- No trust that they won't nerf the tool/model behind the feature
- No trust they won't sunset the feature (the graveyard of LLM-features is vast and growing quickly while they throw stuff at the wall to see what sticks)
- No trust in the company long-term. Both in them being around at all and them not rug-pulling. I don't want to build on their "platform". I'll use their harness and their models but I don't want more lock-in than that.
If Anthropic goes "bad" I want to pick up and move to another harness and/or model with minimal fuss. Buying in to things like this would make that much harder.
I'm not going to build my business or my development flows on things I can't replicate myself. Also, I imagine debugging any of this would be maddening. The value add is just not there IMHO.
EDIT: Put another way, LLM companies are trying to climb the ladder to be a platform, I have zero interest in that, I was a "dumb pipe", I want a commodity, I want a provider, not a platform. Claude Code is as far into the dragon's lair that I want to venture and I'm only okay with that because I know I can jump to OpenCode/Codex/etc if/when Anthropic "goes bad".
I also clearly see the lock-in/moat strategy playing out here, and I don't like it. It's classic SV tactics. I've been burned too many times to let it happen again if I can help it.
To the contrary, they've proven again and again and again they'll absolutely do that the first chance they get.
For example, this demo (https://github.com/barnum-circus/barnum/tree/master/demos/co...) converts a folder of files from JS to TS. It's something an LLM could (probably) do a decent job of, but 1. not necessarily reliably, and 2. you can write a much more complicated workflow (e.g. retry logic, timeout logic, adding additional checks like "don't use as casts", etc), 3. you can be much more token efficient, and 4. you can be LLM agnostic.
So, IMO, in the presence of tools like that, you shouldn't bother using /loop, code routines, etc.
I see people making similar conclusions about various LLM providers. I suspect in the end it’ll shake out about the same way, the providers will become practically inoperable with each other either due to inconvenience, cost, or whatever. So I’ve not wasted much of my time thinking about it.
I guess I'm one of the people who disagree, specifically about AWS. I think a lot of companies just watch their bill go up because they don't have the appetite to unwind their previous decision to go all-in on AWS.
Ignoring egress fees, migrating storage and compute isn't hard, it's all the auxiliary stuff that's locked in, the IAM, Cognito, CloudFormation, EventBridge, etc... Good luck digging out of that hole. That's not to say that AWS doesn't work well, but unless you have a light footprint and avoided most of their extra services, the lock-in feels pretty real.
That's what it feels like Anthropic is doing here. You could have a cron job under your control, or you could outsource that to a Claude Routine. At some point the outsourced provider has so many hooks into your operations that it's too painful to extract yourself, so you just keep the status quo, even if there's pain.
your experience just hasn’t been my experience I guess. The more managed the service you use, the more costs you are going to pay - for a very long time I’ve got by with paying for compute, network, and storage on the barebones services. If you want to pay for convenience you will pay for it.
One area that was a little shitty that has changed a lot is egress costs, but we mostly have shifted to engineering around it. I’ve never minded all that much, and AWS support is so good at enterprise tiers that they’ll literally help you do it.
What grinds my gears is how Anthropic is actively avoiding standards. Like being the only harness that doesn't read AGENTS.md. I work on AI infra and use different models all the time, Opus is really good, but the competition is very close. There's just enough friction to testing those out though, and that's the point.
My point was, I don't think it mattered much, and it feels like an ok comparison - cloud offerings are mostly the exact same things, at least at their core, but the ecosystem around them is the moat, and how expensive it is to migrate off of them. I would not be surprised at all if frontier AI model providers go much the same way. I'm pretty much there already with how much I prefer claude code CLI, even if half the time I'm using it as a harness for OpenAI calls.
Claude Code routines sounds useful, but at the same time, under AI-codepocalypse, my guess is it would take an afternoon to have codex reimplement it using some existing freemium SaaS Cron platform, assuming I didn't want to roll my own (because of the maintenance overhead vs paying someone else to deal with that).
