MODULE 02 · IDEATION LAB

Ideation with Amazon Quick

A live, seven-prompt walkthrough your trainers can run in the room. From a fuzzy idea to a scored shortlist to a shareable HTML proposal — every prompt scripted, every expected output shown.

~30 min 7 prompts HTML proposal at the end Reusable meta-prompt

Set the room up for the demo

Five minutes of prep, then you can run the whole ideation live. Everything you need is here.

The scenario · use this or swap in your own

PINd.AI's next product bet

You're a small AI-first startup with three fuzzy ideas competing for the next quarter's engineering focus. The founders have some rough notes, a rough sense of market, and no time to write a formal proposal. Your job in the next 30 minutes — using nothing but Amazon Quick — is to score the three ideas, pick a winner, and produce a shareable one-page HTML brief the leadership team can react to.

Product ideation Structured prompting HTML output Shareable brief

Before you start — check these off

A Quick account

Free tier at quick.aws.com is enough for this demo. Enterprise/AWS Console account works too.

The three ideas (one line each)

Have the fuzzy pitch ready in a note. See the scenario above for the built-in example we'll use in the walkthrough.

A fresh chat window

Open Quick chat with a clean session. Do not use a Space for this demo — we're teaching prompting, not RAG.

A projector or shared screen

The room must see your prompt window and the AI output side-by-side. Set the browser zoom to 125% before you start.

The meta-prompt open in a tab

Tab 4 in this page. You'll hand it out at the end so trainees can rerun the flow with their own ideas.

A confident "let's go" opening

Don't over-explain. Trainees learn faster watching you fumble a prompt and fix it than by hearing theory upfront.

Why this ordering works: we start broad ("here are three fuzzy ideas") and progressively add structure (scoring rubric → risks → HTML). Each prompt narrows the funnel while keeping the human in the loop. It's the exact pattern to teach — trainees can reuse it for any decision.
Rehearsal tip

Run the whole seven-prompt sequence solo the day before, on the exact machine and browser you'll present with. You'll catch Wi-Fi lag, model latency quirks, and the one prompt that produces surprising output. Then jump straight to Tab 2.

The seven prompts, one at a time

Step through the exact prompts to paste into Quick, what each returns, and why we prompt in this order. Click a number below to jump; use ◂ / ▸ to walk in sequence.

Prompt walk-through · live in Quick chat Step 1 of 7
The pattern behind these prompts

Notice the arc: frame → score → drill → challenge → pick → HTML. Every step adds one dimension. Trainees who get this arc can reuse it for anything — hiring calls, feature prioritisation, vendor selection. Teach the pattern louder than the prompts.

The HTML proposal Quick generates

What lands after Prompt 7. Quick returns a self-contained HTML file — trainees can download, share, or paste into an email. This is the "wow" moment; make sure everyone sees it.

Why HTML, not Markdown or a slide?

HTML is the most portable rich format on the planet. It renders in every browser, every email client (Gmail, Outlook), and every doc tool. When you ask Quick for an HTML brief, you get styled headings, tables, cards — no fussing with slide software, no losing formatting in a copy-paste. It also travels better than a link to a Google Doc that half the room can't open.

The exact prompt (see Step 7 in Tab 2) instructs Quick to build a single-file HTML brief with inline styling — one file to send, everything included.

Preview · what your trainees will see

pindai-product-brief.html · saved · ready to share

PINd.AI · Product Bet Recommendation

DRAFT · GENERATED BY AMAZON QUICK · HUMAN REVIEW PENDING

Executive summary

Across the three candidate ideas, Grounded Compliance Copilot ranks highest on a weighted score of market pull, feasibility with our current team, and 12-month cost-to-launch. We recommend focusing the next quarter's build on it, with a 2-week discovery spike first.

Idea scoreboard

#3 · 6.8/10
AI Meeting Summariser
Market 7 · Feasibility 9 · Cost 5
#2 · 7.4/10
SME Loan Q&A Assistant
Market 8 · Feasibility 7 · Cost 7
#1 · 8.6/10 · Pick
Grounded Compliance Copilot
Market 9 · Feasibility 8 · Cost 9

Recommendation rationale

DimensionWhy it wins
Market pullCompliance teams already ask for "AI for policy Q&A" — three inbound requests this month alone.
FeasibilityReuses our existing RAG stack; no new model training required.
Cost-to-launchEstimated 8-week MVP; two engineers plus one design partner.

Top three risks

  • Data access: customer compliance corpora are sensitive — need a written data-handling plan before pilot.
  • Answer accuracy: hallucinations in a compliance context are unacceptable — require citations + human review for every answer.
  • Sales cycle: compliance buyers move slowly; a 6-month pilot is normal.
Trainer note: the preview above is illustrative — Quick's real output will match the shape but the wording and scores will differ each run. That variance is a teaching moment: LLMs don't produce identical outputs, and that's fine when the pattern is right.
Save + share the output

After Quick generates the HTML, hit "Download" in the chat to save a .html file. It opens in any browser, prints cleanly to PDF, and pastes into Gmail as rich content. This is the artefact your trainees will forward to their team the same day.

