Me too, I figured I spent more time on the computer than taking pics. Now I shoot jpg and if I have some spare time I go out and shoot. If I take a good pic I share it with basic editing if any instead of waiting to get it "perfect".
I've moved two clients to colo. Dramatic cost savings. So many systems only use VMs and a few basic cloud features. Everyone knows this, but just to make the point, you can still use certain cloud products (cloud storage for example) just fine while running your primary workloads on your own hardware. Sometimes it makes perfect sense, and you just need someone to nudge you and tell you it's going to be ok.
Going to the cloud can't possibly be as cheap as owning your own hardware for obvious reasons - they have to make money somehow. Well, unless you use spot instances, which uses spare nodes.
In any case, you move to the cloud despite the cost if you need the multi-region redundancy, the management/features etc. More commonly it's because the higher ups heard everybody's doing it, but oh well :D
The main thrust of the economic argument has been on the cost of system adminstrators that maintain the hardware. Electricity and cooling being big ongoing costs, but also when AWS released it wasn't uncommon to order a server and have it take 3 months to arrive.
I think in practice the system administrators are still in the company now as AWS engineers, they still keep all that platform stuff running and your paying AWS for their engineers too as well as electricity. It has the advantage of being very quick to spin up another box, but also machines these days can come with 288 cores, its not a big stretch to maintain sufficient surplass and the tools to allow teams to self service.
Things are in a different place to when AWS first released, AWS ought to be charging a lot less for the compute, memory and storage, their business is wildly profitable at current rates because per core machines got cheaper.
I think lower cost of capital is a narrow metric and rarely reflects the real total cost of ownership once you account for ops headcount, provisioning lag, redundancy requirements, patching, and developer time.
Cloud looks expensive on sticker price, but it buys instant provisioning, autoscaling, managed databases and multi-region DR, and those benefits only pay off if you actually exploit autoscaling, reserved or savings plans, spot fleets and cost tooling like Kubecost or AWS Compute Optimizer to enforce right-sizing and kill zombie instances.
If you want cheap dev and UAT keep them on on-prem metal or cheap colo, but automate with Terraform and run reproducible runtimes like k3s or devcontainers so environments stay consistent and you do not trade lower capex for a creeping operations nightmare.
I think it's incentives. Working with AWS is good for your resume. It's also a responsibility thing. If AWS is down then it's Amazon's fault. If there's a problem with your physical server, then it's your problem.
A broader thing here is -- and you may also notice this trend in software -- employees are incentivized towards complex solutions, while business owners are incentivized towards simple solutions.
Everything I've read says that self-hosting doesn't become cheaper than AWS for companies until you reach $1-$3 million per month spending when all costs are accounted for. Then there is the highly overlooked aspect that a good API like AWS has lets your expensive admins actually get things done hundreds of times faster than how most self-hosted IT can do. It usually takes months to buy and install additional capacity for most companies.
> good API like AWS has lets your expensive admins actually get things done hundreds of times faster than how most self-hosted IT can do
Depends. APIs must take into account many more cases than our own specific use case, and I find we are often spending a lot of time going through unnecessary (for us) hoops. And that's leaving aside possible API changes.
I think it can make sense for user-facing services. I host my web and database servers with AWS because unmanaged DBs can be a PITA, Amazon is peered with basically everyone, AWS is way more generous with network speeds than many dedicated / colo providers, and it’s easy to scale capacity up and down. Backend servers are hosted with cheaper providers though.
Not from people using that same phrasing twice within a few sentences.
"""
No warning. No traffic spike. Just… more money gone.
That’s when I finally looked at Hetzner.
I’ve seen too many backend systems fail for the same reasons — and too many teams learn the hard way.
So I turned those incidents into a practical field manual:
real failures, root causes, fixes, and prevention systems.
No theory. No fluff. Just production.
"""
It's clearly slop, they immediately use effectively the same one again:
"""
That last line isn’t a joke. There were charges I genuinely couldn’t explain. Elastic IPs we forgot to release. Snapshots from instances that no longer existed. CloudFront distributions someone set up for testing.
"""
No, human writers don't repeat this pattern every single paragraph. They use it at most across in a whole article.
Repetition is a very common tool in writing (ie 'I have a dream').
I'm just irked that it's being called out for AI slop because "I feel it in my bones!!"
There's a good chance it was written using AI -- should that matter? If the content is wrong/sucks, say that instead. If you're going to dismiss all AI assisted writing: good luck in the next decade.
Reading the same (very annoying marketing blog) style of writing gets old fast.
