AI Agents and the Burden of Repeated Permission Requests

Summary

A Hacker News post asks whether people still use permission prompts with AI agents. The author says they became exhausted by constantly monitoring a terminal because the agent repeatedly asked for approval before using tools. The post argues that existing harnesses often fail to remember permissions after users grant them, turning routine work into a long sequence of manual confirmations. To avoid this, the author says Claude has been configured as an alias for `claude --dangerously-skip-permissions`, while Codex has used `approvals_reviewer = "auto_review"` for about a year. In a follow-up comment, the author says the experience caused anxiety close to a panic attack and that they felt better after changing the workflow. The discussion focuses on the tradeoff between smoother agent operation and the risks of bypassing or automating permission review.