Tag
An AI agent that can be contacted via text, call, Slack, or email, takes actions on your behalf, knows when to ask for approval, and can even make purchases, described as a highly anticipated AI assistant launching today.
The author built Approv, a tool that pauses risky AI agent actions to request human approval via WhatsApp with signed audit logs, seeking feedback on handling human oversight.
OKed is a plugin for OpenClaw that intercepts risky tool calls and requires user approval before execution, preventing agents from performing destructive actions like deleting data or sending payments.
The article presents a simple pattern to classify AI agent tools as safe or dangerous, routing dangerous actions like sending emails or deleting files to a human approval node to prevent unintended execution.