Dev Rishi discusses the limitations of static rules and frequent human approvals when AI agents use multiple tools and can take actions at speed. He describes examples his company observed and treats them as operational anecdotes.
Rishi outlines a proposed approach combining cross-platform visibilityAI agent observability makes an agent's state, actions, tool use, failures, resource use, and outcomes visible enough to understand and operate it., a specialized model for context-sensitive policy decisionsGovernance sets responsibilities, rules and oversight for decisions about how AI systems are used., and recovery from destructive actions. Host Sam Charrington questions the reliability of smaller modelsModel reliability is the degree to which an AI model produces dependable behavior across repeated, varied and operationally realistic use. and the practical burden of recovery.
The conversation considers prompt injection, Model Context Protocol permissions and observability. Rishi describes his company’s product and makes claims about its performance; these are attributed to him and not independently verified.
Watch on YouTube
