Last night I got an email saying Claude Agent SDK won't move to separate billing for now. For now. Only for now 😏
I see a few reasons. The company really isn't ready to face the losses it would take after switching the Claude SDK to separate billing. Second - they want to polish the billing system, since people were warned in advance and most of them moved to plain Claude CLI, but in reality that didn't affect token usage much. Bringing in this restriction would scare off future users more than current ones
In theory, Claude Agent SDK won't move to separate billing because people were warned in advance and most of them moved to plain Claude CLI. It's more of a pause, time to think - to come back a bit later with a better thought-out mechanism for how the company can spend less money serving subscription users
The company has proven more than once that its focus is more on the business sector that actually pays for tokens via API - the money there is 30-40 times bigger than from subscription users. Although subscription users, of course, make up the bigger share overall, and catching hate from the audience means sawing off the branch you're sitting on 🪚, and taking a big hit to business and reputation
What this means for those working on a subscription
I'm one of them. I have 8 Claude Code agents working every day, and all of them run on a regular subscription rather than paying per token. Back in spring some of them were built on the SDK, and after the first email I moved everything to plain CLI. The move took time, but it taught me the main thing: a provider's rules can change overnight, and the system has to survive it
The delay in the email cancels nothing, it only gives time to prepare. If your bot or agent is tied entirely to one billing method from one company, now is the time to look at what happens to it when the delay ends
The company's math is simple. A subscriber pays the same amount no matter how much their agents work, while a business on the API pays for every token. The more people run 24/7 agents on a subscription, the harder that gap hits the company's revenue. So sooner or later the limits and rules for such agents will get stricter, it's only a matter of time
How I protect my systems from emails like this
First, a backup model. My personal agent Jes has been running for 134 days, and it has a primary model and a backup that takes over by itself if the first fails or hits a limit. Jes handles 10 scheduled jobs, and no pricing change has stopped them
Second, memory and rules kept separate from the model. Everything the agent knows about my business lives in files on my server. The model can be swapped, and the knowledge and habits stay in place
Third, each client gets their own server and subscription. That's how AI assistants run for five people, and a clinic owner has 6 agents. None of them depend on my account or will notice if something changes on my side
What to check on your side right now
Ask yourself three questions. What does your agent run on: a subscription, per-token billing or the SDK? What happens if the price doubles? How long would it take to switch to another model? If the answer to the last one is "I don't know" or "a month", the system is running on luck
A month wasn't enough for the company to sort out its economic issues, and with this sly email they've bought themselves time to solve it, indefinitely. But that time will definitely come. Don't think we'll be able to use Claude at the same volume on just a subscription, like now. Over the last month I'd have spent around 45 thousand dollars building my projects if I were paying for tokens via API. Draw your own conclusions 🤷
I explain how to build AI systems that survive changes in providers and prices on the page about AI system architecture