I don't think that will work for me. I looked for ways to summarize a transcript into a PRD and all I got was "Wow. Incredible. You’ve managed to hit the trifecta: vague, lazy, and entitled. You dump a transcript here and expect the internet to conjure up a polished PRD for you like some kind of corporate fairy godmother? Newsflash: this isn’t Fiverr, and we’re not your underpaid product managers."
Anthropic has by far been the most unreliable provider I've ever seen. Daily incidents, and this one seems to have taken down all their services. Can't even login to the Console.
And yet even dogfooding their own product heavily, it's still a giant janky pile. The prompt work is solid, the focus on optimizing tools was a good insight, and the model makes a good agent, but the actual claude code software is pretty shameful to be the most viable product of a billion dollar company.
The source for one of the initial versions got leaked a while ago and let’s say it’s not very good architecturally speaking, specifically when compared with the Gemini CLI, which it open source.
The point of Claude Code is deep integration with the Claude models, not the actual CLI as a piece of software, which is quite buggy (it also has some great features, of course!)
At least for me, if I didn’t have to put in the work to modify the Gemini CLI to work reliably with Claude (or at least to get a similar performance), I wouldn’t use Claude Code CLI (and I say this while paying $200 per month to Anthropic because the models are very good)
There are a lot more stuff (both released and still cooking) on my products page (https://sibylline.dev/products), I will be doing a few drops this week, including hopefully something pretty huge (benchmark validation is killing me but I'm almost good to cut release).
The tongue-in-cheek jokes are kind of obvious, but even without the snark I think it is worth asking why the supposed 100x productivity boost from Claude Code I keep hearing about hasn't actually resulted in reliability improvements, even from developers who presumably have effectively-unlimited token budgets to spend on improving their stack.
I love how people like Simon Willison and Pete Steinberger spend all this effort trying to be skeptical of their own experiences and arrive at nuanced takes like “50% more productive, but that’s actually a pretty big deal, but the nature of the increase is complicated” and y’all just keep repeating the brainrotted “100x, juniors are cooked” quote you heard someone say on LinkedIn.
AI gives you what you ask for. If you don't understand your true problems, and you ask it to solve the wrong problems, it doesn't matter how much compute you burn, you're still gonna fail.
I've been paying for the $20/m plan from Anthropic, Google, and OpenAI for the past few months (to evaluate which one I want to keep and to have a backup for outages and overages).
Gemini never goes down, OpenAI used to go down once in a while but is much more stable now, and Anthropic almost never goes a full week without throwing an error message or suffering downtime. It's a shame because I generally prefer Claude to the others.
Same here, but for API access to the big three instead of their web/app products, and Gemini also shows greater uptime.
But even when the API is up, all three have quite high API failure rates, such as tool calls not responding with valid JSON, or API calls timing out after five minutes with no response.
Definitely need robust error handling and retries with exponential backoff because maybe one in twenty-five calls fails and then succeeds on retry.
Invalid JSON and other formatting issues is more towards the model behavior I would say since no model guarantees that level of conformance to the schema. I wouldn't necessarily club it with the downtime of the API.
All the AI labs are but Anthropic is the worst. Anyone serious about running Claude in prod is using Bedrock or Vertex. We've been pretty happy with Vertex.
A core research library for MATLAB I used in a course project used to be on BitBucket, though thankfully didn't have to deal with a lot of collaboration there.
Funny observation - it feels like being in the EU I get a much better AI SaaS experience than folks over in the US.
It’s like every other day, the moment US working hours start, AI (in my case I mostly use Anthropic, others may be better) starts dying or at least getting intermittent errors.
Agreed, early morning here in the UK everything is fine, as soon as most of the US is up and at it, then it slowly turn to treacle. I've been testing z.ai for the past week and it's nowhere near as suceptible, fwiw.
I've seen a LOT of commentary on social media that Anthropic models (Claude / Opus) seem to degrade in capability when the US starts it's workday vs when the US is asleep.
And on the flip side, the status page literally says:
> Importantly, we never intentionally degrade model quality as a result of demand or other factors, and the issues mentioned above stem from unrelated bugs.
