Akatama's Slice of the Internet

SCaLE 23x: The Parts That Don't Make It onto YouTube

By Jimmy Lindsey

March 18, 2026 | Categories: devops, open-source, conference, Linux

A couple weeks ago, I spent four days at the 23rd edition of the Southern California Linux Expo (SCaLE) surrounded by immigration lawyers, quantum computing researchers, veterinary technicians, and the occasional software engineer, all united by a love of open source and Linux. Most tech conferences feel like someone is trying to sell you something, but SCaLE is completely different. Instead, it's community-run, most of the booths are staffed by volunteers, and the people running them are just as likely to want to talk to you about your interests as their own. This is less a "what I learned" post and more a "what I did" one, but the two turned out to be harder to separate than I expected.

A Short History Lesson About My Experience with SCaLE

All the way back in SCaLE 21x, just a few weeks before my second daughter was born, I met Drew Adams. I went to a panel that talked about how to contribute to open source. Drew talked extensively about contributing to KDE, which is my go-to desktop environment for Linux. Right after the panel, I went to talk with him about how to get started. At the time, I didn't imagine that Drew would essentially "adopt" me into this world of open source.

I don't want to go too much into the details of Drew and the other members of his crew, mostly for their privacy. Essentially, Drew founded a sub-organization of openSUSE called SoCalSUSE. When I went to talk to him, he eventually brought me to his booths at SCaLE. Yes, there are multiple. SoCalSUSE handles the openSUSE, KDE and NextCloud booths. After talking to him a bit at my first SCaLE, he had thoroughly convinced me to try out openSUSE Tumbleweed, a rolling release distro I still use to this day.

Last year, at SCaLE 22x, I met Avni and Adam from Internet-in-a-Box. Similar to meeting Drew a year before, this would also be very fortuitous. I ended up becoming a major contributor to Internet-in-a-Box, and a large part of my success in the last year has been due to things I learned working on that project.

While Drew, SoCalSUSE, Avni, Adam and Internet-in-a-Box's other contributors are all amazing people, they are not the only ones. Pretty much everyone at SCaLE is excited to share their passions and listen to yours. SCaLE is much better if you do it with friends, but the nice thing is that you will not have a hard time making friends if you listen to people and share your passions with them in turn.

This year, SoCalSUSE invited me to stay with them at their Airbnb. Not only did I have more fun, it also ended up being much cheaper (thank you openSUSE Travel Support Program)!

Working at the Booth

I helped set up the booth on Thursday, then spent almost all of Saturday and all of Sunday helping out. You may think that spending so much time at a conference without going to any talks is a waste of time, but in actuality it may be more worth your time. First of all, all of the talks at SCaLE are streamed to YouTube and eventually they are even edited and uploaded. Second of all, you meet a lot of people by running booths like this, and since the people you talk to have at least some interest in your booth, you likely have something in common.

Other than all the new people I met, it is also a great way to get closer to people you have met before. Especially early in the mornings, or when a big talk was running, the booths would often be empty. That made it a great time to talk to those who were currently working at the booth. Given the topic of the conference, you might think everyone who worked at the booth was a software engineer by trade, but this wasn't the case at all. Besides me, there was a former manager/mechanic, a non-software quantum computing researcher, an immigration lawyer, a Registered Vet Tech, a former system administrator for a school district, a stay at home mom and more. It is fascinating to learn what makes such a diverse array of people so interested in Linux and open source to devote their time to running the booths at SCaLE.

Internet-in-a-Box also had a booth there, so there was quite a bit of time I got to spend with Adam and Avni. Unfortunately, I was not aware they would have a booth at SCaLE until long after I had already committed to help out with SoCalSUSE. Internet-in-a-Box also had a talk that I went to, but I was already aware of most of the details since I contribute, so I went to show support. I also got to have a delicious dinner with them on Sunday night after SCaLE had ended.

Talks

Most of the talks I went to were part of DevOpsDay LA, which was the main reason I went to SCaLE originally. This year, it definitely wasn't as good as it has been in the past. Three or four of the talks pretty much broke down to this:

Replace (A) with your topic and X, Y and Z with things humans still need to do, most of which are the same no matter the topic. That's not to say that even in these talks there was nothing to learn, but it did get kind of tiring hearing it over and over again.

Overall, I think LLMs are really interesting and I want to discover new ways to use them. Yet, a one-hour talk has to spend so much time setting up the problem, that when it comes to how the LLM is used to solve that problem, it ends up being shallow. The best way to learn about LLMs is to use them. The next best way is to read about how people use them and try it out yourself. A talk just gets in the way of that loop.

These Are NOT the Vulnerabilities You are Looking For: Hiding Vulnerabilities in Containers

The core message here is that vulnerability scanners are just a tool. If someone tampers with them enough, they stop doing their job. Kyle Quest demonstrated that by obfuscating OS and package information in a container image, you can fool scanners like grype or trivy into missing real vulnerabilities. The scarier implication is that malicious actors could hide this in a base image before it ever reaches you. I'm not losing sleep over it since I stick to trusted images, but it's a good reminder that your scanner is only as reliable as the environment it's scanning. Kyle also made it very easy to recreate at home if you want to see it for yourself.

Build a Better Loop: A Guide to Platform Engineering

The biggest takeaway here is that we're too focused on the specific tools we use to deploy and create applications, when the goal underneath those tools has never changed. AI is the next wave, and the right move is to build systems that make it easier for AI to operate, not to perfect what you already have. As the talk put it: "When cloud came along, the right move wasn't to perfect your VMWare infrastructure."

What I found most interesting was the framing around how your customer is changing. Today it's developers using AI as a tool. In the near future, it will be AI agents as direct consumers of your platform. That shift has real implications for how you build. The talk identified three assumptions worth challenging:

The practical upshot: audit your tests so an agent can verify correctness without a human, audit your legibility so an agent can answer "who owns this and who depends on it" from a service name alone, and map your value stream around decisions rather than code.

Why Your Kubernetes Cluster Will Fail: Lessons from 1 Million Real-World Incidents

I'll believe a well-built agentic system can handle simple Kubernetes problems quickly. Whether it holds up on genuinely complex incidents, I'm more skeptical, but the approach Mickael Alliel's team took is worth understanding. After analyzing a million real-world incidents, they found failures clustered into six categories:

Their first instinct was to build a big runbook. That failed, because production environments are messier than any runbook can anticipate and the same error rarely has the same root cause twice. The solution they landed on was a multi-agent AI SRE called Klaudia, which has narrow agents with deep domain expertise in specific areas, grounded in the organization's own knowledge base and architecture. They claim a 99.2% success rate. I agree that if you're going to build something to wrangle Kubernetes, a system of focused agents is probably your best bet. Whether those numbers hold in the wild is another question.

Conclusion

The irony of SCaLE is that the parts hardest to justify on a calendar, such as an extra hour at the booth during a slow stretch, are the parts that make it worth going back to. The talks are streamed and uploaded to YouTube. The technical details will be there whenever I want them. What you can't get from a recording is the camaraderie of hurting feet from standing at a booth all day, or a game night that runs way too late. I already miss it. See you at 24x.