Reliability & operations · Take
The AI argument at work is really about who's on call
The fight over AI-written code inside a company comes down to who gets paged, and the old 'you build it, you run it' rule would make the argument more honest pretty quickly.
See, the problem is that neither side is making it up. They are seeing really scary trends. They’re grappling with real hard problems that are getting worse.
If both sides are right, how can we ever hope to resolve the argument? In June, Charity Majors wrote AI enthusiasts are in a race against time, AI skeptics are in a race against entropy. The enthusiasts don’t want to get left behind. The skeptics see the costs of quality and reliability piling up.
One of the things that’s ironic though is that both of these sides feel like they are the tiny minority and they’re outmanned and they’re being suppressed and they are standing up for what is truth and valor in the face of the big AI folks or the big skeptics.
They both think of themselves as rebels, and somebody isn’t doing a good job of identifying who the empire is. I think a better way to divide people might not be by how optimistic they are, but by who gets paged when there are alerts.
This often starts to come down to the group that is on call and the group that is not.
The people who the buck stops with them, they are seeing melting mental models, they’re seeing slop, they’re seeing all their hard work just dissolve and they don’t see any end in sight.
It’s not surprising that there are skeptics. They get the 3 a.m. side of the productivity gain.
It’s always weird to me just how much software engineers really seem to believe that the world exists in the repo. It doesn’t. It’s production, you know?
The enthusiast often measures the repo (how many PRs got merged, how fast they’re landing). The skeptic measures production, where the cost of that speed eventually shows up.
If you’re shipping code written by an agent, I think you should be carrying the pager for it. It’s “you build it, you run it” for agents, and it makes the discussion honest pretty quickly.. the person counting merged pull requests ends up being the one reading the alerts.
Imagine an organization where a separate ops group carries the pager for all application teams. The app teams start using agents, merging code faster, and the ops team absorbs the consequences. There’s a long conversation about this dynamic on The Pragmatic Engineer. But who’s on call for the code that nobody read?