← Learning paths

6 lessons · take it step by step

Agentic AI

Design bounded agents: workflows, execution loops, permissions, durable state, MCP version boundaries and reliability evaluation.

Start here

Follow the order if you are new, or jump straight to a topic you need.

Start the first lesson →

  1. Agents vs Workflows

    25–35 min

    Choose between deterministic code, fixed LLM workflows and model-directed tool loops using task uncertainty and operational cost.

  2. Bounded Agent Loop

    25–35 min

    Run model → tool → observation cycles with explicit stop conditions, reservations, deadlines and inspectable outcomes.

  3. Tool Design & Permissions

    25–35 min

    Turn model proposals into narrow, authorized actions with payload-bound idempotency and host-controlled approvals.

  4. Context, Memory & Run State

    25–35 min

    Keep model context, private conversation memory and durable execution checkpoints distinct.

  5. MCP Fundamentals & Version Boundaries

    25–35 min

    Separate host, client and server responsibilities and distinguish modern per-request metadata from legacy initialization.

  6. Agent Evaluation & Reliability

    25–35 min

    Evaluate the full action path: permitted tools, bounded work, recoverable state and justified terminal outcomes.