Claude Code makes auto mode the default for paid plans
Two stories this evening, both about who gets to decide what an AI coding agent is allowed to do — Anthropic is handing that decision to a classifier, and Meta is finally entering the arena.
Anthropic will make auto mode the default in Claude Code for Pro, Max, and Team plans starting August 14, replacing the per-command approval flow with a classifier that decides when a human actually needs to be asked. Auto mode routes every action through a guard that blocks anything irreversible, destructive, or aimed outside the working environment, and only interrupts when it can't find a safe path itself. Anthropic's numbers are the hook: in a controlled study with 1,053 paid professional testers, human reviewers caught just 13.6% of dangerous commands while auto mode caught 89% — a block rate that held steady even in long sessions. The company also notes that users currently approve 97% of permission prompts, the polite way of saying manual approval had become rubber-stamping.
That's the real story: Anthropic is arguing a classifier is a better gatekeeper than the humans who were supposed to be watching. Early evidence supports it — auto mode users on Team and Enterprise plans ship about 25% more pull requests — and the company has stopped charging for the classifier's extra tokens on paid plans. But the trade-off is a governance shift, not a safety win: a classifier's judgment is only as good as its trust boundaries, and teams that haven't defined hard-deny rules for production and secrets are silently delegating that definition to Anthropic. Enterprise accounts, the Claude API, and cloud deployments stay opt-in for now, with defaults planned "in the coming month."
Meta is entering the coding-agent race with Muse Code, a terminal agent powered by the new Muse Spark 1.2 model, now in beta for macOS and Linux. On Terminal-Bench 2.1, Meta reports 82.9% — behind Claude Code on Opus 5 (86.7%) but ahead of Codex on GPT-5.6 Terra (81.8%) and Grok Build (81.6%); on the longer-horizon DeepSWE benchmark the gap to the leaders widens (59.3% vs. 65.0%). Those are Meta's own harness numbers, and the announcement notably doesn't tout placements — the charts just show a strong, clear second place. Meta's pitch instead is resilience and price: persistent background agents, an append-only event log that makes long-running tasks restart-safe after crashes, and aggressive pricing that positions Muse Code as the budget option in a field Anthropic and OpenAI currently own.
What to watch: whether security teams push back on the classifier default before August 14 — and whether Enterprise gets the same treatment next month.
Do you trust a classifier to decide when your coding agent should ask permission? Tell us in the comments.