Yes, I expect that is very much the point here. A bunch of product guys got on a whiteboard and said, okay the thing is in wide use but the main moat is that our competitors are even more distrusted in the market than we are; other than that it's completely undifferentiated and can be swapped out in a heartbeat for multiple other offerings. How do we do we persuade our investors we have a locked in customer base that won't just up-stakes in favour of other options or just running open source models themselves?
Too bad we've now managed to turn programming into the same annoying guesswork.
I actually trust that they will.
1) that AI will be more advanced in the future
2) that the AI I am using will be worse in the future
# Note: This is inefficient, but deterministic and predictable. Previous
attempts at improvements led to hard-to-predict bugs and were
scrapped. TODO improve this function when AI gets better
* make sure the model maxes out all benchmarks
* release it
* after some time, nerf it
* repeat the same with the next model
However, the net sum is positive: in general, models from 2026 are better than those from 2024.
Not just that, but there’s really no way to come to an objective consensus of how well the model is performing in the first place. See: literally every thread discussing a Claude outage or change of some kind. “Opus is absolutely incredible, it’s one shotting work that would take me months” immediately followed by “no it’s totally nerfed now, it can’t even implement bubble sort for me.”
Sometimes you have to keep starting new session until it works. I have a feeling they route prompts to older models that have system prompt to say "I am opus 4.6", but really it's something older and more basic. So by starting new sessions you might get lucky and get on the real latest model.
Suuuuuuure it was.
That said, I had way better experiences with old (but contemporary) Apple hardware than any other kind of old hardware.
I never asked for a 1M context window, then I got it and it was nice, now it's as if it was gone again .. no biggie but if they had advertised it as a free-trial (which it feels like) I wouldn't have opted in.
Anyways, seems I'm just ranting, I still like Claude, yes but nonetheless it still feels like the game you described above.
https://x.com/lydiahallie/status/2039800718371307603
--- start quote ---
Digging into reports, most of the fastest burn came down to a few token-heavy patterns. Some tips:
• Sonnet 4.6 is the better default on Pro. Opus burns roughly twice as fast. Switch at session start.
• Lower the effort level or turn off extended thinking when you don't need deep reasoning. Switch at session start.
• Start fresh instead of resuming large sessions that have been idle ~1h
• Cap your context window, long sessions cost more CLAUDE_CODE_AUTO_COMPACT_WINDOW=200000
--- end quote ---
https://x.com/bcherny/status/2043163965648515234
--- start quote ---
We defaulted to medium [reasoning] as a result of user feedback about Claude using too many tokens. When we made the change, we (1) included it in the changelog and (2) showed a dialog when you opened Claude Code so you could choose to opt out. Literally nothing sneaky about it — this was us addressing user feedback in an obvious and explicit way.
--- end quote ---
but you can replicate these yourself! i'm happy that ant/oai are experimenting to find pmf for "llm for dev-tools". After they figure out the proper stickyness, (or if they go away or nerf or raise prices, etc) you can always take the off-ramp and implement your own llm/agent using the existing open-source models. The cost of building dev-tools is near zero. it is not like codegen where you need the frontier performance.
Chinese models (GLM, MiniMax) are better.
It changes a number of things. Not all tasks require very high intelligence, but a lot of data may be sensitive enough to avoid sharing it with a third party.
I can’t use Claude Code at all anymore, not even for simple tasks. The output genuinely disgusts me. Like a friend who constantly stabs you in the back.
My favorite AI feature at the moment is the JetBrains predict next edit. It‘s so fast that I don’t lose attention and I’m still fully under control.
I'm currently hosting, on very reasonable consumer grade hardware, an LLM that is on par performance wise what every anyone was paying for about a year ago. Including all the layers in between the model and the user.
Llama.cpp serves up Gemma-4-26B-A4B, Open WebUI handles the client details: system prompt, web search, image gen, file uploading etc. With Conduit and Tailscale providing the last layer so I can have a mobile experience as robust as anything I get from Anthropic, plus I know how all the pieces works and can upgrade, enhance, etc to my hearts delight. All this runs from a pretty standard MBP at > 70 tokens/sec.
