ICD-10, the International Classification of Diseases tenth revision, is the diagnosis code set every clinical claim is coded against. A diagnosis of "type 2 diabetes without complications" maps to E11.9; "essential hypertension" to I10. There are about 70,000 codes in the US clinical modification (ICD-10-CM).
For a clinical AI, ICD-10 mapping is a non-trivial NLP task. The patient says "my sugar has been high," which the agent has to map to a diabetes-related code if the provider confirms it, while resisting the temptation to commit to a code the provider has not yet selected. Auto-coding before clinical sign-off is exactly the kind of regulated work an AI should not do unsupervised.
The right pattern is suggestion, not authoring. The AI proposes a candidate code with the supporting transcript excerpt, the provider confirms or overrides, and the EHR records both the final code and the provider attestation. A vendor that auto-codes without provider attestation is creating billing risk.
Vorel clinical agents suggest ICD-10 codes from transcripts and link the supporting language. Provider attestation is required before the code is committed to the encounter record.