I think, it really depends on the question you are asking!

In case you're asking, what's the transcriptome of all small RNAs (including miRNAs), then go for bulck sequencing.

In case you're asking, which miRNAs are currently and (probably) effectively acting in post transcriptional silencing of target mRNAs, then go for an AGO pull down and extraction of RNA from that pull.

Both are justified questions that will result in fruitful outcome, as long as the efficiency and quality of downstream methods work.

I am no miR-seq expert, but have a rather large experience with standrard RNA-seq. As I view it, your question is similar to should we do ribosome pull-down before RNA-seq. It might give you a more biologically significant view of what genes are expressed (only translated genes are pulled-down compared to total RNA), but quantities are reduced and it may be more difficult to synthesize the sequencing libraries. My conclusion is that AGO pull-down is not mandatory but may give you a better answer to your biological question although with more experimental difficulties.

In my opinion, that's not necessary.

According to our experiences, miRNA accounts for more than 80% of the sequence reads when the samples are prepared with the standard Illumina protocol and sequenced with MiSeq or HiSeq.

So, when MiSeq or HiSeq generates 5M total sequence reads, the user have 3~4M miRNA reads, which is necessary for un-biased miRNA expression profiling.

I think, it really depends on the question you are asking!

In case you're asking, what's the transcriptome of all small RNAs (including miRNAs), then go for bulck sequencing.

In case you're asking, which miRNAs are currently and (probably) effectively acting in post transcriptional silencing of target mRNAs, then go for an AGO pull down and extraction of RNA from that pull.

Both are justified questions that will result in fruitful outcome, as long as the efficiency and quality of downstream methods work.

It's not a simple manner to extract enough RNA from an Ago pulldown for good small RNA library preparation, but it is feasible. Our group compared miRNA libraries synthesized from total mouse heart RNA with libraries synthesized from Ago pulldown RNA and found that the relative abundance of miRNA fraction was similar (see Hu et al., PNAS 2012, PMID: 23150554, in the Supplement). So my feeling would be to at least start with libraries from total RNA.  If you are challenged by a reviewer (as we were!) to demonstrate what is happening in the Ago fraction, then save it for that stage.

Best way we have done this is a size selection to exclude anything that is too large to be miRNA. This can be done by manipulating the EtOH concentrations during RNA precipitation (you need higher concentration of EtOH to immobilize smaller RNA molecules than larger ones), or gel extraction.