Comment by joshstrange
6 hours ago
Well that's clear as mud.
I've complained, extensively, about this before but Anthropic really needs to make it clear what is and is not supported with or without a subscription. Until then, it's hard to know where you stand with using their products.
I say all of this as someone who doesn't use OpenClaw or any Claw-like product currently. I just want to know what I can and can't do and currently it's impossible to know.
They really need to figure out the rules, look I'd love to use a custom harness with Claude Code that I can extend, or build my own (which I'm doing) and use it with my Claude Code license, I don't want to overspend on tokens if I can help it. They really need to set their bar for the next model releases to use less tokens, or to trim their own cost for how these models are run. I'd be okay with a slightly slower experience with Claude Code if it meant similar throughput, but less cost, especially if I can build my own harness for it.
They might've just hit a ceiling with the quality they can get per token? Maybe the only real way left to scale quality is to increase token usage?
Out of curiosity, what kind of custom harness are you thinking?
Anthropic changing what you get for your subscription week to week is why I would never spend beyond a hobby-tier license. Great product. Probably. But maybe depending on what hours of the day you use it. If it suits them.
I can’t tell you how relieved I am that there are many capable open weight models in the wild to keep a ceiling on bad behavior.
I don’t get why people are so surprised. Didn’t they learn anything from Twitter APIs and the like. The APIs are open as long as they serve the short term problem then Anthropic builds the features people actually use (more or less) and ban the usage of APIs for competing clients
The poor communication and flip-flopping are what concern me.
How can I buy into an ecosystem that might disallow one of my main workflows? I currently use several hook scripts to route specific work to different models. Will they disallow that at some point? We don't know because they can't get their story straight.
Keep in mind this is hearsay, since we are reading something through a non-official channel, it's maybe not right to call it "flip-flopping"?
Given the lack of clear communication and the fact that their primary competitor openly supports the use of bespoke harnesses, I highly doubt this is an incorrect announcement.
Anthropic is destroying goodwill that is hard-won in this space. At the end of the day, people just need to do their work in a way that makes sense for them. In my case (someone who has been building ML/AI tools for 25 years @ MS & Apple), I have much better results using my bespoke harness. If I'm paying $200/month for compute, I should be able to use it in a way that works for me. Given the push back, I'm not alone.
3 replies →
I have no trouble believing that all labs are trying really hard to come up with an enticing bundle of something works for a wide variety of users, but it's hard to anticipate the popularity of something like OpenClaw, which completely blows through all previous usage patterns at population level.
It seems like a tall order to set lasting rules in this space at this point, where nobody really understands what is going to happen in a few weeks.
I think a good corollary idea to "vibe coding" is the "vibe product". There is so much stuff popping in and out of existence and my excitement has declined.
Same building on their API. You design around what you think is allowed, then a blog post shifts everything. A proper developer policy page would fix this.
Stealibg OAuth keys from first party app to impersonate it in order to not have to pay for usage with properly generated API key was never part of normal use anywhere.
Yeah, the main point here is they had a CLI specifically that allowed you to call Claude, and that was being used. The CLI giving you access should kind of indicate that you should be able to use it as it is defined in the help.
I do agree, though, that the parts of this that were actually using the Claude system to generate OAuth keys themselves are a little sus.
That makes sense to say “must use Claude harness to login before calling Claude cli or using Claude code sdk”
You're not stealing oauth keys, they are your keys??
They probably decreased the cost and limited these external calls