I'm sure I'm preaching to the crowd, but there really should be some safeguards that can trip along the way for what is clearly an individual account. Also there should always be hard limits available to turn on (maybe turned on by default).
Perhaps some kind of "we only let your bill be X% higher than the rolling average for your account" coupled with "First time you hit $10, $50, $100, $500, $1K .... firebreaks, alarms coupled with auto-stop/shutdown if you don't respond with a day or two". All of this would be available to turn off so it would just serve as a safety net.
> Also there should always be hard limits available to turn on (maybe turned on by default).
Absolutely.
As I said on here a few days ago in a discussion about AWS finally launching hard caps ... I really don't buy the stories the US providers tell you that "its too difficult" or "what if you suddenly go viral".
The "viral" bit is easily solved through basic monitoring of metrics that everybody should be doing. I believe the cool-kids give it the fancy name of Site Reliability Engineering (SRE). All you need to do is top-up your balance / adjust your cap if your metrics are trending upwards for an explainable reason. Its not rocket science. Just set a reasonable yet affordable buffer above your most ambitiously expected figures.
As for the "too difficult" that's just a lie. Its 2026. You're a cloud provider.
> All of this would be available to turn off so it would just serve as a safety net.
You can implement this yourself with the API.[0] Unlike a naïf disable everything setting, you can shutdown services selectively and trigger specific shutdown stuff in your services so you don't lose any data.
I think the OP just points out that if you never read the code you are playing with your wallet. It's a gamble and in this case it resulted in considerable losses for an individual.
I agree, up until the question is how transparent the risks are.
For example, I think that OpenAI's system of incentivizing automated credit purchases borders on being predatory. Not because it would be hidden info. Just because the product is clearly aimed at consumers who do things that they might not fully understand.
For me, I don't habe any issues with their current design.
But for John Doe vibe-coding their nutrition analysis app, I feel it's wrong to advertise automatic payments and long-running tasks without providing a clear fee cap option.
"The market" is not a benevolent fairy godmother with your best interest at heart.
It is a machine that optimizes profit and absolutely nothing else. It does not and cannot consider what is "good" or "right" or even "legal" unless and until one of those things directly and indisputably hurts profits too much.
This is magical thinking. "The market" is fundamentally incapable of making things better for you, the consumer.
> You can't demand scaling and reliability and a hard fee cap.
Why not? Just because you can scale to million or billions doesn't mean that you should do it automatically without any supervision. Just like because your BMW M5 can reach a top speed of 305 km/h, you shouldn't drive at that speed all the time. Heck, some should never drive at that speed.
Auto-scale to one hundred bucks cause I'm a poor student and that's all I have.Auto-scale to a million cause I'm a big corporation and I can deal with a bad fiscal year. By the way, I wonder how is the debt collection auto-scaling so to speak.
Often the most expensive lesson of your life (as OP put it) is also the most valuable lesson. Everybody makes mistakes, learn to expect failures so you can mitigate the effects when they happen.
Use fixed price hosting resources, those cannot bite you and if they're too small you can always expand by buying more virtual hosts, or even program your own simple scaler (which i am 99.9% sure you will not need this or next year after starting a project). AI is capable of setting up local database, and everything you need without paying 10x to a cloud provider.
The vast majority of my personal stuff runs on $6-$12 VMs. No matter what happens, what dumb mistake I make or an agent makes on my behalf, I can only ever be out $6-$12. And, this is exactly why. The "free tier" all these big cloud companies offer is kind of a trap. Sure, you can do a lot for free, but mistakes happen and they've got your card on file.
I had the exact same thing happen a few weeks ago - a DO alarm accidentally set to a past time, retriggering immediately. Claude caught it within a couple minutes from its log tail, though.
Learning to "trade" options probably? A lot of engineers nowadays go thru an X/Twitter inspired "I'm an amateur quant" phase that usually doesn't go well.
33 comments
[ 0.23 ms ] story [ 83.4 ms ] threadPerhaps some kind of "we only let your bill be X% higher than the rolling average for your account" coupled with "First time you hit $10, $50, $100, $500, $1K .... firebreaks, alarms coupled with auto-stop/shutdown if you don't respond with a day or two". All of this would be available to turn off so it would just serve as a safety net.
Absolutely.
As I said on here a few days ago in a discussion about AWS finally launching hard caps ... I really don't buy the stories the US providers tell you that "its too difficult" or "what if you suddenly go viral".
The "viral" bit is easily solved through basic monitoring of metrics that everybody should be doing. I believe the cool-kids give it the fancy name of Site Reliability Engineering (SRE). All you need to do is top-up your balance / adjust your cap if your metrics are trending upwards for an explainable reason. Its not rocket science. Just set a reasonable yet affordable buffer above your most ambitiously expected figures.
As for the "too difficult" that's just a lie. Its 2026. You're a cloud provider.
You can implement this yourself with the API.[0] Unlike a naïf disable everything setting, you can shutdown services selectively and trigger specific shutdown stuff in your services so you don't lose any data.
For example, I think that OpenAI's system of incentivizing automated credit purchases borders on being predatory. Not because it would be hidden info. Just because the product is clearly aimed at consumers who do things that they might not fully understand.
For me, I don't habe any issues with their current design.
But for John Doe vibe-coding their nutrition analysis app, I feel it's wrong to advertise automatic payments and long-running tasks without providing a clear fee cap option.
It's similar to gambling.
It is a machine that optimizes profit and absolutely nothing else. It does not and cannot consider what is "good" or "right" or even "legal" unless and until one of those things directly and indisputably hurts profits too much.
This is magical thinking. "The market" is fundamentally incapable of making things better for you, the consumer.
If you have oligopoly without the regulations who do you think benefits more from that?
Why not? Just because you can scale to million or billions doesn't mean that you should do it automatically without any supervision. Just like because your BMW M5 can reach a top speed of 305 km/h, you shouldn't drive at that speed all the time. Heck, some should never drive at that speed.
I think some warnings would be welcome.
But yes, these platforms should default to hard caps that you have to explicitly disable.