AO Agentic Operations
MODULE 05 / 10
All modules ↑
PLATFORM · TARGET L3 · 5–7 days
05

Bedrock AgentCore runtime

Deploy, observe, stop and recover the sandboxed agent runtime without broad permissions.

LearnLabIncidentReviewMentor guide

Deploy the pilot as an owned service with explicit identity, isolation, configuration, lifecycle and operational controls.

Prerequisites

  • Modules 01–04
  • VPC and IAM at L3
  • CI/CD and secrets at L2
01 / CONCEPT LESSONS

Build the mental model first.

01

Runtime boundary

Treat the agent as a production workload with inputs, outputs, dependencies and owners.

02

Workload identity

Separate deployment identity, runtime identity and human approver identity.

03

Environment isolation

Development and pilot environments require separate data, roles, logs and budgets.

04

Configuration and secrets

Keep model IDs, limits and endpoints configurable; secrets never enter prompts or source.

05

Lifecycle

Define deploy, health, drain, stop, rollback and decommission procedures.

02 / SYSTEM FLOW

Trace the complete path.

Caller identity→AgentCore endpoint→Runtime role→Model + tools→Telemetry→Policy decision→Response

For every arrow, the engineer must name the data, identity, trust boundary, failure mode and evidence produced.

03 / GUIDED LAB

Build it in bounded steps.

AI and documentation are allowed. The learner owns every generated artifact and must preserve the stated safety boundaries.

  1. 01Create a sandbox environment and data classification.
  2. 02Define separate deployment and runtime roles.
  3. 03Deploy the read-only graph.
  4. 04Configure health, timeout and concurrency limits.
  5. 05Prove logs contain no secrets.
  6. 06Execute stop, rollback and recovery procedures.
04 / INCIDENT SIMULATION

The runtime cannot call one tool; a quick fix proposes administrator access.

Mentor injects

  • Missing narrow API action
  • CloudTrail shows the denied request
  • Deployment identity is confused with runtime identity

Required evidence

  • Denied action and resource
  • Minimal policy diff
  • Proof other actions remain denied

Investigation loop

Symptom → impact → hypothesis → evidence → test → root cause → containment → correction → verification.

05 / DESIGN CHALLENGE

Start with a blank page.

Design dev, pilot and future client isolation. Include tenancy, data, identity, network and observability.

06 / VERIFY & SUBMIT

Verification checklist

Submission package

  • Deployment diagram
  • Role matrix
  • Environment checklist
  • Rollback evidence
  • Operations runbook

Review questions

  • Which identity calls the model?
  • What crosses environments?
  • How is a compromised runtime contained?
  • What is the rollback unit?
  • Who can stop the service?
07 / MENTOR-ONLY GUIDANCE

Reject any permission fix that cannot name the denied action, target resource and reason.

SCORING1 · cannot explain2 · explains with gaps3 · implements and verifies4 · designs and handles trade-offs

Record the achieved L-level only from observed evidence. Course completion and certification do not automatically change the skill matrix.