Reusable meta-prompt

Hand this to your trainees at the end of the session. Fill in the four bracketed slots, paste into Quick, and the whole ideation runs in one shot instead of seven prompts.

When to use the meta-prompt vs the seven-step flow

The seven-step flow is for learning — every step teaches a prompting habit (add constraints, ask for scoring, challenge assumptions, request format). The meta-prompt is for doing — once you know the pattern, you don't need to walk through it every time.

Teach the seven steps in the room. Send the meta-prompt in the follow-up email. That way trainees can rerun the flow next Monday with their own three ideas.

You are a strategic product-ideation partner for an AI-first startup. CONTEXT - Company: [COMPANY NAME + one-sentence description] - Constraints: [team size / budget / timeline] - Decision maker: [e.g. founder team, product committee] INPUTS - Three candidate product ideas we're weighing this quarter: 1. [Idea 1 · one line] 2. [Idea 2 · one line] 3. [Idea 3 · one line] TASK 1. For each idea, score 1–10 on: market pull, feasibility with our team, 12-month cost-to-launch. Compute a weighted total (market 40% · feasibility 30% · cost 30%). 2. Rank the ideas from strongest to weakest with a one-line justification each. 3. For the top-ranked idea, list the top three risks and one concrete mitigation for each. 4. Draft a set of next-step actions for the next 7 days (max five bullets, each with a named owner slot). 5. Package everything as a single-file HTML brief — inline CSS, one file, no external assets. Suitable for emailing to a leadership audience. Include: title, executive summary, scoreboard, rationale table, risks, next steps. RULES - Every claim must be traceable to my inputs above — if you're inferring, say "assumption:". - If any input is missing or ambiguous, ask me one clarifying question before proceeding. - Keep the tone confident but calibrated — this is a decision aid, not a marketing pitch. Output the HTML file directly in the chat, ready to download.

What each slot changes

SlotWhat to writeExample
COMPANY NAMEYour one-sentence identity"PINd.AI, an AI-first fintech tooling startup"
ConstraintsTeam + money + time boundaries"5 engineers · $200k budget · Q1 launch"
Decision makerWho reads the brief"Two co-founders + one investor advisor"
Ideas 1–3One-line pitches"An AI compliance copilot grounded in our customer's policy library"
Teaching angle

Point out the four constraint types baked into the meta-prompt: context, inputs, task, rules. That structure is transferable — it works for pricing analysis, hiring shortlists, feature prioritisation, vendor selection. The template is more valuable than the specific words in it.

Human still decides. Quick's scoring is a starting point, not a decision. The point of the brief is to surface reasoning — the leadership team must read it, argue with it, and sign off. Bake that message into how you present the output.

Run this in a room · 30-minute playbook

Timing, cues, transitions. Follow the row-by-row playbook — it keeps you honest on the clock and gives you exact things to say.

Room setup · 5 min before start

Quick chat open at 125% browser zoom · this page open on a second tab, ready to switch to the HTML preview · scenario notes on paper next to you · timer visible.

00:00OpenOne line: "You've watched me explain Quick. Now let's actually build something with it — a real product decision in 30 minutes." Then paste Prompt 1.
02:00Prompt 1–2 landRead the AI's response out loud with slight commentary. Trainees are learning to read AI output, not just consume it.
08:00Prompt 3–4 (the drill)Slow down. Ask the room: "Which score do you disagree with?" — the point isn't to argue, it's to model the mindset of challenging AI output.
16:00Prompt 5 (challenge)This is the crux of the whole demo. Ask Quick to argue against its own top pick. When it does, say: "That's why we don't skip this step — the AI's own doubts are useful."
22:00Prompt 7 (HTML)Cue the drum-roll energy. When the HTML lands, download it and open in a new tab so the whole room sees it rendered — not raw markup.
26:00Debrief · 3 minAsk two questions: "What surprised you?" and "What would you change about how I prompted?" — the second question is where the learning consolidates.
29:00Hand-offPoint to Tab 4 (meta-prompt). Tell trainees: "You have this in your follow-up email. Rerun the flow on Monday with your own three ideas."

If something goes wrong

SymptomWhat to do live
Quick is slow — the response is taking 30+ secondsNarrate. "Notice it's still thinking — this is Research doing its work, not a bug." Talk through what you'd expect to see. Never fill silence with apology.
Output is generic / low-qualityPerfect teaching moment. Ask the room: "What extra constraint would make this better?" Take a suggestion, add it to the prompt, rerun. Trainees learn faster from a fix than a first-time-perfect demo.
Quick refuses (Guardrails or content policy)Also a teaching moment. "See — Guardrails do their job. Let me rephrase." Rephrase to remove the trigger and continue. Don't dwell.
HTML output is brokenSay "the AI got the wireframe right, the styling didn't render — let me ask it to fix that inline". One follow-up prompt: "regenerate with all styles inline and no external CSS references."
Wi-Fi drops mid-demoSwitch to this page's HTML preview (Tab 3) and walk the room through the shape while you reconnect. The preview is your backup.
The one thing to leave in the room

Trainees will forget most of the specific prompts by next week. What they'll remember is the arc: frame the decision, score with a rubric, challenge the AI's confidence, ask for a shareable format. Repeat that arc in your closing sentence. Everything else is scaffolding.