Outsource the heavy thinking. Keep the choosing. A tiny, repeatable loop beats infinite scrolling on a foggy day.

Quick Path ⚡

• Pick one domain. Books, groceries, or shows. Start narrow.
• List 20 real examples. Add why you liked it in 3–8 words.
• Paste the Prompt Kit. Run Analyst → Evaluator → Summarizer → Ranker.
• Choose one pick. Time‑box 5 minutes. Stop after the first good answer.
• Log the result. Save the pick + a one‑line reason to your notes.

Full Path 🛠️

Phase 1 — Prepare your seed data (10–15 min · Easy)

• Choose a domain: streaming, books, groceries, or gadgets.
• Gather 20–40 items you actually chose or enjoyed.
• Add micro‑reasons after each item: “reduced strain,” “fast weeknight,” “cozy mystery.”
• Keep it local in a note first. Paste to AI only what you’re comfortable sharing.

Alt text: Prepare seed data — past items and short reasons listed — notes app.

Optional exports (Unverified):

• Amazon: Order history → copy item names; add your reasons manually.
• Spotify/Goodreads/Netflix: Export lists if available; otherwise copy favorites by hand.

Phase 2 — Run the Analyst (3 min · Easy)

Prompt:

Analyze my history and infer 3–5 preference patterns.
Look across style, function, price, constraints (MS/energy), and contexts.
Return bullets: Pattern | Evidence | Why it matters for me.

Outcome: A simple preference map you can sanity‑check.

Alt text: Run Analyst — patterns extracted from list — AI chat window.

Phase 3 — Evaluate candidates (4–6 min · Easy)

• Collect 6–12 options you’re considering now (links or brief blurbs).
• Run Evaluator with a strict rubric and short outputs.

Prompt:

Using the patterns above, score each candidate 1–10 with 1–2 reasons.
Add flags: comfort (C) vs novelty (N); physical/energy fit (Y/N).
Return a table: Item | Score | Reasons | Flags.
Limit to 25 words per row.

Alt text: Evaluate candidates — table of scores and reasons — AI chat window.

Phase 4 — Summarize & rank (3–4 min · Easy)

• Summarize themes from items scoring ≥7.
• Rank top 3 with trade‑offs.

Prompt:

From items ≥7, extract the common themes.
Then rank the top 3 for me with trade‑offs.
Return: Rank | Pick | Why it fits | Trade‑off | Starter step (<5 min).

Alt text: Summarize & rank — top picks listed with trade‑offs — AI output.

Phase 5 — Decide and log (3 min · Easy)

• Pick the top fit or the best comfort choice if energy is low.
• Log one line: “Chose ___ because ___.”
• If unsure: Start the 5‑minute Starter step and reassess.

Alt text: Decide and log — choice recorded — notes app.

Filled Example: Dummy Data → Result ✅

Seed data sample you can paste

The Bear (TV) — loved tight pacing; low emotional labor on weeknights
Kindle Paperwhite — reduces eye strain; reads in bed
Ergonomic mouse pad — wrist relief; boosts coding comfort
OXO 12" tongs — one‑hand use; durable
Aldi veggie dumplings — 8‑min dinner; low cleanup
Anker USB‑C hub — reliable brand; ports for travel
OXO measuring cup set — clear markings; dishwasher‑safe
Fjällräven Kånken — light; structured; holds laptop
Book: Project Hail Mary — propulsive; smart but warm
Book: Cozy mystery series — gentle stakes; bedtime friendly
Sennheiser HD 560S — neutral sound; comfy pads
Govee LED strip — easy install; room mood
Hario V60 — ritual; bright taste; weekends only
Instant Pot — set‑and‑forget; batch soups

Analyst output (expected)

• Functional comfort bias: Items reduce strain or time; evidence: ergonomic gear, dumplings, Instant Pot.
• Trusted brands for tech: Anker/Sennheiser; avoid flashy no‑names.
• Weeknight energy guardrails: Low cleanup, gentle stakes, short episodes.
• Ritual pockets: Coffee gear and weekend projects okay when energy permits.
• Sound + reading care: Eye strain and fatigue matter; prefer neutral audio and e‑ink.

Candidates to evaluate (example)

• TV: Slow Horses, Only Murders in the Building, Andor.
• Gadgets: Logitech MX Master 3S, Random RGB keyboard, Generic USB hub.
• Groceries: Trader Joe’s gyoza, Fresh meal kit, New spicy ramen.

Evaluator table (example)

Summary + ranking (example)

Themes: Comfort + function wins. Trusted brands for tech. Weeknight‑friendly media and meals.

If you do only one thing now, do: Order or test MX Master 3S.

The Prompt Kit 🧰

1) Analyst

You are my personal preference analyst.
From my history, infer 3–5 patterns with evidence and why each matters for me (MS/energy constraints included).
Return: Pattern | Evidence | Why it matters.

2) Evaluator

Score each candidate 1–10 against my patterns.
Add flags: comfort (C) vs novelty (N); physical/energy fit (Y/N).
Return: Item | Score | 1–2 Reasons | Flags.
Limit 25 words per row. Be blunt.

3) Summarizer

From items scoring ≥7, extract common themes in 3 bullets.

4) Ranker

Rank top 3 with trade‑offs and a 5‑minute starter step for each.
Return: Rank | Pick | Why it fits | Trade‑off | Starter step.
If nothing scores ≥7, say “No strong recommendation.”

Troubleshooting 🧰

1. Generic outputs.

   • Fix: Add your why after each history item. Specify constraints: budget, time, strain.

2. Over‑novel picks.

   • Fix: Add a comfort/novelty mix rule: 2 comfort to 1 novelty.

3. Too many ties.

   • Fix: Break ties with daily comfort gain and setup friction scores.

Friction Fix 🔧

• One automation: Text expander /arag → pastes the four prompts.
• One simplification: Limit candidate lists to 9 items. Stop at first good answer.

Next Action ▶️

Open a note titled “ARAG — Books”. List 20 titles you finished and add why you liked each in 3–8 words. Run the Analyst prompt.

Privacy & Care ♿

• Redact sensitive data before pasting. Use placeholders like [CLIENT], [ADDRESS], [LINK].
• Prefer local tools for confidential material. Share only what you’re comfortable with.
• Voice‑only path: Dictate items and reasons; ask AI to clean the list.
• Assistive tip: Use Do Not Disturb and a visible 5‑minute timer during decisions.

Handling “No strong recommendation” 🧯

• Maintenance mode: Choose a comfort pick you already like.
• Or rest: Defer the choice. Schedule a fresh 10‑minute slot tomorrow.

Sources

• Inspired by multi‑agent reasoning patterns from retail research.
• Works with Claude or any modern LLM that follows structured prompts.