← The Quick Suite AI Layer

Q9 lab — the AI layer, hands on

Before you start

• Account: an AWS-account-connected Quick subscription (Professional or Enterprise; the 30-day trial covers up to 25 users). Admin or Author Pro role for the space/agent sections; Enterprise Edition if you attempt the topic section.
• Cost: seat + $250/account/month infrastructure fee apply while subscribed. This lab's usage stays inside included allowances except Part 5 (one Research run ≈ 20–40 min of your pooled Research hours) and any index storage you upload (a few MB — negligible against 25 GB/user). The teardown at the end removes everything.
• Have the AWS CLI configured with a role that can call quicksight:* in your home Region.

Write your answers to Q1–Q10 down. The point is evidence, not vibes.

Part 1 — Spaces and the permission exception

1. Create a space named lab-q9 (console: Spaces → Create). Q1: After creation, run aws quicksight list-spaces --aws-account-id <acct> — what is the space's ID and your permission level on it?
2. Upload two files: any PDF and a .csv. Watch the status walk Uploading → Processing → Text ready → Ready. Q2: How long did each stage take, and could you query the PDF's text before its status hit Ready?
3. Link one existing dashboard into the space (skip if none — note it).
4. Share the space with a colleague (or test user) as Viewer. Have them open it. Q3: Which can they open — the uploaded files, the linked dashboard, or both? Explain the difference using the rule from lesson 2.
5. Deliberate failure #1: As the viewer, try to delete the space. Record the exact error. Q4: What permission level is required to delete, and what happens to the files vs the linked dashboard when a delete succeeds?

Part 2 — Chat scoping and the audit trail

6. Open chat. Ask the same question ("what were our top items last month?" or similar) three times: once in General knowledge, once in All data and apps, once in Specific data and apps pinned to lab-q9. Q5: For each mode, note what sources (if any) the citations point at, and which answers carry an Explanations panel.
7. If you have a SPICE dataset: pin chat to it, ask a numeric question, open Explanations → Generated SQL. Q6: Paste the SQL. Does it match what you'd have written?

Part 3 — Build a custom agent

8. Chat agents → Create chat agent → natural-language flow: "An agent that answers questions about the files in the lab-q9 space, in a formal tone, and refuses questions outside that scope." Review what Generate produced in the builder.
9. Move the "refuses questions outside scope" instruction into Persona instructions (if the generator didn't), link lab-q9 under Knowledge sources, and add one suggested prompt. Launch it, then share it to your test user as Viewer.
10. Q7: Ask the agent something answerable only from your uploaded PDF, and something clearly out of scope. Record both behaviors. Then run aws quicksight describe-agent --aws-account-id <acct> --agent-id <id> (get the ID from list-agents) — what AgentLifecycle state is it in?
11. Deliberate failure #2: Try to attach an 11th resource (or in the API, an 11th space) to the agent. Record the error. Q8: Which documented limit did you hit, and is it adjustable?

Part 4 — A flow with a real limit

12. Flows → create from natural language: "Ask the user for a topic, search the web for it, summarize the top findings in three bullets." Run it in run mode.
13. Open the visual editor and count the steps; note the step types against lesson 5's list.
14. Q9: Try to add a schedule. Does your home Region allow it? Relate what you see to the documented Region list (us-east-1, us-west-2, eu-west-1).

Part 5 — Research (optional, burns ~0.5 agent hour)

15. Start a Research run on a topic you know well, sources = web + the lab-q9 space. Note the progress display while it runs (20–40 min — go do Part 6).
16. When it lands: click two citations, then use Understand the statement on one claim. Q10: Did the evidence actually support the claim? Export the report to PDF.

Part 6 — Where the logs are

17. In CloudTrail (home Region), find the events for your CreateSpace / CreateAgent / DescribeAgent calls from this lab. Record one event ID.
18. Look for your chat questions from Part 2 in CloudTrail. You won't find them. Then check whether CloudWatch vended logs (CHAT_LOGS) are enabled in your account — if not, enable them now if you're authorized; that's the day-one governance task from lesson 6.

Teardown

• Delete the flow, the custom agent (aws quicksight delete-agent …), and the space (console or delete-space) — recall Q4: files go with it, the dashboard survives.
• Delete the Research report if you don't want it stored (personal-space artifact storage).
• If this was a trial account spun up for the lab, cancel the subscription before day 30 — the $250 infrastructure fee bills monthly.

Done when you can

• State, from memory, which two things a space Viewer can always access.
• Show a numeric chat answer's generated SQL in under 30 seconds.
• Explain why the agent refused (or failed to refuse) out-of-scope questions, and which builder section controls it.
• Say which limit stopped the 11th resource and where it's documented.
• Point at the CloudTrail event for CreateAgent and explain why the chat content isn't there.

Facilitator notes

• Run Parts 1–3 as the core 60-minute block; 4 and 6 add 30; Research (Part 5) is a background thread started early so it lands before wrap-up.
• The teachable moment is Q3 — pair people up so someone actually experiences the viewer's perspective on space files vs linked dashboards.
• If the org has no test user, the sharing steps work with a second browser profile on the facilitator's account — worse, but workable.
• Expect the Q8 failure to behave differently console vs API (console may hide the option, API errors) — that difference is itself worth two minutes of discussion.