I disagree. Amazon is an utter failure when I don’t know the exact product I want to buy: reviews are faked and many of the products are of low quality, with no way to test them except for a terribly inconvenient return process. I’d rather just try on clothes in a store and not buy ones that I don’t like instead of having to sigh and package them up and wait for a refund.

You may see retail employees as unnecessary middlemen, but they provide a value a machine cannot.

I can talk to GameStop employees about which new games they played to see if they’ll be fun; I can talk to petsmart employees about proper pet care; I can go over installation prodcedures with the Geek Squad folks at BestBuy; I can get recommendations on outfits at the upscale men’s clothing store I frequent...and so on.

There is no superior specialization in this regard. Many of these examples are because of the relationships I have cultivated with long-term retail employees, who know me and my preferences far better than any machine learning algorithm has attempted to match.

> Aggregation, filtering, sorting, recommendations could be best handled by machines.

That only changes who those middlemen are and their scope. Based on what I'm currently seeing on search machines can only deliver exactly what I'm asking for. Rarely do they recommendations that aren't patterns. For example, I was looking for backup software a few months ago, I get tons of ad pages on backup software now. I purchased that software months ago and no longer need backup software. It's probably going to be many more months before my search history cycles out of ad networks.

Another example. A road was recently converted to pedestrian only in my neighborhood. Would you like to know the number of Lyft/Uber drivers that follow Google Maps end up at a dead end and have to double back? And that is coming from the world's most powerful search engine.

> best handled by machines

I don't think best is correct in a lot of contexts, but just good enough is enough in most cases.

Potentially but when you're at a store you can immediately tell the if the store has a good or bad purchasing department. It does provide tangible positive value.

You're silly if you think that the industry isn't already using large data analysis to do this job. There's still plenty of value by having a human at the end of the tunnel to actually interpret it though.

Do you foresee middlemen like Amazon eventually going away, or being even more essential?

Middlemen will go away, The Middleman will stay.

Yeah but presumably as the end user you'd still want to interface with a single entity, even if behind the scenes they're then aggregating, filtering, etc. But yeah any humans employed by them doing that work are probably at risk.