What happened in February 2026?

Time flies. February was a productive month, and I got many things done—with the help of my coding assistant , that is. I haven’t joined the “I haven’t written a single line of code for quite some time” flock just yet, but I am close. At the same time, I feel like I’m still locked in beginner mode. Like I have still to really tap into what agentic tools like Claude Code have to offer.

For now, the tools may have changed, but the journeys feel somewhat the same as they used to a few months back. It’s a strange feeling, and I am still trying to wrap my head around it.

Yet Another Claude Code Wrapper

In my previous retrospective, I mentioned CeCe and Vee. The former is a plugin for turning Anthropic’s model into a “modal” agent that I had vibe “prompted” with Claude itself. The latter was a second take at the exact same concept.

A month later, Vee has evolved quite a bit, to a point where the “modal” aspect of the project lost the front seat. More precisely, Vee has become a session multiplexer for Claude Code, with built-in support for spawning it in ephemeral containers with --dangerously-skip-permissions enabled. It is also a playground for me to experiment with the agent. For instance, I built a feedback system where I can point it to the good and bad things it’s doing—my feedbacks get saved in a SQLite database and appended in the system prompt during the next sessions. I’ve also tried to set up a knowledge base with embeddings. Both approaches have their benefits, but I don’t think I have clear evidence that they can be a game changer.

You may remember Vee started out of my frustration to losing control of CeCe’s internals. Well, Claude Code took over Vee’s codebase quite quickly as well. But this time, the output was a lot more satisfying. This time, I leaned into playing the role of a Tech Lead and a QA tester. Seeing the project taking shape and the tool I wanted becoming a reality that quickly has clear upsides. I also find it interesting that I have both a fairly good understanding of how the software works overall—I called most of the architectural choices—and very little insight on how the code itself is written. It reminds me of $WORK, where I am involved in some projects without being a direct IC.

I Built Myself a Cloudlab

In the first half of the month, a colleague of mine was tasked with building “staging environments à la demande” for the engineers working on $WORK’s Next Big Thing™. Their answer was a managed Kubernetes cluster and a bunch of Helm charts—and a small CLI tool to deploy the latter onto the former. A few hours later, I started building myself a k3s  cluster. It’ll be a fun way to get familiar with the stack, I thought. What emerged almost resembles a managed cluster of sorts. I called it elsa , and it now hosts this very website along with a few other self-hosted services I had under my radar for a while .

This may come as a surprise, considering I had migrated my website to a new setup called tinkerbell barely a month ago. Well, fret not. The little VPS sure has been retired early, but its legacy has not been lost. elsa was built upon the very same tools (Terraform, CoreOS, and Ignition) and promise I got hooked on a month ago: that there is freedom in designing a system whose destruction is a non-event. If anything, working on elsa has only deepened my understanding—and my appreciation—of what Fedora has been pushing for with CoreOS.

I really, really want to publish an article about elsa. I yet have to find my angle, though. Funny story, the first couple of paragraphs of this section were initially intended to be its introduction, until I grew dissatisfied with them and decided to repurpose them. Which means I need to restart from scratch. 😆

Anyway, as I was making good progress on elsa, I ran into some good opportunities to contribute to some open source software. I opened two bugfix PRs over the course of my journey: one for the official BetterStack’s Terraform provider , and the other for external-dns’ Vultr webhook sidecar . That was pretty cool! I want to do that more often.

We are almost a third into March already, and the month is clearly following the trend set by its predecessor. As such, I expect to have plenty to write about in a month’s time, meaning I should be able to keep up with publishing these retrospective pieces for a while. At least, I hope so!