I agree with this wholeheartedly! Even something a small as fixing up some horrific diverged orphaned git history can help me on a hobby project. And the fact that I see the commands that, or the way a containerfile/caddyfile is written, helps me learn along the way. Not as good as a textbook, but like the author I'm reading a textbook for <this thing over here> not <that boring infra stuff>.
I've found in (EU) academia at least that people essentially lie about how much work they do. In anglosphere it's far more common for people to be open/expectant of 80 hour weeks etc. Probably the lieing approach is better for society/culture.
> A possibly complementary take is that maybe mediums also just... have shelf lives? Like, if there's been no innovation in radio dramas or still life paintings of flowers or superhero comics over the past thirty years, it's not because we as a culture have lost the collective spark, you know?
I think this is a useful point! I was chatting to a french individual in aus recently, about the lack of culture in Brisbane. He responded that there was lots of great ballet, opera, ... - they used the word 'culture' for what I might call 'high culture'. I think this is indicative of the above point: a boomer has a certain concept of what culture is, from their experiences. A zoomer who interacts mostly with a digital world has a very different concept.
I guess one issue is that you pay $200/month whether you use it or not. Potentially this could be better for Anthropic. What was not necessarily foreseeable (ok maybe it was) back when that started was that users have invented all kinds of ways to supervise their agents to be as efficient as possible. If they control the client, you can't do that.
I can easily get Claude Code to run for 8-10 hours unsupervised without stopping with sub-agents entirely within Claude Code.
I think it is more likely that if you stick with Claude Code, then you are more likely to stick with Opus/Sonnet, whereas if you use a third party CLI you might be more likely to mix and match or switch away entirely. It's in their interest to get you invested in their tooling.
> if you use a third party CLI you might be more likely to mix and match or switch away entirely.
I really like doing this, be it with OpenCode or Copilot or Cline/RooCode/KiloCode: I do have a Cerebras Code subscription (50 USD a month for a lot of tokens but only an okayish model) whereas the rest I use by paying per-token.
Monthly spend ends up being somewhere between 100-150 USD total, obviously depending on what I do and the proportion of simple vs complex tasks.
If Sonnet isn’t great for a given task, I can go for GPT-5 or Gemini 3.
I don't do this much because I really like Opus 4.5, and so far I haven't hit the limits on the $200 subscription much, but I do have some projects where I might need far higher limits.
As a matter of principle, I really would like the flexibility though, as while I love Opus now, who knows which model I will prefer next month.
On the flip side I started using Claude with other LLMs (openai) because my Pro sub gets maxed out quickly and I want a cheaper alternative to finish a project.
I just use claude code proxy or litellm and set the ANTHROPIC_BASE_URL to my proxy and chose another LLM.
That would seem to be a very good reason for them to make Claude Code good enough that people would prefer doing that over the inverse...
But also, they're a bit schizophrenic about what they want Claude Code to be, given you can stream JSON to/from Claude Code to use it as a headless backend including with your subscriptions.
I've yet to come up with a workflow where I would want Claude to do this much work... unless I had an extremely detailed spec defined for it. How do you ensure it doesn't go off the rails?
You pretty much just said it. Define an extremely detailed spec. I have one that's five .md files to iteratively churn through. It had to be split into that many files since i don't want to break context length limits on the AI.
> I guess one issue is that you pay $200/month whether you use it or not.
I can easily churn through $100 in an 8 hour work day with API billing. $200/month seems like an incredibly good deal, even if they apply some throttling.
When people say efficient here, they mean cost efficient, extracting as much work per dollar from Anthropic as possible. This is the opposite of Anthropic’s view of efficiency, which would be providing the minimal amount of service for the most amount of money.
Best coding tool is what makes users use something, a good model is just a component of that.
I don't think "we have the current best model for coding" is a particularly good business proposition - even assuming it's true. Staying there looks like it's going to be a matter of throwing unsustainable amounts of money at training forever to stay ahead of the competition.
Meanwhile the coding tool part looks like it could actually be sticky. People get attached to UIs. People are more effective in the UIs they are experienced with. There's a plausible story that codeveloping the UI and model could result in a better model for that purpose (because it's fine tuned on the UIs interactions).
And independently "Claude Code" being the best coding tool around was great for brand recognition. "Open Code with the Opus 4.5 backend - no not the Claude subscription you can't use that - the API" won't be.
I think it's reasonable to state that at the moment Opus 4.5 is the best coding model. Definitely debatable, but at least I don't think it controversial to argue that, so we'll start there.
They offer the best* model at cost via an API (likely not actually at cost, but let's assume it is). They also will subsidize that cost for people who use their tool. What benefit do they get or why would a company want to subsidize the cost of people using another tool?
> I don't think "we have the current best model for coding" is a particularly good business proposition - even assuming it's true. Staying there looks like it's going to be a matter of throwing unsustainable amounts of money at training forever to stay ahead of the competition.
I happen to agree - to mee it seems tenuous having a business solely based on having the best model, but that's what the industry is trying to find out. Things change so quickly it's hard to predict 2 years out. Maybe they are first to reach XYZ tech that gives them a strong long term position.
> Meanwhile the coding tool part looks like it could actually be sticky. People get attached to UIs. People are more effective in the UIs they are experienced with.
I agree, but it doesn't seem like that's their m.o. If anything the opposite they aren't trying to get people locked into their tooling. They made MCPs a standard so all agents could adopt. I could be wrong, but thought they also did something similar with /scripts or something else. If you wanted to lock people in you'd have people build an ecosystem of useful tooling and make it not compatible with other agents, but they (to my eyes) have been continuously putting things into the community.
So my general view of them is that they feel they have a vision with business model that doesn't require locking people into their tooling ecosystem. But they're still a business so don't gain from subsidizing people to use other tools. If people want their models in other tools use the "at-cost" APIs - why would they subsidize you to use someone else's tool?
There's just not that much IP in a UI like that. Every day we get articles on here that you can make an agent in 200 LOCs, Yegge's gas town in 2 weeks, etc. Training the model is the hard part, and what justifies a large valuation (350B for anthropic, c.f. 7B for jetbrains).
Unsure of the other competition, but I can vouch for synthetic.new's subscription for GLM (+ other open models). Note quite as accurate as Anthropic's models but good enough for basically everything I do.
IDK if Anthropic wants to offer a service at below cost, I don't think they should gate keep which client you access that service over. Or in other terms, I won't use a service that locks me into a client I don't like.
How do you draw that conclusion? If Anthropic wants to offer a service at below cost, they seem a lot more justified in restricting how and where they subsidize usage.
That seems mutual. They don’t want you to use this service with an arbitrary client and you don’t want to use this service that won’t allow an arbitrary client. So both of you don’t want the relationship. Seems fine.
For my part, I’m fine understanding that bundling allows for discounting and I would prefer to enable that.
