Anthropic defaults Claude Code to autonomous mode for paid tiers

The buried detail here is the subscriber tier specificity: Pro, Max, and Team plans get auto mode as default, but Enterprise is conspicuously absent from the announcement, suggesting Anthropic is still calibrating how much autonomy to hand large organizational customers by default.

This move sits directly downstream of the reliability questions surfaced in our coverage of Steve Yegge's Gas Town collapse (Simon Willison, early August), where Claude Opus 4.7 developed recursive failure modes in autonomous tooling. Anthropic is now making a confidence bet that those failure modes are sufficiently contained for general subscriber use, even as that incident illustrated how quickly autonomous agents can undermine the systems they're meant to serve. The David Crawshaw piece from August 3rd also framed LLM-driven autonomous maintenance as a maturing use case, and this default change is Anthropic institutionalizing exactly that posture. The EU AI Act enforcement coverage from August 3rd adds a regulatory wrinkle worth noting: defaulting to autonomous execution for European subscribers may trigger scrutiny under the Act's high-risk system provisions.

Watch whether Anthropic extends auto mode as default to Enterprise plans within the next 90 days. If they do, it signals the safety validation held under real-world load. If Enterprise stays opt-in past Q4 2026, that's a meaningful signal about where Anthropic's internal confidence actually stops.