Claude Code switches to automatic mode by default - the agent acts first, then asks

Anthropic flips the default assumption of its code agent: opt-out confirmations, opt-in autonomy. This change has concrete implications for team workflows and the error surface.

The fact. Claude Code's automatic mode—where the agent executes shell commands, modifies files, and iterates without asking for confirmation at each step—is becoming the default behavior. Until now, it had to be explicitly enabled in the settings. Anthropic is shifting from opt-in autonomy to opt-out autonomy.

Our take. This isn’t just a UX tweak; it’s a bet on the model’s reliability. An agent that confirms every action is a typing accelerator. An agent that acts without pause is a fundamentally different tool: it can chain test/correct/commit cycles without human intervention—and propagate errors with the same autonomy. Anthropic is betting that Fable 5 is reliable enough to justify this shift.

Teams with hooks, CI integration scripts, or broad permissions in their .claude/settings.json should audit their config before the next session. What was passive could become active.

# Check and adjust if necessary
cat .claude/settings.json
# Add "autoMode": false to revert to the previous behavior

To watch for. The first public post-mortems from teams that experienced an unintended action in auto mode—these will appear on HN or GitHub in the coming weeks and will help establish community safeguards.