Using HumanLayer to give agents safe access to high-stakes tools

I’ve been hacking on a python toolkit to enable AI agents to communicate with humans in tool-based and asynchronous workflows. I’ll show how HumanLayer can be used to gate access to high-stakes function calls like “email_customer” or “run_production_sql_query”, requiring human approval in slack before executing the tool. Time permitting, I’ll also go into a (very WIP) version of a more generic “human as tool” that enables an agent to consult various members of a team for advice and questions. I can talk about challenges with different tool-calling models and how we improved agent accuracy in calling the correct tool.