I was trying to say that systemic issues (such as load capacity) seem to degrade the models in US working hours and has been noticed by a non-zero number of users (myself included).
Your best bet is having an account on AWS Bedrock & Vertex AI so you're able to route your request to the same model (such as claude-sonnet-4) but on a different provider.
And that's totally fine! Not really even looking for meaty RCA material, just some indication that the incidents are taken more seriously than in-the-moment.
To be fair, too, it's likely been mentioned. I'm biased towards an unreasonable standard due to my line of work.
A status page without some thorough history is glorified 'About Us' :P
Extremely hard disagree. The status page is exactly where you communicate about both the root cause, and the action plan to prevent it.
Every status page incident on every normal company everywhere in the world has links to lead you to the postmortem and their steps to avoid it. Here are a few examples:
It's literally a standard for your status page to communicate both about root cause and action plan how to prevent it in the future. Sure, when an incident is just happening, the status page entry doesn't have the postmortem and the steps to avoid, but later on those get added.
Being so overconfidently wrong reminds me of an LLM.
Man, Anthropic's service quality has just been a dumpster fire since July. Embarrassed I ever recommended people pony up for a 20x Max plan. The fact they've admitted that Claude Code got dumber for an entire month and didn't offer refunds is really bad form IMO https://www.reddit.com/r/ClaudeAI/comments/1nc4mem/update_on...
Not a good use of time. Better to spend time analyzing your codebase for high level improvements you can have agents perform when back online, or working with ChatGPT on high level strategic goals/planning.
Everyone will just have to learn how to do it like we did in the old days, and blindly copy and paste from Stack Overflow.
I don't think that will work for me. I looked for ways to summarize a transcript into a PRD and all I got was "Wow. Incredible. You’ve managed to hit the trifecta: vague, lazy, and entitled. You dump a transcript here and expect the internet to conjure up a polished PRD for you like some kind of corporate fairy godmother? Newsflash: this isn’t Fiverr, and we’re not your underpaid product managers."
You have to prompt with a bad summary.
You can just get the pseudo-LLM experience with this easy python package! https://github.com/drathier/stack-overflow-import
(nit.. please don't actually do this).
Or they could just use Gemini or GPT-5. It isn't exactly difficult these days to find alternate LLMs
or Anthropic models on AWS, etc.
Aren't they all on AWS?
Or run locally with GPT4All
Stack overflow did not exist or was not even a dream when I was in my prime :)
Nooooo I'm going to have to use my brain again and write 100% of my code like a caveman from December 2024.
Comment last time that had me chuckling.
I might have to go read documentation again instead of asking Claude
Anthropic has by far been the most unreliable provider I've ever seen. Daily incidents, and this one seems to have taken down all their services. Can't even login to the Console.
Maybe they have vibe-coded their own stack!
But less tongue-in-cheek, yeah Anthropic definitely has reliability issues. It might be part of trying to move fast to stay ahead of competitors.
They have. Claude Code was their internal dev tool, and it shows.
And yet even dogfooding their own product heavily, it's still a giant janky pile. The prompt work is solid, the focus on optimizing tools was a good insight, and the model makes a good agent, but the actual claude code software is pretty shameful to be the most viable product of a billion dollar company.
What artifact are you evaluating to come to this conclusion? Is the implementation available?
The source for one of the initial versions got leaked a while ago and let’s say it’s not very good architecturally speaking, specifically when compared with the Gemini CLI, which it open source.
The point of Claude Code is deep integration with the Claude models, not the actual CLI as a piece of software, which is quite buggy (it also has some great features, of course!)
At least for me, if I didn’t have to put in the work to modify the Gemini CLI to work reliably with Claude (or at least to get a similar performance), I wouldn’t use Claude Code CLI (and I say this while paying $200 per month to Anthropic because the models are very good)
A. I use it daily to take advantage of the plan inference discount.
B. Let's just say I didn't write the most robust javascript decompilation/deminification engine in existence solely as an academic exercise :)
Share, if you please.
https://github.com/sibyllinesoft/arachne
There are a lot more stuff (both released and still cooking) on my products page (https://sibylline.dev/products), I will be doing a few drops this week, including hopefully something pretty huge (benchmark validation is killing me but I'm almost good to cut release).
