There are huge swaths of azure engineering teams that deploy things entirely manually via azure portal. Insane.

Maybe years ago but definitely not today. But nice FUD

How much of it is deployed with nondeterministic tooling, though?

I remember their customer-side push for semantically merging Office and some other suites into something-or-other like 365 or Copilot, creating enormous confusion. I have difficulty believing MSFT distinguishes between handmade and machine-learning tooling in that type of environment.

I suspect there's also a good amount of LLMs/Agents in the mix. They have been rechristened to Microslop for a reason.

There is nothing micro about the amount of slop. MacroSlop.

Even as recently as 2022 (when I used to work there), infra-as-code use at Microsoft was still very nascent and non-standard. Infra at Microsoft was mostly "ClickOps", with a ton of tribal undocumented knowledge about the right order to do things, changing the defaults, and things like that.

The amount of times we've had to reach out to an experienced person who had transferred to another team/org about processes in our team was ridiculous.

> We use Azure at work and it’s wild how much stuff is just manual labor behind the scenes on their platform.

AWS is no different. Everything is on fire all the time, but customers usually don't notice.

That means AWS is doing it right, though

Or just not as badly.

at something AWS scale something is always on fire, they probably measure disk head crashes per second, and they're supposed to insulate it from their clients with layers of redundancy. If those layers are working, they're doing it okay.

Above those layers is a ton of jank, e.g. Elastic Beanstalk. But my understanding is the fundamentals on AWS - compute, storage, networking - are pretty much rock-solid.

Fair.

Can you clarify what you mean? Do you work there, or have you had negative experiences as a customer?

Disclosure: I work at AWS, at the core EC2 dataplane, no less. We monitor our production environment very aggressively with very tight SLAs, and our technical leadership is competent as well.

The petrol in a four stroke internal combustion engine is on fire all the time too. The important part is nobody notices.

What is MBO?

Management by Objectives?

(Although I like the idea of Management BuyOut, too.)

I read opinions like those but then I worked with Azure at previous jobs a lot and never had any really big issues. On the other side, I just tried using AWS for a small Saas of mine and after setting up SES, I got a shady email, from non-defult support amazon domain that I need to send them scans of my personal documents. Link in that email was blocked by uBlock Origin, domain was not the same as other emails, not a single notification in AWS Support Console anywhere so I ignored it as phishing. Got suspended two weeks later, not a single human in sight within 48 hours of me making a ticket, I just gave up and moved emails to Cloudflare. If I had actual real clients at that phase I'd be pretty pissed. I genuinely don't know how people blindly trust AWS with everything these days, it's 2026 and there's still not even a hard billing limit built into AWS.

I will never understand people crying about MS and Azure when AWS and GCP support are 10x worse. At least Microsoft usually has decent docs and support.

Hardly, Microsofts docs will actively lie to you on functionality and features. Even pointing claude at them will run opus around in circles realizing it's not concrete at all.

If the docs are correct I'm sure they'll depreciate the offering and rename it to something else and leave you on an island either way.

> I worked with Azure at previous jobs a lot and never had any really big issues

to what extent, though? over the course of my working with azure, ive been disappointed constantly, by just about every service i've used. things that should work a particular way don't, and when you try to investigate you end up in a maze of out-of-date and/or totally wrong documentation. i've started a project before, followed their precise "quick start!" guide to the letter, and find it totally non-functional and wrong.

things that DO work end getting updated (often due to financial incentives) and break, or shuddered entirely. some values that you might want to change for certain services are hard-coded (which, to their credit, you can verify from their public github repos) for ostensibly no reason. a lot of their documentation is internal only, and the public facing stuff feels perennially out-of-season.

(opinion) i feel like fundamentally microsoft has embraced ai development so wholly because they've not been able to describe their services in documentation very well, and thus rely heavily on the ai's ability to digest what scraps of truth remain and the various source code to be effective at delivering output