rirze's comments

On: The Emacsification of Software

[rirze]

It was also the most prominent ecosystem for "software-for-one". Lots of custom-bespoke, "it works for me" packages that were only created for personal use.

On: My graduation cap runs Rust

[rirze]

A lot of graduation ceremonies have restrictions on how you can decorate your cap. I would be surprised if this was allowed.

On: Leaving GitHub for Forgejo

[rirze]

It's too decentralized for my taste IMO.

I like using radicle.xyz instead.

On: GitLab announces workforce reduction and end of their CREDIT values

[rirze]

While hosting internal services for 4 years, Gitlab was the only service that ran hybrid. Wish they could get their act together and focus on actual engineering again.

If anyone at Gitlab management is reading this; getting your microservices to run fully stateless in a Kubernetes cluster should the #1 goal. No disclaimers about potential risk. It's been 5+ years. Get it together. Stop bolting on minor package management features no one is going to end up using anyways.

[tapoxi → rirze]

They actually fixed the Gitaly thing a month or two ago

[rirze → tapoxi]

Thank god. I left that job 9 months ago so that's good to hear.

Sucks that it's `(Beta)`, but progress I guess.

Edit: Nvm, I found the asterisk; Full HA requires Gitlab to make Praefect Kubernetes-native and that is not done yet. [1]

[1]: https://about.gitlab.com/blog/gitaly-on-kubernetes-generally...

On: Vibe coding and agentic engineering are getting closer than I'd like

[rirze]

I didn't think it was sarcastic till I read your comment, upon which point, I got confused and read it twice to make sure it wasn't sarcastic.

Nevertheless, it is refreshing to see nuanced positive energy. I agree that AI is going to be the great multiplier.

On: Vibe coding and agentic engineering are getting closer than I'd like

[rirze]

This take is too premature. We forget that AI is seamless for contexts that are in the training datasets (popular programming languages, open source libraries, well-documented algorithms, etc..).

It is very obviously hallucinogenic when it comes to new programming languages, new domains, and uncommon/poorly documented contexts. And AI is very poor at (3D) spatial visualization (making AI assisted CAD development incredibly hard).

AI is not capable of genuine logical thinking from fundamentals yet; these are highly trained, curated models.

On: Show HN: Agent-skills-eval – Test whether Agent Skills improve outputs

[rirze]

It's hard to make hooks work here, since the default approach it's using is call the URL directly.

I think it's better to have a repo-level skill instead, titled something like "connecting_to_db.md" and demonstrate exactly how to connect. Codex has been pretty good at referring to skills but it depends on context at the end of the day.

On: Behavior-Oriented Concurrency for Python

[rirze]

It's researchers trying to contrive an example for why their research effort deserves a budget every year.

Of course their example will be detached from reality. But it doesn't matter.

On: Lessons for Agentic Coding: What should we do when code is cheap?

[rirze]

Yah, that comment is odd.

Sysadmins, Devops engineers will the be the last ones replaced by AI. The context window for their problems are huge.

Unless you define Sysadmins and Devops as fiddling with YAML all day, which might be the case here.

[faangguyindia → rirze]

>Sysadmins, Devops engineers will the be the last ones replaced by AI.

Most setups aren't properly documented which makes the discovery and exploitability part the major bottleneck when this is facilitated by AI, the sysadmin/devops team is downsized.

On: PyInfra 3.8.0

[rirze]

Yeah, you’re gonna eat your words when you do something that’s not “install this package” and “create this user”.

For anything dynamic and sufficiently complicated, ansible is horrible. Pyinfra is much better.

[vardalab → rirze]

I disagree. The rigidity of YAML and stuff like that is what actually makes LLMs work better. I have strict linting rules and file size limits and it imposes discipline on LLMs. That's why it worked even last year. Even before Opus 4.0 it worked to some extent as long as you imposed discipline on these models Trust me, I do pretty complicated things with Ansible, key thing is to have decent established patterns and these models truly are getting better.

[rirze → vardalab]

That's not my point at all.

When you have 6 stanzas to perform a dynamic if/else branch, the underlying system if flawed.

Models can overcome the complexity of ansible-- I argue that they shouldn't be. Ansible is a flawed framework.