Arrested DevOps · Software Engineering & Infra · November 2026
Host Matt Stratton had asked how engineering and security can enable non-engineers (marketing, product, CEOs) to ship software with AI without losing control of production. Stephan offers the phone's app-permission model as a mental model he uses with clients.
…even if I forget it, everything is okay. So there's always just like a back door of everything going to be okay, because that's already been thought through. Like all the worst case scenarios have been thought through. So this thing is really bulletproof with me being able to push and do things.
…And we in engineering have not really thought about like how you market that and share that and communicate that outside of your department. So what are the ways to solve for that? Or what have you kind of seen about doing that? Because I imagine that's a key part of unblocking that bottleneck.
Yeah. One thing, one mental model that helped me a lot and helps clients a lot in discussing this is basically we already, the company already solved this with your phone. So you on your phone, you run apps from people you don't know and they are next to your banking data and no one is worried. So why are you allowing people to run stuff on your phone, but not let people run stuff on your production system? And if you look into this, there are some attributes to the phone that can. Carry over to normal development, which is, for example, app permission. If you usually, if someone writes some code, it can do everything. It can delete the orders, it can delete customers, it can read all the customers, it can do everything. If you model it with permissions, then Tanya could write something and it could be deployed, but her app or her module or her package does not have the rights to read all the customers. And then there's no need to review the code if it does this. You just need to review the permissions. You're just like the phone pops up. Can I read…
Yeah, I think it keeps going back to that these are solved problems, but the scope of where they have to be extended to, you know, because we think about things like safety two. And the idea is if we look at a blameless and a safety two concept is when something bad happens, it's not what did Tanya…
That was like four questions all in one. Yes, it was. Okay. We're going to start with question one, which is, let's go back actually. Can you repeat to me, Matt? Yeah, the first one was too many. What are
some things you understand more now about engineering and building systems and stuff that just you didn't sit down and read a book, you know? No, I didn't