autonomy grading scheme. Five failure modes specific to this positioning are analyzed: missed migration points, attribute misjudgment, non−compliant recommendations, stale rule bases and over−trust. Using an open−source system with financial characteristics as a case, the paper further examines the actual scale and difficulty of migration point discovery, and presents an evaluation design that requires no production environment.