For some reason LinkedIn thinks I'm super into maritime content.

What is Koreo? The platform engineering toolkit for Kubernetes Last month we open sourced Koreo, our “platform engineering toolkit for Kubernetes.” Since then, we’ve seen a lot of interest from folks in the platf…

What is Koreo? bravenewgeek.com/what-is-koreo/

CNCF and Synadia Align on Securing the Future of the NATS.io Project SAN FRANCISCO and San Mateo, CA – May 1, 2025 – The Cloud Native Computing Foundation® (CNCF®), which builds sustainable ecosystems for cloud native software, and leading edge innovator Synadia today…

Glad to see this resolve www.cncf.io/announcement...

They tightly couple business logic to the database.

Protecting NATS and the integrity of open source: CNCF’s commitment to the community When a company contributes a project to the Cloud Native Computing Foundation (CNCF), it’s not just sharing code—it’s making a commitment to the open source community. It’s a pledge to uphold open…

Highly disappointing www.cncf.io/blog/2025/04...

The YAML Engineer A new approach to Kubernetes configuration management and resource orchestration.

The future of infrastructure management is controller-driven. theyamlengineer.com/controller-d...

The YAML Engineer A new approach to Kubernetes configuration management and resource orchestration.

Stop Treating YAML Like a String theyamlengineer.com/stop-treatin...

Bulk importing and exporting existing Google Cloud resources  |  Config Connector Documentation

Yes, it's a common ask (as well as going the opposite direction with exporting to TF). Historically, for GCP we've used Config Connector to do bulk imports of existing resources as well as exports to Terraform. cloud.google.com/config-conne...

I don't know if there is similar tooling for AWS/ACK?

The YAML Engineer A new approach to Kubernetes configuration management and resource orchestration.

For a bit more on the motivation and thinking behind Koreo: theyamlengineer.com

With Koreo, you can now viably implement your own internal developer platform in a way that is manageable, scalable, and—critically—actually testable.

Koreo enables platform teams to implement powerful workflows and resource orchestrations that enable the automation of everything from simple deployments to entire cloud environments.

The platform engineering toolkit for Kubernetes | Koreo Koreo is a new approach to Kubernetes configuration management empowering developers and platform teams through programmable workflows and structured data

A core part of Real Kinetic's business is helping organizations establish platform engineering as a practice, but the existing tooling available today is lacking. That's why we're excited to open source Koreo, the platform engineering toolkit for Kubernetes. koreo.dev

Pluralizing rest resource names in APIs is a mistake · Issue #18622 · kubernetes/kubernetes The pluralization of rest resource names in URL paths is mostly human conceit that complicates every interaction from clients and leads to little practical benefit for end users. In addition, it's ...

Indeed it was a mistake. github.com/kubernetes/k...

If you've ever built Kubernetes integrations you'd know that "plural" might be the dumbest fucking thing perhaps in the entirety of Kubernetes.

Whispers: most companies don't run at Netflix scale and can very economically run their core business on serverless just fine. bsky.app/profile/theb...

All in all, what a wet fart of a product announcement. Less graphs, more use cases and *showing* the capabilities. Seems like they desperately need a Steve Jobs type person who actually has a vision because I'm not sure if Altman does. This problem doesn't seem to be specific to OpenAI though.

But I'm also guessing $200/mo is the only way this is even economically viable until compute gets wildly cheaper, so I'm not sure about the long term prospects of this stuff.

Unless these AI companies are just content with trying to sell this stuff to engineers, idk how this will ever break into normal consumers until someone has an "iPhone moment" but for GPTs.

Is this what happens when you have a bunch of PhD researchers trying to build an economically viable business? Like wtf is this. At least show a side-by-side comparison telling me why this is so much better than the other versions versus the alphabet soup of model names.

Introducing ChatGPT Pro Broadening usage of frontier AI

Oh sweet for $200/mo I can get access to GPT-4o with o1-mini Advanced Voice which also includes o1 pro mode! And there are graphs that show how much gooder this is for me. openai.com/index/introd...

If you're working on something really new, I don't think you really know what your idea is until you've built a prototype.

There's a critical tangibility to seeing some part of the vision made real. Accept no substitutes, and be skeptical of big claims without concrete demos.

Prompt engineering but the prompt goes to an API instead of a text box.

Lambda is only 10 years old?? What even is time

1. It's super easy to quickly make a thing that does stuff OK (this is like 10% of the work)

2. 90% of the work is the "refinement"

3. I would be hesitant if it was the core of my business and not just a bells-and-whistles feature or productivity enhancement

Probably these are well-understood problems to someone who is not AI-ignorant like me, but my experience led me to a few conclusions...

It's like describing a problem to someone who is writing code to solve it, except you don't get to see the code, only the output of the program, which is a major PITA to debug. And worse, minor changes to the description is like having them starting from scratch each time.

The other problem I saw was, again in very specific circumstances, I could not get the model to change a certain behavior no matter what I put in the prompt.

I'm curious how people are testing these things. For us, it's completely viable to have a battery of tests and then parse the output and check it because we're just generating YAML. But for anyone doing unstructured output, I'd imagine it would be difficult to test things reliably.

The biggest problem I see is idk how you handle model stability. Just switching from Gemini 1.5 Flash 001 to 002 resulted in completely different output in certain circumstances.

Which means the entire AI assistant we built is just a one-shot API call with no infrastructure whatsoever. HOWEVER our problem space is very neatly scoped which I think makes it a lot more tractable.