No, usually you have to validate some kind of authentication tag after agreeing keys. So together with the key establishment you can already calculate e.g. a HMAC value over the previous messages and verify that HMAC once the shared secret is computed over there as well. This is e.g. how TLS works.
Of course, you can also detect tampering after receiving an authenticated message, but in that case you have the trouble that you don't know which part of the protocol was perturbed. This may not sound that big of a problem, but in many protocols it is the difference between entity and message authentication, and this may for sure complicate error handling within the protocol.
This was for instance an issue with the EAC/PACE protocol in ePassports which was fixed in the German EAC2.