Claude Code tops $2.5B, ships auto mode

This is less a convenience feature than a trust contract: Anthropic is asking developers to let a black-box classifier decide when an agent can touch their machine. That will feel liberating in side projects and awkward fast in production unless the system becomes explainable.

• –Auto mode sits between manual approvals and `--dangerously-skip-permissions`, which is the right ergonomic middle ground for long coding sessions.
• –Anthropic says the classifier blocks destructive actions like mass deletion, exfiltration, and malicious execution, but it can still let ambiguous risky actions through.
• –The real adoption bottleneck is explainability: if blocks feel arbitrary, teams will either ignore them or fall back to manual approvals.
• –The $2.5B run-rate figure says Claude Code is now a serious business line, which helps explain why Anthropic is pushing harder on autonomy.
• –More surfaces mean more risk: as Claude moves beyond the terminal into web and messaging workflows, permission UX becomes part of the product, not just a settings page.