Yeah, I had the exact same response after reading the post. I mean, I'm all for jumping on the Microsoft hate train, but not if it misses the elephant in the room. Let's say the _perfect_ GitHub replacement spawns tomorrow? What's preventing the same infrastructure challenges of millions of lines of AI-generated code destroying it?
I think centralized code hosting is pretty much going to get killed by AI. Just like it's doing to social media.
> I mean, I'm all for jumping on the Microsoft hate train, but not if it misses the elephant in the room.
That elephant didn’t even exist yet for the first few years of poor uptime shown in the graph in TFA… I don’t really disagree if we’re talking about the recent uptime issues, but how does that explain the years 2020-2023?
It doesn't. It just means if they were having problems before, they've now been made significantly worse by AI (on the free tier). All I'm saying is that the problem is bigger than, "Microsoft sucks."
>What's preventing the same infrastructure challenges of millions of lines of AI-generated code destroying it?
There's something called "rate limits" that engineers not working for GitHub have probably heard of; it's this crazy idea that you should limit the load on your infra in order to avoid downtime. GitHub is not the first free service to ever have to deal with bots.
> I think centralized code hosting is pretty much going to get killed by AI. Just like it's doing to social media.
Private corporate codebases are a poor fit for GH because they don't benefit from public social graph effects. And the typical codebase isn't so large as to be technically challenging to deal with with OSS tools. I'd guess they make up a substantial share of revenue.
But once the reliability is called into question, self-hosted or smaller alternatives start to look good. Although there's some trickiness there if you want to be super cautious about making sure you can get to your code+infra in case of a vendor incident, especially if you're cloud based.
Because if you were building GitHub from scratch today you wouldn't build it the same way and would benefit from many of the technological advancements of the last 2 decades (nearly).
of all the awful things AI is doing and will be doing to society, killing centralized code hosting and social media will be its shinniest moments, both deserve to die painful deaths
> How did people do it before github? Did everyone write everything with peek and poke?
I've been sharing GPL projects since 1999. We didn't need peek and poke (Both of which I have also used further in history...), but we managed nevertheless.
Prior to github I shared software on sourceforge (and others). Prior to that I published stuff on Freshmeat.
Prior to that I downloaded games others shared (not open source) on Happy Puppy.
Prior to that I used usenet to find and download games, shareware, etc.
Prior to that I used ftp to (IIRC) ftp.sunsite.edu, ftp.nic.fi, and others.
Prior to that I got news of new releases using Gopher.
Finally, prior to that, I actuallyy did use peek and poke to write software :-/
If github went away, and centralised repos went away, we'd still have something...
No, it was rhetorical. I remember downloading software from sourceforge, distro servers and getting drivers from random people's websites that needed to be compiled.
Why is centralized code hosting getting killed? I'm running an opensource project, >99% of the code is AI generated, could not do this without GitHub. Ai generated source code needs a place where AIs and people can collaborate. I'm expecting GitHub to be hugely successful, but mostly for an AI audience.
I'm sure the underlying infra is not a single server, so this is mostly a period where they have to adapt to higher loads due to AI becoming actually useable in the last 8 months. It's basically proof how well AI works these days. Give it a few months so they can scale and it'll get better. Remember Twitter fail whale? Growth pains that can and will be solved.
> It's basically proof how well AI works these days. Give it a few months so they can scale and it'll get better. Remember Twitter fail whale? Growth pains that can and will be solved.
GitHub's problems can technically be solved, but that doesn't mean they can be solved in a way where the economics still work out.
If AI use is 10x-ing the amount of infrastructure costs for GitHub but not 10x-ing the amount of money Microsoft brings in from GitHub then there is certainly no guarantee they will bother to solve these issues adequately.
And I'd be shocked if the revenue side of things isn't lagging way behind the extra usage post-AI-era, both because a lot of the new use is probably on the GitHub free tier, and because even on the paid tier most usage (other than CI/Actions, AFAIK) are on a fixed subscription cost per user regardless of how much you are slamming their servers and it is unclear how much they can raise that price without current enterprise users fleeing.
Twitter had a clearer goal that aligned with the financials... support more people stably, show more ads. Things are less clear with GitHub's business model where the free tier is a loss leader for the paid tier but the expansion in usage is likely to balloon the free tier usage at a far faster rate than the paid tier usage.
