[Discussion] Agent SDK no longer available under Pro/Max subscription billing starting June 15

[gcr]

Anthropic is changing the rules around Claude SDK billing: https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan

Starting Jun 15, Claude Pro and Max users won't be able to use Claude SDK with their subscription usage.

As a silver lining Anthropic will be offering a monthly credit for Agent SDK use, but users must explicitly opt-in via email to claim it and the language in the doc mentioning "instructions to claim their credit before June 15, 2026" makes me wonder whether the sign-up to claim this credit will be available after then.

I think this means Claude use in Zed will count toward extra usage, is that right? Does this match others' interpretation?

Wanted to flag as a potential issue to start some discussion about what to do.

[pan93412]

Related: anthropics/claude-code#32362

[dmeehan1968]

For starters, 20x is no longer 20x - the maximum credit is only 10x the Pro plan.

Given that they've just added agent manager to make orchestrating easier within the TUI, and that usage is still going to be allowed, the implication is not that they are trying to reign in compute, but make the official Claude clients (Code, Desktop) act as a gatekeeper for subscription usage.

That seems like a step backwards, because one reason (for me) using ACP is to allow clients to add functionality that the official client doesn't provide (such as a GUI, issue tracker integration, rich (HTML) agent responses that interact with the client, integrated terminal etc, etc).

This seems like a massive step backwards, not just in access to compute, but in developer experience.

I don't really want to go back to using an agent specific client, because of vendor lock in, as well as the loss of functionality.

Clearly the compute provided by the subscription is, at face value, an order of magnitude greater than the subscription cost. If Anthropic are really subsidising that, and the retail API prices aren't just inflated to push people towards subs, then clearly they are going to go bust one day. We should be paying for what we use, but its currently unclear whether the pricing is reflective of costs or just a scheme to switch our development behaviours until we have no choice but to pay to stay as productive as the competition.

I guess at least and ACP client gives me a choice.

[benbrandt]

Hey all yes usage in Zed via ACP (since we use the "blessed" path of SDK usage) will count against this credit. We have more on our blog here: https://zed.dev/blog/anthropic-subscription-changes

[matthieublanchard]

Adding a data point from a different angle than the Zed/commercial-client discussion above.

@dmeehan1968 nailed it — third-party clients add things the official one doesn't. My case is narrower: I'm a solo dev running a personal, non-commercial web UI over the
Agent SDK on my own subscription, used interactively, live, one prompt at a time (multi-session, per-session attachments, steer-mid-turn, file-based memory). No CI, no
cron, no resale.

The split keys off "first-party client vs SDK" rather than "human-in-the-loop vs automated", so this purely interactive personal usage gets metered like industrial
automation. If the credit pool stays, could there at least be a path for personal, human-paced SDK use to count against the normal subscription (real TTY / interactive
cadence / logged-in human)? Right now June 15 just taxes having a custom UI.

[dmeehan1968]

@matthieublanchard This is pretty much my situation, my ACP client has so far been for personal use, to build my own workflow for multi-agent use, but I am intending on releasing it, possible as open source ultimately. I don't think I have a way of commercialising it and ultimately its really a proof of concept around what IDE's look like in an agentic world.

I haven't added any automated loop features, so I can only take advantage of the agent inherent capabilities, such as parallel sub-agents. I don't have hand off.

As you rightly observe, the Anthropic change effectively says "you can only use our UI" which stiffles rather than promotes innovation. They are clearly ok with subsidisation of compute for certain classes of user (non-enterprise, human in the loop).

Ultimately its clear the market will go in the direction of non-subsidisation, just like everything once eyeballs are captured. I think its a bit dumb, and dangerous, to say that there is only one lens through which that can happen. AI may well turn out to be more compelling than mere attention, given its potential for a productivity multiplier rather than a productivity drain like other technologies. Also, its pretty clear that once they've milked the productivity enhancements with a human in the loop, their obvious growth opportunity is to remove the human, so further automation will come, and perhaps with the intention that only they will control how that's applied and monetised. It's pretty clear that this is an attempt to capture the value of human labour, rather than benefit humanity (well unless governments are prepared to tax the profits and redistribute, but I'm not holding my breath for that).