Hacker News Reader

Comment by spaceman_2020 | Hacker News Reader

spaceman_2020

is it me or ever since AI coding became the norm, there have been way more outages with otherwise reliable services?

I get downtime on Supabase every few weeks. Even Cloudflare. And now Github

Yes, because that caused the usage of the services to skyrocket, GitHub runs on Azure and Azure is experiencing capacity strain due to AI, so GitHub's services are struggling to auto-scale

Per a report that came out the other day, the GitHub move to Azure has been slowed down (i.e. I don't think it's done). But maybe you have newer/better info than me

GitHub’s data architecture (lots and lots of projects in a hierarchical distribution with immutable, highly transaction-based data elements) ought to be easily shardable to allow for massive scale-out. Especially a decade after acquisition by Microsoft. Even a major increase in activity (which was surely predicted internally years ago, given the marketing around AI) should be easily handled by a company with their resources. These repeated failures are indicative of massive mismanagement and misinvestment.

Microsoft has been letting GitHub operate independently since up to very recently. Seems to imply the problems could be systemically GitHub’s and Microsoft is now stepping in and correcting the ship.

GitHub’s instability started when they announced they were wholesale migrating to azure. They’ve been struggling with uptime since before the enormous surge in agentic coding they’re seeing. I don’t doubt that AI has massively increased their load, but u think this is more suspect than that.

I think they’re referring to services using AI for coding and shipping bugs more often as a result.

They are definitely more outages but the question is if these outages are due to the providers using LLMs to build there products and are therefore not delivering the quality they did before or have LLMs enabled a completely new user base to create projects which they deploy in the free tiers of named providers and they simply cannot keep up with the growth and the new influx of free users is skewing their mixed calculations (free vs paid) so heavily that they cannot scale without losing money. I'd probably say it's a mix of both.

Enterprise software and support services have consideraby (in my personal experience) degraded over the last year.

Less care on process, or quality, more focus on "just ship".

I've also seen and heard from peers this happening in multiple smaller outsorcing companies.

Correct. There’s no incentive to be careful anymore when you can just prompt an LLM to fix it

No but everyone is pretending that everything is fine. Actually, no, no one is pretending anything. No one cares, really.

Not just you, but uncertain whether it's due to unreviewed slop going to production, or increased demand due to slop generation.

> is it me or

No, of course not.

No, it's not just you. It is fairly obvious what's happening - the same old Entshittificators now have a great tool to up the speed of entshitification by 100x - thus these crappy outages every other day.

Reply on news.ycombinator.com