AI Agents in Practice

Building the loop that survives production: observe → decide → act → check → repeat, the patterns that extend it, and the boundaries that keep an agent inside its lane.

For engineers building LLM-powered agents that have to hold up in production — not demo-only loops.

8 parts. Start at Part 1.

  1. Part 1 — The Demo Worked. Production Didn't.

    Beginner

    Why agent demos break the moment they meet production, and what the demo hid.

  2. Part 2 — What Makes Something an Agent

    Beginner

    A control loop with tools, state, and boundaries; MCP, RAG, and Skills as the primitives an agent composes.

  3. Part 3 — How the Control Loop Actually Works

    Intermediate

    State across turns, stopping conditions, and context as a finite resource.

  4. Part 4 — Five Agent Patterns and the Control Surfaces That Make Them Safe

    Intermediate

    Prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer, and the control surfaces that keep each shape shippable.

  5. Part 5 — Workflow, Agent, or Single LLM Call: How to Decide

    Intermediate

    Who decides the next step, and why hybrid is the steady-state shape for most production systems.

  6. Part 6 — Building the Production Agent Loop

    Intermediate

    Tool contracts, state, budgets, traces, and why a 200 OK is not proof the world changed.

  7. Part 7 — When the Loop Goes Wrong: Reading Agent Failures from the Trace

    Intermediate

    Classify the failure from the trace before reaching for a bigger model.

  8. Part 8 — The Boundaries That Keep Agents Safe

    Advanced

    Bound what an agent can see, do, remember, and prove.