orthologs are copies of the same genes within the same species.

This is wrong. By definition, orthologs are genes derived from a speciation event, paralogs are derived from a duplication event. Orthologs are what is usually meant when people talk about "the same gene in different species". Homologs are genes that derive from the same ancestor (= orthologs+paralogs). Formally, these relationships can only be inferred from a phylogenetic tree although short cuts and approximations are often used for computational reasons. How much effort should go into assembling gene families depends on how much one cares about the evolutionary history of the genes.

So, from your response and that of Jean, it would be ok to use HMMer or blastp to identify a gene family within the same assembly or even a group of assemblies based on domain structure, but you'd want to use something like orthomcl to infer history of the genes?

In this case my ultimate goal is to create a tree showing the members of a large and diverse gene family. If I'm understanding the orthomcl > phylogenetic would be somewhat redundant and heavily focused on history, while the blastp/hmmer > phylogenetic tree method would first choose based on structural/functional similarity and then add a layer of history. Am I misunderstanding?