While there are some good bits in the dragon book, some of its advice on the engineering of a compiler is pretty dated. For example, most of the parsing chapter(s) seem to have the notion that there is a single (global?) symbol table to which tokens and various other things refer, and which is gradually refined and updated throughout compilation. I think that may have been a widespread design once, but it's a bizarre anachronism now. Nowadays, information is often embedded directly into the AST, and compiler passes often create and destroy multiple ad-hoc symbol tables as necessary.

I would primarily trust the dragon book for its theoretical content, although even that likely no longer represents the state of the art. I'm not all that knowledgeable about the state of the art in parsers, but I know that compiler optimisations have certainly moved beyond what is in the dragon book.

In short, while the dragon book still contains useful information, I don't think its notion of how a compiler is constructed is close to how compilers are actually constructed today.