Others have already pointed to HN comments arguing that 23andme is mostly for novelty, but for those just skimming lw discussion that don't want to wade through pages of material, I'll highlight the strongest argument against taking 23andme seriously:

Recent research hints that 10% of ordinary healthy people have genes that we understand to be indicative of major disease. In other words, if these people bought 23andme's service, they would receive results that would be extraordinarily distressing, even while being nonetheless healthy.

See the study in question. Relevant quote: "[O]ur current best mean estimates of ∼400 damaging variants and ∼2 bona fide disease mutations per individual [is an underestimate]". (The study was brought to my attention by NPR. Note that I have not read the actual paper, but only listened to a news report on it and read the abstract.)