← All roadmaps
Intermediate~8 weeks7 steps

Agentic Engineering

Design and ship AI agents: the agent loop, tool use, MCP, evaluation, and guardrails that keep autonomous systems safe. Best if you're comfortable with Python or JavaScript and have called an LLM API before.

  1. What is an agent

    The model-tools-loop anatomy, and when an agent beats a simple workflow.

  2. Tool use fundamentals

    Define tool schemas, execute calls, and return results the model can act on.

  3. Workflow & agent patterns

    Prompt chaining, routing, parallelization, and orchestrator-worker designs.

  4. Model Context Protocol (MCP)

    Connect agents to data and tools through the open MCP standard.

  5. Evaluating agents

    Trace, test, and score multi-step behavior — not just single responses.

  6. Guardrails & safety

    Sandboxing, permissioning, human-in-the-loop, and defending against prompt injection.

  7. Build a capstone agent

    Put it together: build a working agent with an SDK and real tools.