LEARNING PATH

AI agent engineering

Build verifiable AI agents from task decomposition and tools through evaluation and production governance.

← Back to knowledge center
Reading progress0%0/10 completed
Continue reading →
01

AI agent engineering

Build verifiable AI agents from task decomposition and tools through evaluation and production governance.

Open collection
TUTORIAL_CHAPTERAI agent foundations: tasks, state, and boundaries

Understand how agents differ from ordinary chat and build a minimal loop with explicit state and permission boundaries.

Open collection →
TUTORIAL_CHAPTERTool calls and function contracts: keep agents controllable

Design tool schemas, error responses, and idempotency rules to reduce argument drift and side effects.

Open collection →
TUTORIAL_CHAPTERPlanning and state machines: split long agent tasks into recoverable steps

Compare one-shot plans, staged execution, and state machines for replayable agent orchestration.

Open collection →
TUTORIAL_CHAPTERAgent evaluation and regression testing: verify actions, not only answers

Build task sets, tool traces, and safety assertions so Agent behavior can be checked after every change.

Open collection →
TUTORIAL_CHAPTERAgent observability and cost controls

Use traces, budgets, and redacted logs to find slow requests, loops, and unexpected model cost.

Open collection →
TUTORIAL_CHAPTERAgent production release: canary, rollback, and human handoff

Create release checks, a canary plan, rollback signals, and a human handoff path before an Agent reaches production.

Open collection →
casesCase study: an auditable support-ticket triage agent

Classify support tickets with read-only lookups and structured output, then hand uncertain cases to a person.

Open collection →
casesCase study: a deployment-advice agent with human approval

Let an agent prepare a release checklist while requiring a person to approve every production change.

Open collection →
TUTORIAL_CHAPTERAgent 重试、幂等与证据:让失败可以恢复 (original language)

用预算、幂等键和最小证据记录控制 Agent 重试,避免重复副作用。

Open collection →
casesCase study: an evidence package and human handoff after an agent failure

Turn tool traces, failure reasons, and next-step advice into an auditable handoff record.

Open collection →
7 chapters3 cases
Product Hub Demo · V0.3