It’s like suddenly all memes are just the same meme and nobody makes their own memes because “AI does it better”.
The style of writing is an intrinsic part of communication, if you can’t critique that then what is content? We’re not machines sharing pieces of data with each other.
AI-written text is not necessarily incorrect, but if the author did not take their time to remove the AI slop, they probably did not put much effort into it elsewhere. In addition, the text is often over two times longer than without the slop, which disrespects the reader's time (even worse in this case, since a significant fraction of the article is an ad for the author's books).
Maintaining and updating your own hardware comes with so much operational overhead compared to magically spinning up and down resources as needed. I don’t think this really needs to be said.
It’s absolutely not a one time cost. Once you have it you need to hire people full time to maintain it and eventually upgrade it. Hardware fails constantly
I had the same inclination back in the 90s when I upgraded my Cyrix 486 SLC2 50MHz without a heat sink (which seems like a no-no in retrospect) to Cyrix MediaGX 133MHz. The stocker fan was immediately noticeable. I thought I had done something wrong.
Upgrading and Repairing PCs 4th edition even says directly, that some shady resellers will put a heatsink on a chip that they're running beyond spec, but that Intel designs all their processors to run at rated speed without one.
The cpu and heatsink was fully integrated into what looked like a NES cart, with an integrated fan and everything. It was not really possible to separate the cpu and the heatsink as the locking mechanism to keep the cart in place on the motherboard interfaced with the heatsink assembly.
So I'm a little dubious of that no-heatsink claim.
No, Windows NT until 4.0 had the same interface design as Windows 3.x (although there existed a semi-official SP/addon to give NT 3.5 the Chicago interface, making it quite similar to 95), and NT 4.0 came later than 95
Both OS lines were developed concurrently up until XP release where DOS-based 9x was abandoned and NT became the basis for every subsequent product. Plus of course there's that whole part of the story where MS teamed up with IBM and worked on OS/2.
NT got new 9x shell with 4.0 release but a beta package could be installed on 3.51 as well - tho, that could render some compatibility issues.
Actually I'd say Win2K was the merge point: its internal version is NT 5.0, while XP is NT 5.1 & 5.2. The Win2K UI is the first NT truly usable in a home situation, and last and best iteration of the Win95 UI before the plasti-color of XP. (Yes, that UI was last available as XP's "Classic Theme", but I'm giving it to 2K because XP doesn't really change anything.)
However Win2K wasn't really marketed to consumers as there were still some minor compatibility issues: particularly the DOS emulation was not great, so getting things like older games to work often required lots of tweaking to the launch process, and some things still might never work. Those compatibility settings got more options, saner defaults, and more automatic settings, in XP and later to go along with the full commitment of NT for everyone.
So, while DOS-based WinMe was actually released after Win2K, it was just a stop-gap to bring some more internet things, directx, and media player stuff, to home users while NT 5.x got it's compatibility and driver model in order. Except it was notoriously unstable, generally hated, and mostly forgotten.
I use Opus 4.6 (for complex refactoring), Gemini 3.1 Pro (for html/css/web stuff) and GPT Codex 5.3 (workhorse, replaced Sonnet for me because in Copilot it has larger context) mostly.
For small tools. But also for large projects.
Current projects are:
1) .NET C#, Angular, Oracle database. Around 300k LoC.
2) Full stack TypeScript with Hono on backend, React on frontend glued by trpc, kysely and PostgreSQL. Around 120k LoC.
Works well in both. I'm using plan mode and agent mode.
What helps a ton are e2e playright tests which are executed by the agent after each code change.
My only complain is that it tends to get stutters after many sessions/hours. A restart fixes it.
As long as we're on the subject, I'll take the opportunity here to vent about how embarrassingly buggy and unusable VS Code is in general. It throws me for a loop that pros voluntarily use it on the rare occasions I'm forced to use it instead of JetBrains.
I use Claude Code in Zed via ACP and have issues all the time. It pushes me towards using the CLI, but I don’t want to do things that way because it’s a vibe coding workflow. I want to be in the drivers seat, see what the agent has done and be able apply or reject hunks.
I’m in the same situation. Zed’s Claude Code is better in terms of control, but it’s wildly buggy and unreliable. Definitely not a drop in replacement.
I’m in the same boat. I use it to save me from going to a browser to lookup simple syntax references, that’s about it. Its agent mode is terrifying, and asking it anything remotely complex has been a fool’s errand.
This article is mostly based on NBER working paper 34836, which was published this month, and the data was collected from September 2025 to January 2026[0]
reply