Transcription
GitHub took down lovable.dev today, and I want to talk about it because, at the end of the day, we are in a situation where that kind of thing is going to get much, much worse in 2025.
I'll explain the competing exponential growth curves that we're on here and how this all broke down.
So, overnight in the US, lovable.dev suddenly was unable to continue to create GitHub repos. This quickly snowballed into a major outage because, at the end of the day, their exponential growth has been so fast. Lovable.dev has been creating GitHub repos at the rate of one every two seconds—absolutely insane speed. They said that they had previously checked before the holidays with GitHub and asked, "Look, we're growing really fast. We're depending on GitHub to be up. Are you good? Are we going to hit quotas or rate limits?" GitHub told them no.
Well, no one is quite saying what happened. I've dug in, and GitHub won't say. Lovable is being polite about it all. What is actually occurring is that there's a terms of service violation that lovable ran into sometime during the night hours on January 2nd in the US. GitHub was all asleep, and it's the holiday, so you're really waking up a developer who's somewhere skiing at this point. No one was on call, really, and certainly, they weren't responsive because lovable was hard down for hours overnight—a major client of theirs hard down for hours.
They said they tried and tried and tried to wake up lovable and got nobody. They attempted to build a workaround on Amazon S3 during the night to keep people up because they're a European startup. It's like the first working day of the year in Europe, and people are coming in with their New Year's resolutions, wanting to build stuff, but it's hard down. So, they're desperately scrambling to get stuff over to S3, building a workaround. You can read on X the different updates that come through, and it's always like more bad news until they finally got GitHub to wake up this morning. GitHub took them back out of jail and allowed them to start creating repositories again.
Now, reading between the lines, I strongly suspect that one of the takeaways for lovable is that they can't be that dependent on GitHub. They probably will eventually move to Amazon S3 as a way to start to scale this, which is not exactly ideal. One of the things that's really nice about GitHub is that everyone knows what it is. Everyone goes over to GitHub and knows what you've done, and there's just not a good substitute for that social quality that GitHub brings to code.
So, it's sort of sad to see a situation where you have an AI tool trying to build on top of GitHub, trying to reinforce the GitHub flywheel, and GitHub just isn't set up for the volume. This is something that I want to call out because I think there are two exponential curves that are happening in 2025 that are stacking on each other, and they are going to make life even worse for providers like GitHub.
If you are in the infrastructure business that has anything to do with development at all, brace yourself. Do an architectural refactor now. The reason why is that, over the last 20 years, your entire business model has been engineers committing code at the speed that humans can write code. Not anymore.
Now, in 2024, you've had a massive 10x explosion in the number of people who are interested in coding because they can code with LLMs. So, there's a bunch of AI-driven code being committed to repositories like GitHub, and that's about to get even worse. Now, stack on top of that the massive explosion in humans using AI to code. Soon, agents will be using AI to code because we're going to have autonomous agents coding within the next couple of months.
So now, with 10x to 10x, you're going to get even more code being committed. Look, I am not saying that this is all high-quality code; we're not talking about that here. We're just saying that, from a sheer volume perspective, if you are in tech, you should be doing a refactor now. You should be looking at your architecture and examining the ways people use your systems. Ask yourself: if these people suddenly get access to AI tooling that enables them to be much more productive, how will their usage of my system change?
Let me give you an example that's not even GitHub. Let's say you're running a nice little SaaS business for marketers. Imagine your marketer suddenly gets a hold of Project Mariner and a browser that's agentic. They're like, "Oh, thank God! I don't have to go and get the freaking report downloaded off of this tool. I can just have the agent do it for me every morning."
Now they feel free, so they're going to code up five or six different agents, give them different tasks, and they'll be all over your system. They will be using this person's login with or without your knowledge, and suddenly, your usage patterns are going to completely change. That's a light version of what could happen. Imagine the consequences for security and authentication.
So, we have a lot to work on when it comes to agentic. I think the GitHub example is an early 2025 warning of how bad it could get if we are not careful. I'm glad lovable is back up, but unfortunately, because of the fundamentals in the space right now, I expect more of this to happen in the coming year.