Why did one day of AI cost more than a month of servers?

https://junueno.dev/en/retry-storm-rebilled-llm-cost/

14•dxs•1h ago

Comments

ieie3366•1h ago

The article itself is AI slop.

nativeit•1h ago

I find the contextless, reflexive labeling of every article on HN as “AI slop” far sloppier and lazier than the articles being criticized. I am no fan of these trends, nor the companies driving them, but this sort of lazy critique has become counterproductive, IMO.

jsnell•2m ago

But it is AI slop. It's obvious that the text is all AI-generated, there's half a dozen different tells that punch you in the face from the first paragraph and never stop. Humans just don't write like this. (And fwiw, Pangram flags it as 100% AI-generated).

This particular blog also has the benefit of having some pre-LLM history. You can see that the older writing style is totally different.

So the "author" is already being lazy and dishonest in presenting this as their own work. Why would we believe any part of the story? Why are you trying to give the benefit of the doubt to something so egregiously bad?

athrowaway3z•1h ago

Senior Cloud Engineered Pipeline(d) Slop

Frankly i'm fine somebody having an experience, and pulling it through an AI into digestible text to be shared.

But for the love of god just put an AI Use Disclaimer at the top.

whywhywhywhy•1h ago

Keep repeating what happened over and over in different ways and different analogies adding nothing each time.

wdave_t•1h ago

I have the same feeling. Interesting topic but the text feels generated, perhaps because of the constant reiteration of the point

tomrod•39m ago

So so much AI slop in these posts that usually would be very interesting. I don't like Claude's voice in most situations. The LLMisms stand out like nails on a chalkboard.

elzbardico•1h ago

Really folks. That idea of people without experience and knowledge of software engineering vibe coding production code is a little bit not realistic, no matter how much Claude wants to convince investors of the contrary.

Coding agents are fantastic tools, but they are not Jesus: they don't do miracles.

petcat•1h ago

Perhaps because your server fleet is running on commodity Xeons and not a million nvidia GPUs

k08200•1h ago

A big part of the cost explosion is using the frontier model for tasks that don't need to be done. I tried using the gpt-4o and much cheaper models, and the cheaper ones were more accurate in my three — paying for the reasoning depth that I don't use. The other half is asking the model to do what deterministic rules should be. Calls that don't are the cheapest. Starting with profiling what calls the larger model really needs.

initramfs•1h ago

Very interesting. I am guessing the company didn't offer refunds if the receiving computer wasn't able to process the results, unless the issue was on the service...

mid90sahsan•52m ago

If this was done by a dev in my company, they would have fired him that instant.

swiftcoder•52m ago

> A deterministic failure doesn’t get better when you retry it

Among all the other nonsense here, this one is solid advice. Most software with retries just blindly retries N times - in reality, you need a retry/backoff policy based on the specific type of error encountered (I've been rooting these out of our own software stack lately)

TacticalCoder•35m ago

Yup exponential backoff. But this is all just bandaid: like saying there should at least be some tests.

It doesn't change the underlying problem though: slop is slop and that turd the CFO produced would still be a turd.

"Here's a piece of shit that doesn't work, but at least it's cheap" is no way to operate.

jeffbee•16m ago

This operation should not have been retried even once. The overall operation failed due to an unsatisfied precondition that a second attempt could not have rectified (the non-existent target resource). The status of the job should have transitioned to permanent failure, awaiting manual intervention.

Cowboys, Frontiersmen, Settlers, Townspeople, Cityfolk

WSL container is now available for public preview

Silicon Valley Gets High on Its Own Supply

A TacoSprint 2026 Retrospective

Show HN: Cline subscription plan to access GLM-5.2 at 2-5x discount

Flipper Device's new Busy Bar is a customizable display for productivity

How Small Postgres Metadata Tables Throttle Your Largest Queries

A reliable unprivileged container jail escape proof of concept for CentOs/RHEL

Show HN: PMB – local memory for coding agents that shows if it is used

Cyberpunk Edgerunners 2 is releasing this year

Ask HN: Homeless, Former Software Developer, What Now?

Truckloads of Tesla Batteries Keep Getting Stolen Before They Leave the Factory

Testing is f***ing awesome

Adobe to Acquire Topaz Labs

Open Source Software Is the Pastime of the Rich

Sound Effects for Free Use from the Finnish Broadcasting Company, Yle Archives

Ask HN: How do you handle QA at a startup with no QA team? Genuinely curious

Digital Sovereignty at the UN

Loop Engineering Is Just Software Engineering

Subveris – A clean, free dashboard to track your subscriptions

Microsoft Needs Windows Lite

Leaked A20 Pro Image Hints at iPhone 18 Pro Performance Gains

We took away psychological safety and then told everyone to be more productive

AMD contributes their GPU support to tiny-vLLM

Hotels cheat with AI-optimised images

Accurate Decoding of Natural Sentences From Non-Invasive Brain Recordings

Show HN: macOS Native Markdown Reader with Vim Keybinding Using Swift

Projection: A JJ Workflow for splitting public and private files

Show HN: Add visualisations to your data MCP in a few lines of code

Show HN: Timefeed – Auto sync public events to your calendar