This paper provides an analysis of the Korean honorification system as agreement. Under the proposed analysis, subject honorification is triggered by an unvalued honorific feature right below T probing downwards to agree with the subject. Addressee honorification is triggered by another instance of the same feature, which probes upwards and agrees with the c-commanding addressee representation. I adopt the bidirectional agreement model (Arregi & Hanink 2021) that enables a unified analysis of both types of honorification. In the second half of the paper, I discuss a crosslinguistic variation among honorification languages. When the subject is 2nd person and thus co-referent with the addressee, Korean and Japanese express honorification of both subject and addressee while Magahi and Tamil actually disallow the co-occurrence – they only allow subject honorification. This crosslinguistic difference can be captured in a straightforward way under my analysis and two well-motivated assumptions: parametric variation in the position of the allocutive probe (Alok 2020; 2021), and an intuitive re-interpretation of Kinyalolo’s Constraint (Kinyalolo 1991; Carstens 2005).