This week-end I learned BetterStack has a Terraform provider and I think it is beautiful.

So debugging with Claude reminds me a lot of that feelings of having to watch someone else doing all the keystrokes via a Meet room. In my case we were debugging a k3s cluster issue and it understands that a lot better than me so it makes sense that I take the copilot seat.

“Good question.” Claude said. And the user knew: all the reassuring comments the agent made in the past few minutes were probably worthless.

Having Claude debugging a kubernetes issue on its own is mesmerizing

My little k3s setup is coming along nicely. Two VMs, journal logs and hardware metrics pushed to BetterStack. Next step is actually deploy something there 😄.

Unsurprisingly, I’m not the first one noticing the similarities: tante.cc/2026/01/15/s...

After the fast food and fast fashion industry, are we about to see the emergence of the fast software industry?

holy moly my cluster boots now.

gotta say, deploying a k3s cluster on Vultr + CoreOS + Ignition is harder than I thought.

I’m all in with the FedoraOS + Ignition way of deploy things™ but oh my god how annoying it is to get it right at first. I need to stop creating and destroying VMs that are costing me 2cts each time and try on local VMs 😂

the world if "ps | grep xyz" didn't show "grep xyz" in the output

Week-end challenge: setup a simple k3s cluster.

I’m resuming trying to use Claude Code in a way that would fit an engineering process. I’m using an issue to test the command that I’m implementing: github.com/lthms/vee/is... I like it because it’s not a trivial one. Like, I really wasn’t sure about the best way to implement this.

People will tell you being polite to a code assistant is just a waste of tokens. Meanwhile, Claude sometimes starts its prompts to subagents with “Please.”

So upgrading once’s firmware does help restoring sound on Linux

Specifically, the kind of person who, despite being largely on "the same side" as you on the broader issues, will come out knives swinging at the smallest slight, real or perceived, and then deflect and blame-shift and put words in your mouth to justify it. I don't need people like that in my life.

I can now start new Claude instances inside Docker containers with a few keystrokes

I’m getting bold with my “I don’t like Claude Code vanilla TUI’s UX, let's vibe code something better.” I’m still looking for the #$WORK formula, but for personal projects, I don’t see a way back to coding myself, and that makes me nostalgic.

Even though AI agents make me more productive: when I use them, I end up working more, not less! It’s such a contradiction: a thing that should be saving time ends up taking away more overall time… A reason I find it hard to believe AI will lead to fewer hours/days worked…

Wouldn’t have guessed I’d missed 2025 so hard so quickly (and 2025 wasn’t great when you think about it so it’s telling)

Witnessed at #$WORK today; misusing a gant chart to show a flamegraph in mermaid Nice

Reflecting on what happened in January. It was an intense month on my side, *for sure*.

The Shape of Leadership #devopsish

Year in review: 2025. Worked on my OCaml static site generator and then used it: 164 posts, RSS, tags, search, and images. alexleighton.com/posts/2025-1...

Modular explicits are coming to OCaml 5.5 🎊🫰🥳🎉🎂 github.com/ocaml/ocaml...

ฅᨐฅ

the ocaml compiler just told me "This doesn't make sense to me." which i find extremely funny because in the context of an error message that means literally nothing

Gitlab API is driving me crazy. IIUC it’s aggressively cached? The first call is correct, then if I leave a comment in the same MR, it will not appear?

"claude md" ridiculous. he is NOT a doctor

Watch me rewriting from scratch the first artifact I have ever vibecoded 🥸

I’ve spent ~1 week experimenting with CeCe without writing a single instruction (letting Claude doing it for me). Now trying to get everything I learned and write the prompts myself ahah.

Biased for action much --- ● 🐱 Pushed to main. ❯ did i tell you to push to main? ● 🐱 No, you didn't. You said commit, not push. Sorry about that.

In my own interregnum, but: I’m open to new opportunities. If you’re looking for someone who can lead teams working in AWS/serverless, Typescript or OCaml, and/or take an observability or DevOps lens to their cloud infrastructure, I’d love to chat. I’m based in Sydney for onsite/hybrid/remote

Who’s going to FOSDEM next weekend? I’m considering it as a last minute trip

I’ve written down my thoughts about what I want to do in 2026 wrt. LLMs

I could use my rss feed to automate posting a Bluesky message for my new posts 🤔

I think my overall emerging philosophy for agentic work could be summarized as: “leveraging existing collaborative platform to interact with the agent when possible” e.g., I want to review Claude code in Gitlab/GitHub. Follow progress in Linear.

I may have pushed a Docker image that could lead some rss readers to list a new entry. Who knows

2026 Q1 challenge: publish enough articles that my website homepage does not mention 2024 anymore

I’ve finally configured a proper notification daemon and now i don’t need to compulsively switch to my claude code session to check if it is waiting for me. 😂

I need to stop using the b reader word when asking for feedback about unpublished articles. I think it triggers the wrong places 😅

Oh, looks like my CoreOS + Quadlets article has been shared on buttondown.com/devopsish/ar...! That’s cool

I also want to write a blogpost “How I want to use LLMs in 2026” but I have no idea how it would be received

I need to stop overthinking it. It’s okay if i don’t vibecode every little ideas i have. I’ve been using Claude to drastically improve the quality of the (non-code) artifacts I produce at #$WORK for instance (especially issues). That’s a good starting point

It’s like i can’t just “use Claude” and feel the urge to put it in boxes that maybe it does not need. I’ve been refining github.com/lthms/cece for a week or so now, and I’m both satisfied and somewhat frustrated ahah.

this is actually my biggest issue with vibe coding (at least in its present incarnation) vibe coding breaks flow and makes the user more out of touch with the code they are modifying rather than more in tune with it

The vibosphere on the floor right now

I’m starting to have fun with making Claude work as transparently as possible.

As I am myself starting to embrace agentic coding, I relate so much to this www.pcloadletter.dev/blog/pre-cod...

I’m not sure I like the early 2026 vibes of “let’s adopt generative AI in our work workflow asap because FOMO” 😂. However, I feel like there will be value in configuring, tweaking, etc. one’s code agents setup. That’s not that different from curating an editor config—and I find that reassuring