[tao_oat]

A bit surprised by the snarky comments here -- I also want Claude to work reliably but very few (no?) companies have ever seen this level of rapid growth. We're going to go through a long fail-whale-style period and I can imagine very, very few companies that could avoid that.

[al_borland]

When the narrative around AI is that people should rely on it all the time, people will be judged by your token use (it better be high), the AI is smarter than everyone and will take all the jobs, the AI is the best programmer, and more… When things fail repeatedly, it highlights that the emperor has no clothes.

If it’s as good as they say, why can’t it figure out how to not go down every day?

How can people rely on it for their job if it goes down everyday? Maybe they shouldn’t rely on it.

If it’s supposed to be such a good engineer, why should it have the same scaling issues as Twitter did 20 years ago with 20 years of lessons learned and 20 years of development for more modern and scalable infrastructures? Shouldn’t it know all the tricks to scale and have redundancy to keep availability high? Does it not know the demands?

When expectations are out of line with reality, there will be snark when things fail. Those expectations have been force fed to us by these AI companies for years now, so I don’t have much sympathy or patience to offer them. They created these expectations of their platforms and if they can’t live up to them, then maybe it’s time for recalibrate the public image of what AI really is and what it can do… and what its limitations are.

[dgb23]

I have to agree with this. The economic culture around this tech is very toxic.

There seems to be a mass anxiety around the job market even. I‘ve seen a lot of social media content, including videos of people giving advice, especially to younger tech workers.

The most dangerous (psychologically, socially, economically) are people in important positions, who understand just enough to see some of its usefulness, but not enough to assess where its assumptions and guarantees actually are.

Even moreso if they see workers as a mere cost center instead of an asset.

But here is my perhaps naive, hopefully brave prediction: the real winners of this shake up are not decided yet, and neither bean counting nor superficial engagement with the topic will be sufficient or even useful.

[sassymuffinz]

So true, we’re constantly told that we’re now obsolete, a magic robot can do everything we can do without sleeping and for a fraction of the cost. Except occasionally the robot just doesn’t turn up to work or occasionally he appears drunk on the job. The elites think it’s fine now while it’s cheap but just wait until the agents are priced properly and cost 5x or 10x more.

Suddenly the fleshy meat sacks who used to do all this work, just slower, who have persistent memory, who get better and more experienced over time, who only require a few bananas to power their brains start looking like the more reliable option again.

The only reason these chat bots exist is because the upper crust don’t want to pay us to live properly, not because the robots can do it better, they just want to pay as little as possible.

[runeblaze]

I mean I used to work on model reliability with my little PhD degree and the models i manage go down all the time.

Some profs have a team of PhDs and things go to shit all the time. I don’t know why we expect $FRONTIER_LLM to do better

[al_borland]

We accept human errors, limitations, and failures. We can empathize with team of humans doing the best they can, and we know any failure is a chance for them to learn and grow.

The sales pitch of AI is that it’s better than humans and has no real limits; it will make us all obsolete. This framing they created means I expect it not to make errors, not to have limits, and not to fail. I expect it to be able to learn and adapt at the speed of light and solve complex problems beyond what a PhD could do. This is what we’ve been told with the narratives around future jobs, AI performance on PhD level tests, how coding is a solved problem, and pictures painted of what a future with AI will look like. While we may know this isn’t true, this is what they are selling, and that’s the standard I’m going to hold them to.

I don’t blame the customer for being upset the snake oil didn’t live up to its promises, I blame the snake oil salesman. We have every right to be upset with the snake oil salesman and ridicule him when his product doesn’t work. Maybe we don’t need better more reliable snake oil, maybe we need real medicine. If real medicine don’t exist, its better to be honest than to mislead people and say it does.

This isn’t to say AI is completely useless, but it’s not what’s being sold. The downtime just proves that, unless they aren’t using their own product. If that’s the case, why not?

[yakattak]

They’re asking for $100+/mo for the plans that are actually usable at scale. If I’m paying that much I have very high expectations.

There’s also the fact that they’re known for dogfooding heavily, I imagine that contributes to it a lot.

[stingraycharles]

> They’re asking for $100+/mo for the plans that are actually usable at scale. If I’m paying that much I have very high expectations.

If you think $100 is that much and get very high expectations from it, you're not the target customer. You're a loss leader to Anthropic, and the fact that you don't see that / still have high expectations means your expectations are unrealistic.

[dijit]

$100/m for SaaS is very steep.

For an entire productivity suite including mail, meetings and terabytes of backed up redundant storage with nearly no bandwidth limitations it's like $35/m for even the most expensive option.

[tao_oat]

Those products have very low COGS in comparison to this.

[bredren]

Claude Code and Codex are not SaaS products in the traditional sense.

[stingraycharles]

Comparing state of the art LLMs with Office365 / Gsuite is like comparing renting a datacenter vs an airbnb. Entirely different things.

Compare it to hosting models locally, that would be more apt. Or renting GPUs from cloud vendors.

[dijit]

It’s a SaaS, and the most expensive SaaS available.

If you’re saying an LLM provides more value than the office productivity suites , mail platforms and meeting platforms which run essentially the planet: then I am afraid, you have drunk the kool-aid.

