[dehrmann]

Generally, things get harder as you scale because something that would have taken a new Postgres table has become a distributed systems problem.

You also hit diminishing returns with new features and user growth, but when you have hundreds of ~~billions~~ millions of users, even marginal engagement gains can multiply out to something big, so you have a lot of teams working on features that might move the needle a bit.

There are also more regulatory hurdles as you grow, so some of the staff just support that.

[mtnGoat]

Hundreds of billions? Is there a source to this idea that Twitter has these numbers?

[bjtitus]

Surely you aren't questioning that Twitter has users beyond our Solar System.

[1over137]

Come now, it was surely a typo, there aren't even that many humans!

[edflsafoiewq]

The rest are bots.

[dehrmann]

Nice catch. Yes, millions.

[mtnGoat]

No worries. I figured it was, but thought maybe Twitter was making up numbers somewhere.

[Jensson]

Yes, things gets harder at scale which is why he said 50 and not 5. Depending on how much infrastructure code you do inhouse you add another 50 for managing that. If you have your own datacenters add another 100 or so if you have a few around the world. If you want world class recommendation add 100 or so data scientists.

That gets us 300 tech people to run the service.

That is for engineering, those numbers are roughly what I saw at Google for these kinds of things. Then Google typically has 1 non tech person per tech person, so add in at least as many people again. Then since Googles customer support and community management is hardly world famous for being good, rather it is infamous for being bad, you probably want even more non tech people than that. But still, do you really need 8000 people for it?