GitHub’s AI load problem is coming for your platform team

Ajouté : 2026-05-16

[00:00:00]Coté: AI-driven app deploys have driven GitHub traffic up to about 3.5x over the past two years.

[00:00:05]That sounds modest, But as Gergley wrote up over at Pragmatic Engineer, it doesn't seem like they were expecting this.

[00:00:10]To address this, they started a plan in October to handle 10x that traffic, and by February they'd revised it to 30x.

[00:00:16]Their reliability is now sitting around at 86%, probably driving some of that complaining you're seeing.

[00:00:22]Now, this is a huge company.

[00:00:23]It's Microsoft with thousands of engineers probably devoted to just this kind of problem, and they're struggling with the onslaught of AI-generated apps and code.

[00:00:32]Their reliability is now sitting around at 86%, probably driving some of that complaining you're seeing.

[00:00:37]If you run a platform team at a bank, insurer, whatever type of enterprise you might be in, the question that you should be concerned with is what your version of that curve looks like.

[00:00:45]As you introduce AI coding into your organization, do you have the capacity to handle that kind of increase so quickly in app delivery?

[00:00:51]If you don't, do you have the budget to build that out?

[00:00:53]Now, AI-generated code is obviously an oncoming flood because it works and people are hungry for it, especially on the business side.

[00:01:01]And we know from the cloud shadow IT decade that if the platform, the infrastructure isn't ready, developers with the business's support are gonna find what they need, usually somewhere out of policy and insecure.

[00:01:12]people in the ops world also know that the ops people are gonna be the ones left over to handle it a year or two later, following that motto, "not my mistake, now my problem," and all that.

[00:01:22]My friend Cora wrote a piece this week making a related point.

[00:01:25]The platforms that already did the boring integration work, service catalogs, credential rotation, multi-tenancy, have all been running quietly but successfully in large organizations.

[00:01:35]Those are the ones that are well-positioned to deal with what's coming because they're used to this kind of scale and they're already installed.

[00:01:40]Now is not the moment to start building your own platform or start with a new platform.

[00:01:44]You need to focus on handling this coming wave of AI-generated app releases and providing the services that they need for data and authorization, just the basics of delivering apps and keeping them up.

[00:01:54]And then you probably also need to figure out hosting access to the models and inference, sorting through that, dealing with the performance of it.

[00:02:01]That's a lot to do with, and tinkering with a platform for just the basics that hasn't been hardened by years of enterprise use isn't gonna be that much fun.

[00:02:08]So go check out Cora's full piece.

[00:02:10]There's a punch list and more details about what you should be looking for here.

[00:02:13]You can find the link below or you can find it linked off of TryTanzu.ai.