If you want to better understand the agent side of things, look into Hermes agent and you can start understanding the internals of how all this stuff is done. You can run a very competitive coding agent using modest hardware and open models. In a similar note, image/video gen on local hardware has come a long way.
Just like Linux, you're going to exchanging time for this level of control, but it's something anyone who takes LLMs seriously and has the same concerns can easily get started with.
Yet I still see comments like this that seem to complete ignore the incredible work in the open model community that has been perpetually improving and is starting to really be competitive. If you relax the "local" requirement and just want more performance from an LLM backend you can replace the llama.cpp part with a call to Kimi 2.5 or Minimax 2.7 (which you could feasibly run at home, not kimi though). You can still control all the additional part of the experience but run models that are very competitive with current proprietary SoTA offering, 100% under your control still and a fraction of the price.
This isn’t an LLM. It’s a product powered by an LLM. You don’t get access to the model you get access to the product.
An LLM can’t do a web search, an LLM can’t convert Excel files into something and then into PDF. Products do that.
I think it’s a mistake to say I don’t trust this engine to get me here, rather than it is to say I don’t trust this car. Because for the most part, the engine, despite giving you a different performance all the time is roughly doing the same thing over and over.
The product is the curious entity you have no control over.
The funniest thing Ive heard is that now we have LLMs, Humanoid robots are on the horizon. Like wtf? People who jump to these conclusions were never deep thinkers in the first place. And thats OK, its good to signal that. So we know who to avoid.
Increasingly, we're also seeing the moat shrink somewhat. Frontier models are converging in performance (and I bet even Mythos will get matched) and harnesses are improving too across the board (OpenCode and Codex for example).
I get why they're trying to do that (a perception of a moat bloats the IPO price) but I have little faith there's any real moat at all (especially as competitors are still flush with cash).
So yeah, doesn't bode well for being a pure play model producer.
This Routines feature notably works with the subscription, and it also has API callbacks. So if my Telegram bot calls that API... do I get my Anthropic account nuked or not?
The Chilling Effect of this is real and it gets more and more frustrating that they can't or won't clarify.
I downgraded my $200/mo sub to $20 this past week and I’m going to try out Codex’s Pro plans. Between the cache TTL (does it even affect me? No idea), changes in the rate limit, 429 rate limit HTTP status code during business hours, adaptive thinking (literally the worst decision they’ve ever made, as far as my line of work is concerned), dumb agent behavior silently creating batshit insane fallthroughs, clearly vibe coded harness/infrastructure, and their total lack of transparency, I think I’m done. It was fun while it lasted but I’m tired of paying for their mistakes in capacity planning and I feel like the big rug pull (from all three SOTA providers) is coming like a freight train.
I didn't even know what opencode was prior to that drama, yet now here i am using opencode and a ton of crafted openai agents in my projects. Would love to have some claude agents in that mix, but i guess im stuck in Claude Code if i wanna even touch their models... I'd love to go back to just claude as i "trust" them more in a sorta less evil vibe manner, but if they are gonna prevent subscription usage to something people use to allow themselves more freedom, they gotta then close that gap with their own tools rather than pumping out stuff like this which scares me off given the past couple months.
I totally understand why they are cutting off 3pa access to stuff like openclaw, where the avg user is just a power user in comparison to avg claude user or whatever. I haven't kept up a ton with their opencode issues, but I just know i can't get behind a company actively trying to make my potential usage of tokens less optimized to keep me locked into their ecosystem.
Really just kinda hoping local models kill it all for devs after a few years, I'm not interested in perma relying on data centers for my workflow.
I've got a setup where GPT5-mini (Free on GH) talks to you to refine and draw the outline of your feature, calls a single Opus subagent to plan it in depth, then calls a single sonnet subagent to implement it.
