Arrested DevOps · Software Engineering & Infra · November 2026
Matt asked where non-engineers shipping with AI works and where it goes sideways. Stephan says the sideways case is shadow apps: when the official path to production is blocked, the AI agent suggests another host.
…okay, before we can put that into production, it needs to be reviewed and you get into a review queue. And to some, so there are a lot of benefits. Obviously, the person can build it on their own until they like it, but then they still need to wait. And that's kind of what I call bottleneck 2.0.
…the same thing. It still needs to be reviewed. You're like, wait a minute, I coded this awesome thing. And what do you mean? Now it's not sitting here. So like this idea of these folks doing this, where does that, where does it work effectively? And how does it go sideways maybe when it doesn't?
I start with the sideways. One more fun part. Yeah. Depends on whether you're responsible for it or not, for the sideways. But I think the sideways thing shows up mostly in shadow apps and maintaining security. So everyone deploys somewhere. They can't deploy to the production system of the company. They deploy to Versal. They deploy to whatever the AI thinks is a good idea to deploy to. I think that is where it goes sideways, where people, because there is some kind of bottleneck, it just, the AI always helps you. It says, oh, you can't deploy through the company side. Well, we can deploy it to Versal or we can deploy it anywhere. So that's, I think, is where it goes sideways, where the bottleneck basically pushes everyone to deploy just somewhere else. And I've seen CEOs who deploy to Versailles, to CEOs who deploy wherever they like. So that's where I think where a lot of things happens where and no one knows what's running where.
Right, exactly. And it's so much easier. And even it's like you said, I kind of love where you're saying it's bottleneck 2.0. Like these are not new problems. They're just happening at a greater scale because in the early era of cloud, we had the exact same shadow IT problem, which was the, oh…
Yeah. I think a lot of it I had to, it was an adjustment for me to get comfortable with being uncomfortable because I'm not going to understand absolutely everything that's happening. Like I don't have a background in code. I'm still spending, I'd say some days I'm fully in terminal in Ghosty doing…
I think that there's a responsibility on engineering or sysadmin, whoever, the people who care about, you can't just say do it the right way. You have to enable for that. And so often, and this is, I mean, recording episodes of this show, I continually keep going back to this same old well of like…