Day 2: Giving Agents Real Capabilities

An agent that can only talk is a chatbot with a system prompt. What makes it capable is what it can reach: real APIs, real code execution, and other agents it does not control directly. Today you give your agent hands — not by writing boilerplate API wrappers for hours, but by describing what you want it to connect to and watching it wire up the integration. By the end of today, your agent calls APIs, handles rate limits gracefully, executes code without taking your machine down, and talks to tools it does not own.

What you'll cover

• Calling APIs Reliably , errors, retries, and rate limits across Claude, OpenAI, and other providers, without your pipeline falling over.
• Rate Limits & Backoff , the specific retry pattern that keeps a busy agent from getting itself throttled.
• Automating Workflows with Events , triggering agent work from something real happening, not a chat window.
• Running Agent Code Safely , a locked-down sandbox so agent-generated code can't take your machine down with it.
• Agent-to-Agent Communication , connecting an agent to tools and other agents it doesn't own, using the emerging standard for it.

Next: Day 3 , Agent Skills.