Github will only charge you for one opus request (With a 3x multiplier) and one sonnet request, whether they consume 50 or 500.000 tokens. I'm running this setup for 9 hours a day at work and I've barely consumed 40% of my monthly allowance.
Another comparison would be "unlimited storage", where "unlimited" means some people will abuse it and the company will soon limit the "unlimited."
- SDK that allows you to use OAuth authentication!
- Docs updated to say DO NOT USE OAUTH authentication unless authorized! [0]
- Anthropic employee Tweeting "That's not what we meant! It's fine for personal use!" [1]
- An email sent out to everyone saying it's NOT fine do NOT use it [2]
Sigh.
[0] https://code.claude.com/docs/en/agent-sdk/overview#get-start...
[1] https://www.reddit.com/r/ClaudeAI/comments/1r8et0d/update_fr...
edit: And specifically i'm making an IDE, and trying to get ClaudeCode into it. I frankly have no clue when Claude usage is simply part of an IDE and "okay" and when it becomes a third party harness..
It says in the prohibited use section:
> Except when you are accessing our Services via an Anthropic API Key or where we otherwise explicitly permit it, to access the Services through automated or non-human means, whether through a bot, script, or otherwise.
So it seems like using a harness or your own tools to call claude -p is fine, AS LONG AS A HUMAN TRIGGERS IT. They don’t want you using the subscription to automate things calling claude -p… unless you do it through their automation tools I guess? But what if you use their automation tool to call your harness that calls claude -p? I don’t actually know. Does it matter if your tool loops to call claude -p? Or if your automation just makes repeated calls to a routine that uses your harness to make one claude -p call?
It is not nearly as clear as I thought 10 minutes ago.
Edit: Well, I was just checking my usage page and noticed the new 'Daily included routine runs' section, where it says you get 15 free routine runs with your subscription (at least with my max one), and then it switches to extra usage after that. So I guess that answers some of the questions... by using their routine functionality they are able to limit your automation potential (at least somewhat) in terms of maxing out your subscription usage.
It would be absurd to me if the same application is somehow allowed via ACP but not via official SDK. Though perhaps the official SDK offers data/features that they don't want you to use for certain scenarios? If that were they case though it would be nice if they actually published a per-SDK-API restrictions list.
That we're having to guess at this feels painful.
edit: Hah, hilariously you're still using the SDK even if you use ACP, since Claude doesn't have ACP support i believe? https://github.com/agentclientprotocol/claude-agent-acp
> My point is that they must apply these restrictions.
I fully understand and respect they need restrictions on how you can use your subscription (or any of their offerings). My issue is not there there _are_ restrictions but that the restrictions themselves are unclear which leads to people being unsure where the line is (that they are trying not to cross).
Put simply: At what point is `claude -p` usage not allowed on a subscription:
- Running `claude -p` from the CLI?
- Running `claude -p` on a Cron?
- Running `claude -p` as a response to some external event? (GH action, webhook, etc?)
- Running `claude -p` when I receive a Telegram/Discord/etc message (from myself)?
Different people will draw the line in different places and Anthropic is not forthcoming about what is or is not allowed. Essentially, there is a spectrum between "Running claude by hand on the command line" and "OpenClaw" [0] and we don't know where they draw the line. Because of that, and because the banning process is draconian and final with no appeals, it leads to a lot of frustration.
[0] I do not use OpenClaw nor am I arguing it should be allowed on the subscription. It would be nice if it was but I'm not saying it should be. I'm just saying that OpenClaw clearly is _not_ allowed but `claude -p` wouldn't be usable at all with a subscription if it was completely banned so what can it (safely) be used for?
The new reality of coding took away one of the best things for me - that the computer always just does what it is told to do. If the results are wrong it means I'm wrong, I made a bug and I can debug it. Here.. I'm not a hater, it's a powerful tool, but.. it's different.
... until this week! Opus is struggling worse than Sonnet those last two weeks.
There's utility in LLMs for coding, but having literally the entire platform vibe-coded is too much for me. At this point, I might genuinely believe they're not intentionally watering anything down, because it's incredibly believable that they just have no clue how any of it works anymore.