If you’re evaluating software licenses you have to weigh the price to value, there can be value to these LLMs but its not 3x the productivity of Mail+Spreadsheets+Live Meetings+presentations+wordprocessing+filesharing.

Its just not.

[stingraycharles]

If you really think the value add by LLMs is comparable to email and calendar, I don’t understand why you don’t understand my point that you’re not the customer Anthropic cares about.

[baq]

you could arguably ditch the productivity suite and a few other 'essential' subscriptions to make room for this one, except the price point will get enshittified to hell in the coming months and years.

[dgb23]

I think the bigger point is that the price tag is simply not competitive, especially given all of the issues, downsides and dangers.

Whether Anthropic makes money from the $100 subscriptions or not, is their problem.

[tristanj]

Those subscription plans are a loss leader. Anthropic would rather have you pay per-token for their API, where they actually make money. By cutting subscription limits, they push people towards using their API. And it's working, there are people spending thousands per month on their API.

[forrestthewoods]

> They’re asking for $100+/mo for the plans that are actually usable at scale. If I’m paying that much I have very high expectations.

They’re losing money on you at that price point.

Or more precisely you’re paying for it by giving them training data.

[omega3]

I'm not convinced, Kimi 2.5, GLM 5.1, Minimax M2.7 are all fraction of the price and still make money on inference.

[baq]

they can adjust prices or you can adjust your expectations.

...and let's be realistic, it'll be both.

[trvz]

They always have the option to stop accepting new customers when their infrastructure is peaked out instead of lowering quality for everyone.

[stingraycharles]

You don't know whether this is due to infrastructure capacity or has other reasons (organizational). Also, "let's stop accepting new customers" is probably not a realistic choice for a hundred reasons.

[alkonaut]

That would mean in a way accepting that they are suddenly a service company with the aim to create revenue by selling services to customers for money.

[arcfour]

You can't stop accepting new customers unless you're fine with killing your potential future customer base. That's a ridiculous suggestion.

Either your current customers or your potential future customers are going to be unhappy so long as compute resources are finite. Take your pick.

[ed_mercer]

> That's a ridiculous suggestion.

Is it though? Claude's reliability is now at an all-time low of 98.7%. It's not a stretch to think that large companies will have second doubts about about adopting claude for their production environments.

[trvz]

Waiting lists are a thing.

[baq]

> You can't stop accepting new customers unless you're fine with killing your potential future customer base. That's a ridiculous suggestion.

what? they already have, they aren't releasing mythos except to a limited pre-approved customer base who is practically begging them to take their money. they can do that for lower tier models and at this point they should.

[arcfour]

Their rationale provided for that is safety-based, not infrastructure-based.

[coldtea]

>You can't stop accepting new customers unless you're fine with killing your potential future customer base. That's a ridiculous suggestion.

And yet, it's what any business with limited stock or slots (from restaurants and car companies to airlines) have done since forever...

[techpression]

Doesn’t matter, they are not handling it correctly, but instead keep selling while far over capacity. They should not accept more users until they can supply the service. We solved this thousands of years ago, it’s called waiting in line. And yes, it’s not common to see, but that doesn’t excuse not doing it.

[logicchains]

>I also want Claude to work reliably but very few (no?) companies have ever seen this level of rapid growth. We're going to go through a long fail-whale-style period and I can imagine very, very few companies that could avoid that.

Their main competitor OpenAI has much better uptime and more generous usage limits.

[code51]

Exactly this. OpenAI is running huge workloads silently, without anybody patting their back.

[rvz]

How can Claude work reliably if Claude keeps going on vacation for several hours?

Maybe it is recovering from the weekend a few days ago, but wanted to take an extra day off like it did on Monday, hence the "outage".

[ben_w]

> How can Claude work reliably if Claude keeps going on vacation for several hours?

Not that I wish to anthropomorphise it in this answer, but businesses have managed just fine when humans do this for "lunch breaks" and "going home for the evening to sleep".

(And even mandatory meetings which should have been emails).

[sassymuffinz]

Thing is, if Dave the programmer goes on vacation or calls in sick for the day, hopefully you have a larger team to fall back on and your business doesn't grind to a halt.

No one is apparently noticing that if they build their entire business model around AI being a certain price and availability they're essentially building one giant point of failure into their productivity.

What if the price shoots up 10x or Claude goes down for a day, or what if he's occasionally drunk (hallucinating). Reliability is sometimes a more important facet of business than ultra speed and productivity.

[ben_w]

Aye, correlated failure is not something to be overlooked. Mistaking correlated risk for uncorrelated risk was a big part of the global financial crisis.

There are fallback mechanisms when the risk is per model provider (as in, "What if the price shoots up 10x or Claude goes down for a day" is a manageable concern), but I'd be more worried about the way all models regardless of provider have similar failure modes, i.e. that some tasks fail in similar ways for all models. In some ways, LLMs are collectively like Star Trek's Borg: you've met one, you've met all of them.

[coldtea]

>I also want Claude to work reliably but very few (no?) companies have ever seen this level of rapid growth.

You do understand however that aside from the growth/maturity path, this is also a path to enshittification and skinning their users, which might come even faster to LMMs than say Google , because the latter managed to have hundres of billions in investments in record time to recoup and IPOs on sight.

[pllbnk]

They have this new Mythos model. I am sure it can fix all the bugs and reliability issues since it's nearly AGI. /s