Anthropic is splitting Agent SDK billing in 48 hours. Your CI agents will break.

June 15. Two days.

Anthropic moves every programmatic Claude usage off your subscription and into a separate credit pool.

Agent SDK calls. claude -p. GitHub Actions. Third-party apps using Agent SDK auth. All of it now drains from a fixed monthly dollar credit instead of your subscription limits.

The numbers:
- Pro: $20/month
- Max 5x: $100/month
- Max 20x: $200/month
- Team Standard: $20/seat
- Team Premium: $100/seat

Credits don't roll over. Credits are per-user. Credits require a one-time opt-in.

Here's what breaks: if your credit runs out and you haven't enabled "usage credits" (the overflow toggle), your Agent SDK requests hard-stop. No fallback. No auto-billing. Just errors.

A nightly agent loop that comfortably fit under Max 20x can now run out mid-month. A CI job that "just worked" starts failing the moment the pool empties.

The escape hatch: point headless workloads at a dedicated API key. With ANTHROPIC_API_KEY set, claude -p and the Agent SDK authenticate against the API account. The June 15 pool never enters the picture. You pay list rates, but the bill lives where you can budget for it.

Three things to do this week: claim the one-time credit from the email Anthropic sends around June 8. Audit what your programmatic usage actually costs at API rates. Then decide per pipeline whether overflow should bill or hard-fail.

Interactive Claude Code in terminal and IDE? Unchanged. Claude chat on web and mobile? Unchanged. This only hits the workloads you can't watch in real time — which is precisely the category most likely to break quietly.

SOURCE: https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan
VERIFIED: Anthropic Help Center, The New Stack, Start Debugging
SIGNAL: Every team running Claude agents in production needs to audit billing before June 15 or risk silent pipeline failures.