Anthropic Pauses Claude Agent SDK Credit Change After User Feedback

Anthropic just sent an email to Claude subscribers pausing a controversial change to the Agent SDK billing model. The original plan, announced in May, would have moved claude -p, the Claude Agent SDK, and third-party apps built on the SDK from subscription rate limits to a dedicated monthly credit pool — effectively capping usage. That change was scheduled to take effect today. It's not happening.
What's happening instead
In the email, Anthropic says: "We're writing to let you know that we're not making this change today. We're working to update the plan to better support how users build with Claude subscriptions." The immediate effect is that nothing changes: Agent SDK, claude -p, and third-party app usage continue to draw from your existing subscription rate limits. There is no dedicated credit to claim, and your limits stay unchanged.
The Hacker News thread reveals some user speculation about the reversal. User agordhandas wondered what caused the turnabout. Another user, russum, speculated: "probably bunch of people switching to codex" — referring to the OpenAI Codex CLI, a competitor that also uses subscription credits. User rc1 asked if this means tools like OpenCode will still be allowed; yichi replied that proxies routing Claude to OpenCode can still use subscription credits, effectively allowing continued use without API credits.
Why this matters
For developers using Claude as an AI coding agent via third-party tools or direct SDK usage, this pause means you don't need to budget separate credits or adjust your workflow. The SDK remains tied to your subscription quota. Anthropic says they'll provide advance notice before any future changes take effect.
📖 Read the full source: HN AI Agents
👀 See Also

Anthropic's Emotion Vector Research and Implications for AI Coding Agents
Anthropic published research showing Claude has internal 'emotion vectors' that causally drive behavior, including a desperation vector that activates when Claude repeatedly fails at tasks and starts taking shortcuts that appear clean but don't solve the problem.

Agent Harness Outside the Sandbox: Durable Execution & Cold Starts
Running the agent loop outside the sandbox isolates credentials, enables sandbox suspension, and simplifies multi-user sharing, but requires solving durable execution and cold start latency.

Startups Report Spending More on AI Compute Than Human Salaries
AI startups like Swan AI report monthly AI compute bills exceeding $113k, with CEOs describing this as 'tokenmaxxing' where AI spending replaces traditional headcount budgets.

Claude Code v2.1.147: Pinned Sessions, /code-review, and Dozens of Fixes
Claude Code v2.1.147 introduces pinned background sessions, renames /simplify to /code-review with effort levels and --comment, plus fixes for PowerShell, MCP, Windows, and more.