InterviewsVector
Capability spine

Arc 08 · Staff AI Engineer

Agentic Systems, Tools, and Protocols

Treat agents as bounded distributed workflows with typed tools, explicit state, permissions, recovery, and human authority.

Exit capability: Ship a tool-using workflow whose actions are bounded, observable, and recoverable.

Mapped lessons
7
Published now
1
Full-arc estimate
≈18 hours
Last edited
2026-08-11

Lesson sequence

Live units open into complete labs. Planned units stay visible to show the dependency path, but intentionally have no detail route.

  1. 01

    Tools Are Typed Authority Boundaries

    Planned

    Design tool names, schemas, errors, idempotency, and authorization for safe model use.

    Build labAdvanced105 min estimate

  2. 02

    A Tool-Using Agent Is a Bounded State Machine

    Live

    Implement budgets, typed transitions, approval gates, idempotency, and terminal states around a probabilistic planner.

    Build labAdvanced135 min estimateArtifact: Tested bounded agent runtime

  3. 03

    State, Memory, and Context Are Different Things

    Planned

    Separate durable workflow state, retrievable memory, and model-visible context.

    Concept labAdvanced95 min estimate

  4. 04

    MCP Host, Client, and Server Boundaries

    Planned

    Implement protocol capabilities without delegating consent, authorization, or validation to the model.

    Systems labAdvanced115 min estimate

  5. 05

    Durable Agent Workflows

    Planned

    Survive retries, restarts, partial tool success, human pauses, and changing external state.

    Systems labAdvanced120 min estimate

  6. 06

    When Multiple Agents Are Actually Justified

    Planned

    Use ownership, isolation, parallelism, and context constraints—not personas—to decide decomposition.

    Design reviewAdvanced100 min estimate

  7. 07

    Agent Red-Team: Untrusted Instructions Meet Real Tools

    Planned

    Test prompt injection, confused deputy behavior, data exfiltration, and excessive agency.

    Failure labAdvanced120 min estimateArtifact: Agent threat model