As somebody who deals with AWS professionally on a daily basis, I will 100% admit that it can be confusing for people to figure out pricing and scale. I have to talk with engineers every week about design decisions and their cost impacts, and the implications of the things they want to run, and it's clear that AWS doesn't give enough guidance in the UI or documentation structured in a way that a layperson could find what they're looking for quickly and easily for that information.

That said? This article rubs me the wrong way. The suggestion that this problem is caused by "dark patterns" and Amazon being misleading about "pay as you go" screams FUD to me. The screenshots provided right in the article clearly show that what was being done wasn't eligible for free tier. And not understanding that launching a resource, regardless of whether or not you're interacting with it, is consuming the service which launches the resource is a problem with the user, not the marketing. Yes, the author admits that they didn't scrutinize, but that doesn't excuse the position of the article that AWS is somehow doing this to intentionally bilk people.

So, let me get this straight: - OP ignores the obvious "This does not fit the free tier" warnings when setting up their app. - OP does not pay attention to any billing metrics or even bother to try and understand the pricing beforehand. - OP gets hit with a $990 bill. - Amazon gives all the money back, plus free credits. - OP complains.

AWS should have kept their money and OP should've learned their lesson proper.

If we are trading war stories:

I was playing around with some tutorial to learn something (probably something cool like programming your own robotic drone using functional erlang or whatever), pushed to github and went to sleep. Woke up a few short hours later and had lots of emails about the machines I was spinning up.

Checked and saw that my account had wracked up thousands of dollars overnight (I think 6-8 hours), and I started to shut down the machines.

I didn't get them all, there were more machines hidden, and the bills continued to pile up for another hour or two.

I contacted Amazon who shut it all down, and I reset my password.

Then I realized I had pushed my credentials to github (I should really put this under a pseudonym, but I was new to the whole thing and hadn't even looked into Amazon's authentication system. Obviously, billing credentials and sysadmin credentials should never be the same.) Someone had a scraper going that picked them up almost right away.

To Amazon's credit, they cancelled the charges within a few hours, and if memory serves the person investigating gave me a sympathetic but stern message.

I don't know who the credential-stealer was and what they were using it for, but I would guess crypto mining. I did some calculation at the time and I think they would have extracted about 1/3rd the value of my bill, but those were rough calculations.

I really don't think this is a UX problem. It was pretty clear that there were two options, production-ready and free. If you want to be picky, I suppose you can be upset that RDS is just a couple of VMs that you can't run other things on, or to question how much performance benefit a certain number of provisioned IOPS gives you. I don't think that's a dark pattern so much as "we don't know what your workload looks like, you don't know what your workload looks like, so just provision a bunch of IOPS and hopefully we never speak again."

I am less surprised that the mental model fell apart. I guess a lot of people think cloud resources are something that is efficiently shared (consider S3, you pay per byte you store, store 0 bytes, pay $0). But that's actually a rare case, most of the time you are provisioning something for your exclusive use; if you have a database server it costs you the same whether it's doing 10000 transactions per second or sitting completely idle and never logged into.

(Incidentally, the true sharing model used to be popular. Shared hosting with no isolation between tenants predated AWS by a decade. You got a chunk of a computer and shared Apache, MySQL, and PHP with hundreds of other randoms. Very cheap!)

If using AWS for personal use the first step should always be to set up a billing alarm. https://docs.aws.amazon.com/AmazonCloudWatch/latest/monitori...

It’s very helpful. You can still end up overspending but at least you get an email within a day letting you know what’s going on, which can solve a lot of the cost overruns by giving you a chance to act quickly and only get hit with 1/30 the monthly fee.

My girlfriend's first "Free Tier" hosting build was $5200 for ingesting a single document on Kendra. After nearly a month of working with the service team she was able to get the bill removed (even the service team did not know how to delete the app). Its insane that an overage charge on a "Free tier" service can be the price of a used car or multiple months of rent.

I'm sorry, but that's on him. Look at the first screenshot in "Default Configuration". It says so very clearly. Twice. He admitted he didn't care, this is what happens when you don't.

I wish AWS had hard spending limits. Azure have one - you spend over the set limit(probably per billing account?) and your services are suspended. Already saved my from unexpected bill this month.