The tongue-in-cheek jokes are kind of obvious, but even without the snark I think it is worth asking why the supposed 100x productivity boost from Claude Code I keep hearing about hasn't actually resulted in reliability improvements, even from developers who presumably have effectively-unlimited token budgets to spend on improving their stack.
I love how people like Simon Willison and Pete Steinberger spend all this effort trying to be skeptical of their own experiences and arrive at nuanced takes like “50% more productive, but that’s actually a pretty big deal, but the nature of the increase is complicated” and y’all just keep repeating the brainrotted “100x, juniors are cooked” quote you heard someone say on LinkedIn.
AI gives you what you ask for. If you don't understand your true problems, and you ask it to solve the wrong problems, it doesn't matter how much compute you burn, you're still gonna fail.
I've been paying for the $20/m plan from Anthropic, Google, and OpenAI for the past few months (to evaluate which one I want to keep and to have a backup for outages and overages).
Gemini never goes down, OpenAI used to go down once in a while but is much more stable now, and Anthropic almost never goes a full week without throwing an error message or suffering downtime. It's a shame because I generally prefer Claude to the others.
Same here, but for API access to the big three instead of their web/app products, and Gemini also shows greater uptime.
But even when the API is up, all three have quite high API failure rates, such as tool calls not responding with valid JSON, or API calls timing out after five minutes with no response.
Definitely need robust error handling and retries with exponential backoff because maybe one in twenty-five calls fails and then succeeds on retry.
Invalid JSON and other formatting issues is more towards the model behavior I would say since no model guarantees that level of conformance to the schema. I wouldn't necessarily club it with the downtime of the API.
A lot of people might be discovering their preference for Claude.
All the AI labs are but Anthropic is the worst. Anyone serious about running Claude in prod is using Bedrock or Vertex. We've been pretty happy with Vertex.
I wonder why they haven't invested a lot more in the inference stack? Is it really that different from Google, OpenAI and other open weight models?
Have you used Bitbucket?
A core research library for MATLAB I used in a course project used to be on BitBucket, though thankfully didn't have to deal with a lot of collaboration there.
OpenAI used to be just as bad if not worse.
But they've stabilized the past 5 months.
Funny observation - it feels like being in the EU I get a much better AI SaaS experience than folks over in the US.
It’s like every other day, the moment US working hours start, AI (in my case I mostly use Anthropic, others may be better) starts dying or at least getting intermittent errors.
In EU working hours there’s rarely any outages.
By the time San Francisco comes online, your day is already done.
This is exactly my experience. It’s like Claude Code had a stroke during lunch, and when I return working it forgot how anything works.
Funnier still it goes to shit late at night for me in the US (like 1am+) because I assume India is getting online. Can't win.
Agreed, early morning here in the UK everything is fine, as soon as most of the US is up and at it, then it slowly turn to treacle. I've been testing z.ai for the past week and it's nowhere near as suceptible, fwiw.
To back up that observation:
I've seen a LOT of commentary on social media that Anthropic models (Claude / Opus) seem to degrade in capability when the US starts it's workday vs when the US is asleep.
And on the flip side, the status page literally says:
> Importantly, we never intentionally degrade model quality as a result of demand or other factors, and the issues mentioned above stem from unrelated bugs.
keyword: intentionally
the statement is carefully worded to avoid the true issue: an influx of traffic resulting in service quality unintentionally degrading
I wasn't trying to say they intentionally do it.
I was trying to say that systemic issues (such as load capacity) seem to degrade the models in US working hours and has been noticed by a non-zero number of users (myself included).
Yeah, also the CI queues start to get longer towards the end of the EU day as the Americans start their day.
Funny, my friend told me the same thing happens to Figma.
And just like that, the world became a little bit of a better place for a short while.
The Sentient Hyper-Optimized Data Access Network has acquired a meat suit and was last seen shambling toward In-n-Out.
SHODAN
It's okay, this FAQ helpfully highlights that 503 errors are common: https://claudelog.com/faqs/why-is-claude-code-showing-503-er...
FYI: It's not an official Claude / Anthropic website.
Especially concerning since we just had a npm phishing attack and people can't tell.
