Chatbots answer questions. AI agents take actions — they look up orders, issue refunds, update tickets and call internal APIs on their own. That is exactly why companies want them — and why they need careful engineering: an agent that can act can also act wrongly. In this 45-minute live session, we build a support-operations agent for a fictional e-commerce company on Claude, connect it to real tools through the Model Context Protocol (MCP), and watch it work. Then we try to break it — a refund it shouldn't give, a tricked instruction hidden in a customer message, an action it was never allowed to take — and add the guardrails engineers use to stop each one: permissions, human approval, audit logs and agent evals.
Software Engineer & AI Practitioner from Microsoft

20, 100 Feet Rd, Vivek Nagar,
Chandra Reddy Layout, AVS Layout Ejipura,
Bengaluru, Karnataka 560095




