Are Coding Agents Changing Where Our Code Lives?

So OpenAI built its own Git backend too. Pando took 10 weeks with Codex, and OpenAI says median pushes are nearly 5x faster. After Anthropic built Scion, I’m starting to wonder how much of our development setup survives a company full of coding agents. The reason was pretty practical. Codex meant more branches, more uploads and more downloads happening at once. OpenAI says reliability sometimes fell below 90%. Imagine hiring faster and faster coworkers until the place you share code starts struggling. One engineer designed Pando’s core. Teams of agents implemented it, and a handful of engineers deployed it. How long would your team budget for that? I’d want to hear as much about how they checked it as how quickly they wrote it. It’s been serving production traffic since April and keeps the familiar Git workflow. OpenAI is contributing client improvements back to Git. Your everyday commands can survive even when the machinery answering them changes completely. Anthropic went further. Seven people built Scion, its review tool and related systems, then migrated the monorepo in five months. Scion replaces Git there. Anthropic also says its scale and internal token budget make this a poor fit for most teams. GitHub is rebuilding its Git infrastructure for agents too. I’d be careful about predicting its death while it’s adapting to the same pressure. It could absorb a lot of this change before most companies ever need to think about moving. Then there’s Cursor’s Origin, in early beta, and Amp’s Git hosting. Both keep Git. But if your coding tool also offers a home for your code, how often will teams choose a separate service? Git and GitHub are different things. Git tracks the history of your code. GitHub hosts it and adds reviews, automation and collaboration. You can keep Git and change hosts. OpenAI shows you can also keep the workflow and rebuild what serves it. My guess is Git survives longer than some of the habits around it. Familiar commands are useful. But a process built around people submitting changes occasionally may need work when agents are submitting them all day. For your company, I’d look at where finished work waits. Is it review? Tests? Permission to release? A faster coding agent can leave you with a bigger pile of changes nobody has time to understand. Would faster pushes actually help your team today? I can also see the temptation to build more tools yourself. Ten weeks makes a custom Git backend sound almost reasonable. But who maintains it next year? Are we removing a recurring problem or signing up to run another service? For your own career, there’s a lot in that single engineer designing Pando’s core. I’d read it as a reason to get better at understanding systems, making tradeoffs and checking the result. Someone has to know what must stay true while agents change the code. I’d like source control to make the next questions easier: Why did this change happen? What proves it works? Who approved it? Can we undo it safely? If code gets easier to produce, how do we make it easier for the next person to trust?