I don't think this is just the occasional 503s, and it is not just Claude Code. Their console is also down.
> APIs and Claude.ai are down. Services will be restored as soon as possible.
> This incident affects: claude.ai, console.anthropic.com, and api.anthropic.com.
Glad we have an internal load balancing/failover LLM inference system, so this outage won't take prod down for us.
Should I be looking at someting like an OpenRouter or AI gateway to ensure uptime for something that currently relies on Anthropic API?
Or is there a better alternative to address this availability concern?
OpenRouter works great! I wrote a coding agent CLI that uses it, new models get added all the time. You can check out the code here: https://github.com/nerds-with-keyboards/flite/blob/main/bin/...
Your best bet is having an account on AWS Bedrock & Vertex AI so you're able to route your request to the same model (such as claude-sonnet-4) but on a different provider.
Are the vibes off? (pun intended, sorry)
I've noticed a trend with their incident reports... "all fixed", basically. Little mind/words to prevention
The status page is not where you communicate about either the root cause or about the action plan for preventing recurrences.
You sure? That's exactly where I found this. Note the domain :)
https://status.cloud.google.com/incidents/ow5i3PPK96RduMcb1S...
edit: before some drive-by datamining nerd thinks I do/did SRE for Google, no
Fair enough! But that's not real-time communication during an active incident. It's communication O(days) later.
And that's totally fine! Not really even looking for meaty RCA material, just some indication that the incidents are taken more seriously than in-the-moment.
To be fair, too, it's likely been mentioned. I'm biased towards an unreasonable standard due to my line of work.
A status page without some thorough history is glorified 'About Us' :P
Extremely hard disagree. The status page is exactly where you communicate about both the root cause, and the action plan to prevent it.
Every status page incident on every normal company everywhere in the world has links to lead you to the postmortem and their steps to avoid it. Here are a few examples:
https://status.gitlab.com/ -> https://status.gitlab.com/pages/history/5b36dc6502d06804c083...
https://status.hetzner.com/ -> https://status.hetzner.com/incident/2e715748-fddd-427b-a07b-...
https://www.githubstatus.com/ -> https://www.githubstatus.com/incidents/mj067hg9slb4
https://bitbucket.status.atlassian.com/ -> https://bitbucket.status.atlassian.com/incidents/4mcg46242wz...
It's literally a standard for your status page to communicate both about root cause and action plan how to prevent it in the future. Sure, when an incident is just happening, the status page entry doesn't have the postmortem and the steps to avoid, but later on those get added.
Being so overconfidently wrong reminds me of an LLM.
To be fair, we did have rainy weather in Ashburn today (insert some joke about us-east-1 here).
The self feeding AI has destroyed itself at last
Man, Anthropic's service quality has just been a dumpster fire since July. Embarrassed I ever recommended people pony up for a 20x Max plan. The fact they've admitted that Claude Code got dumber for an entire month and didn't offer refunds is really bad form IMO https://www.reddit.com/r/ClaudeAI/comments/1nc4mem/update_on...
Z.AI works fine. Qwen works fine.
Glad I switched.
Qwen code is pretty decent but it’s no Claude. How would you compare Z?
Is it a surprise that a vibe coding company has vibe coded operational excellence practices?
github is down! I can’t write any more code!
The shoulders of giants we stand on are slumped in sadness.
Looks like it’s time to go outside and touch some grass again
"LLM down" is the new "it's compiling" https://xkcd.com/303/
What? Code by hand/brain like your elders. Young whippersnappers.
r/Anthropic/ currently filled with very irate customers.
They should learn to write a few lines of code themselves while they wait.
Hilarious, have you got any new ones?
They should learn to write a few lines of code themselves while they wait.
Not a good use of time. Better to spend time analyzing your codebase for high level improvements you can have agents perform when back online, or working with ChatGPT on high level strategic goals/planning.
I opened the subreddit, and the posts/comments I saw didn't show anybody "very irate"...
Well, there's a number that are cancelling, I'd call that irate.
Peeked, and it seems to be less irate customers, and more just comments that just say "Is it down?"
meanwhile I am amazed by the raw speed of grok in cursor. night and day to claude sonnet, and don't even talk about gpt5