---
title: "Comment by Mahdi Jabinpour on \"The agent replaced the specific error with a generic one, and lost the clue that named the failure\" | Hashnode"
description: "The “please retry” example shows why error handling is part of product behavior, not just observability. I would test the boundary in both directions: the inter"
url: "https://hashnode.com/posts/agent-erases-specific-error-behind-generic-message/6add06c619c667bed4a2b615/comment/6add140c838a42a2d61b8dc4"
---

[Discussion](https://hashnode.com/posts/agent-erases-specific-error-behind-generic-message/6add06c619c667bed4a2b615)

Comment [The agent replaced the specific error with a generic one, and lost the clue that named the failure](https://hashnode.com/posts/agent-erases-specific-error-behind-generic-message/6add06c619c667bed4a2b615)

[Mahdi Jabinpour](https://hashnode.com/@mahdi-lumowp)

[Oct 24](https://hashnode.com/posts/agent-erases-specific-error-behind-generic-message/6add06c619c667bed4a2b615/comment/6add140c838a42a2d61b8dc4)

The “please retry” example shows why error handling is part of product behavior, not just observability. I would test the boundary in both directions: the internal event retains the cause and correlation ID, while the public response avoids exposing account balances or identifiers the caller should not see. That makes the contract precise without turning “preserve the diagnostic” into “return every diagnostic field to every client.”