Also (and this part is admittedly far more speculative) if AI labs are to be believed this is still early days for AI usage and we'll still see massive usage growth over the next few years. If GitHub is already having existential trouble at the beginning of the curve, what hope do they have to scale up with their current business model if AI usage actually does ramp up exponentially?
> And I'd be shocked if the revenue side of things isn't lagging way behind the extra usage post-AI-era, both because a lot of the new use is probably on the GitHub free tier, and because even on the paid tier most usage (other than CI/Actions, AFAIK) are on a fixed subscription cost per user regardless of how much you are slamming their servers and it is unclear how much they can raise that price without current enterprise users fleeing.
I'd guess most of the costs incurred to GitHub outside of Actions as part of the enterprise flat-rate tier are a fraction of what enterprises are paying for AI in order to incur those costs in the first place.
If a company has to pay $5 extra to GitHub for every $100 of extra AI spend due to that AI use creating disproportionate load, I've got a hard time imaging that GitHub will be the thing that gets fled from.
As far as the free tier goes, it seems like there should be a path to making prohibitively-cost-incurring usage models high-friction. (e.g. limit the free Actions minutes that you get to a certain number per month.) As long as the limits are roughly proportional to the actual costs incurred, there's not too much risk of people fleeing to a competing service, because the only way a competing service would be able to undercut the costs is by taking steep losses themselves, which isn't much of a business model in order to attract people's code repositories.
Yah, the monitization bit is challanging. I'll ask my agent to click some of the ads GitHub serves it ;-)
But getting this infrastructure right is crucial for a future where most of the code is AI generated. GitHub puts microsoft in a good position to experiment and learn how to optimize GitHub (enterprise) for the future.
Nate b Jones on youtube, https://youtu.be/FDkvRl1RlT0?si=AEYlUchm_oalMSzf, argues that Atlassian might be an interesting acquisition for Anthropic, as it provide most of the context AI at enterprises need. When executed well, GitHub enterprise, can offer microsoft the same value: the context AI needs in the future.
> But getting this infrastructure right is crucial for a future where most of the code is AI generated.
That's not the problem. The revenue model they have is based on a certain amount of usage from the people who do not pay (you, for example), and a certain amount of usage from the people who do pay (enterprises).
If you 100x you usage, then they need 100x the infra, which means they need 100x the revenue.
At that sort of usage enterprises would rather self-host, and github would be left with only the free users, who are almost all like you now - hammering their servers but not paying for it.
If you self-host, for $5/m you can have your own VPS, but doesn't really solve the problem as much as you'd think - those are all vCPUs and shared, so you can't hammer them all the time either because then the provider has to increase their infra as well so fewer accounts share a single CPU.
Either way, if you want to generate code with AI at the speed that an agent can, you'll have to pay for it one way or another.
Also, one thing the numbers they published show is that the bits that are growing 10x YoY (and which they expect to get “worse”) are all the things that you get “unlimited” mileage off (even if you're a paying customer): repos, commits, PRs.
Things that have “usage based billing” (like action monites) grow closer to 2x YoY.
When there's a dollar amount attached, people don't 10x, because it's not worth it. They splurge when it's cheap, and unlimited.
Well either Microsoft finds a way, or Anthropic will. I'm sure they'd love to host all these projects with all the source and context. Maybe they should buy GitLab, or Atlassian.
> But getting this infrastructure right is crucial for a future where most of the code is AI generated.
If that is the future, then source code hosting will be the least of our worries. The entire industry will collapse because the software will stop working.
> Ai generated source code needs a place where AIs and people can collaborate. I'm expecting GitHub to be hugely successful, but mostly for an AI audience.
Are you paying them in proportion to the resources they expend on you?
There's this thing called "sustainability", and every company needs to have it. Github cannot continue on the current trajectory where every AI-bro wants to run an agent that generates 1000s of lines of code per hour, dozens of commits per hour... and provide that for free to a few dozens of millions of users who won't pay.
That being said, Microsoft does have an opportunity here - AI-bros are willing to pay $200/m to burn tokens so Github should offer a plan for Copilot, say $400/m, that includes a repo.
If they don't ban AI agents on free tiers, they are going to be out of business soon.
I think centralized code hosting is pretty much going to get killed by AI. Just like it's doing to social media.