But this week I've lost count of the times I've had to say something along the lines of: "Can you check our plan/instructions, I'm pretty sure I said we need to do [this thing] but you've done [that thing]..."
And get hit with a "You're absolutely right...", which virtually never happened for me. I think maybe once since Opus 4-6.
We're watching a speed run of growthism, folks.
Oh uh... ok then.
EDIT: This comment is apparently [dead] and idk why.
I think the real issue stems from the 1 Million token context window change. They did not anticipate the amount of load it would give you. That first few days after they released the new token window, I was making amazing things in one single session from nothing, to something (a new .NET based programming language inspired by Python, and a Virtual Actor framework in Rust). I think since then they've been trying too many things to tweak things, whilst irritating their users.
They even added a new "Max" thinking mode, and made "High" the old medium, which is ridiculous because you think you're using "High" but really you're not. There's a hidden config file to change their terrible defaults to let Claude be smarter still, and apparently you can toggle off the 1M tokens.
I think the real fix, and I'm surprised nobody there has done this yet, is to let the user trim down their context window.
Think about it, you used to have what? 350k tokens or so? Now Claude will keep sending your prompt from 30 minutes ago that's completely irrelevant to the back-end, whereas 3 months ago it would have been compacted by now.
Others have noted that similar prompting for some ungodly reason adds tens of thousands of extra garbage tokens (not sure why).
Edit looks like someone figured out that if you downgrade your version of Claude Code and change one single setting it unruins Claude:
I'm thinking they should go back to all their old settings and as a user cap you at their old token limit, and ask you if you want to compact at your "soft" limit or burst for a little longer, to finish a task.
A bit annoying, but not the end of the world.
In a way, it’s true if china has superior AI then it’s dominance over US will materialize. But it’s not hard to see how this scenario is being used to essential lie and scam into trillions of debt.
Its interesting how the cutthroat space of big tech has manifested into an incidious hyper capitalist system where disrupting a system is it’s primary function. The system in this case is world order and western governments
Still using Claude for UX (playgrounds) and language. OpenAI has always been a little more cerebral and stern, which doesn’t suit those areas. When it tries to be friendly it comes off as someone my age trying to be a 20-something.
1. Anthropic realized their models weren't enough of a moat.
2. They built tools so they could expand their moat.
3. People don't want to use their tools, they want their models, and use other, better tools.
4. Anthropic bans the use of better tools, taking advantage of their model superiority to try to lock people into subpar tools.
"I don't have enough of a moat so I'll use my little moat and pretend it's a big one" doesn't sound like a great strategy. All they're doing with this anticonsumer behaviour is making sure that I'll leave the moment another model works for me as well as Claude does.
Ironically, they are now playing against their own models that can relatively easily build wrappers around any API shape into any other API shape.
Luckily you can turn if off pretty easily, but I don't know why it's on by default to begin with. I guess holdover from when people used it with a $20 subscription and didn't care.
It's very light on token usage in general, as well.
Your own, personal, Jevons.
It was a bit buggy, but it seems to work better now. Some use cases that worked for me:
1. Go over a slack channel used for feedback for an internal tool, triage, open issues, fix obvious ones, reply with the PR link. Some devs liked it, some freaked out. I kept it.
2. Surprisingly non code related - give me a daily rundown (GitHub activity, slack messages, emails) - tried it with non Claude Code scheduled tasks (CoWork) not as good, as it seems the GitHub connector only works in Claude Code. Really good correlation between threads that start on slack, related to email (outlook), or even my personal gmail.
I can share the markdowns if anyone is interested, but it's pretty basic.
Very useful, (when it works).
We ought to come up with a term for this new discipline, eg "software engineering" or "programming"
airgramming plusgramming programming maxgramming studiogramming
and recently the brand new way of working: Neogramming !
Personally I stick for now with the "Programming " tier. Maybe will upgrade to "Maxgramming" later this year...