An alarm fires, and now you have findings in three consoles, logs in two more, and no idea which tab to open first.

I built a real incident to work through. An EC2 instance in a throwaway account with SSH exposed to the internet, left running until scanners found it and brute-force attempts started. A CloudWatch alarm fired on its own, with real evidence underneath it across GuardDuty, Inspector, and Security Hub. Nothing staged.

Then I investigated it with two agentic tools that do different jobs.

Amazon Q in the AWS console is where it starts. No install, no subscription — click the icon and ask. It reads CloudWatch Logs, correlates across services, and answers from inside the account with nothing to wire up. For "what happened and where," it's the fastest path there is.

Kiro CLI in CloudShell is where you go when a question needs more than a lookup. I'll show the install, the Builder ID login, and the CloudShell quirks I hit. In return you get an agent that runs commands, chains CLI calls, and writes scripts against the account — and speaks MCP, so you can bring in context AWS has no knowledge of.

The session follows one alarm end to end and shows where the handoff happens: what Q answered instantly, the question where I needed a terminal instead, and what MCP let me reach that the console couldn't. Including the honest part — a prompt that gave me a confidently wrong answer about my own infrastructure, and how I caught it.

Along the way, why the same misconfiguration surfaced in three tools on three different days: Security Hub flagged the open security group before anything happened, Inspector flagged unpatched CVEs, GuardDuty flagged the attack only once it was underway.

To be clear, I never let the instance get compromised — no instance profile, no outbound internet, IMDSv2 enforced. I captured attempts, not successful payloads.

I'll close with the fixes, all free: Session Manager instead of an exposed port, security groups scoped to your own IP, GuardDuty on, a budget alarm, and deleting your lab when you're done.

You'll leave knowing: how to take an AWS alarm from firing to root cause with agentic tooling, which tool fits which stage, how MCP extends an investigation past AWS itself, and how to spot an agent that's confidently wrong.