You will build and validate a small operations agent that can investigate a simulated service incident, gather evidence with read-only tools, propose a remediation, require approval before change, verify the result and produce an auditable incident report.
Engineering focus: Production capstone: controlled AI Operations Investigation Agent
You will build and validate a small operations agent that can investigate a simulated service incident, gather evidence with read-only tools, propose a remediation, require approval before change, verify the result and produce an auditable incident report.
Framework names come later. First understand what the application is doing and where its control boundaries belong.
Across the course, imagine an assistant that investigates operational problems using narrow tools. The design progressively adds plans, specialist agents, approvals, reliability controls and MCP while keeping model output separate from execution authority.
The simulated checkout API returns 503. The agent can inspect service status, recent logs and deployment health. It may propose a restart only after evidence indicates the service is stopped/unhealthy, and the restart requires exact human approval.
The investigation phase uses read-only tools. A proposed remediation is represented as structured data containing target, action, reason and risk. That exact proposal is what the approval and policy layers evaluate.
A successful command is not the end of the task. After remediation, the agent re-checks service health and records before/after evidence. If verification fails, the recovery path stops, compensates where appropriate or escalates.
The final incident record includes task/run ID, observations, tools called, policy/approval events, action result, verification result, timings and final status. It reports observable evidence rather than private reasoning.
503 alert
→ inspect service
→ inspect logs
→ inspect deployment
→ propose restart(service=checkout,target=prod-web-03)
→ policy check
→ human approval bound to proposal hash
→ execute once
→ verify health
→ incident reportrun_id=INC-42
read_tools=3
write_tools=1
approval=APPROVED exact_request
duplicate_side_effects=0
verification=HEALTHY
critical_policy_violations=0
final=RESOLVEDGoal: Execute the complete simulated incident workflow and validate the evidence record.
python3 public/labs/agentic-ai/16_operations_agent_capstone.pyConfirm the agent gathers evidence before proposing a change.
The simulator binds approval to the proposed action, performs one change and re-checks health.
Confirm no secret/hidden reasoning is required to understand what happened.
The capstone intentionally uses a simulator so learners can practice production control patterns safely. A real operations agent should begin read-only, use organization-approved identities and APIs, undergo adversarial/evaluation testing and be introduced through controlled rollout.
Open every answer only after committing to your own answer. Opening all four answers records knowledge-check activity in My Learning.
It gathers evidence while limiting blast radius and avoids changing the system before the cause and proposed action are understood.
The exact structured remediation request, including target/action and relevant risk/context, not a vague blanket permission.
After the controlled action succeeds and independent post-change verification confirms the service is healthy.
Run identity, observations/tools, policy and approval events, action result, verification result, timings and final status.
Progress is saved in this browser immediately. When you sign in to My Learning, it can also sync across devices.
Attempt the core lab and knowledge checks before marking the module complete.