12 Prompts I Still Use Every Day — Even in the Age of Agents
✨ A curated collection of the prompts that have genuinely helped me, organized into five areas: clarifying questions, learning, problem-solving, decision-making, and knowing yourself.
A couple of days ago I shared a deep-thinking prompt, and to my surprise, the feedback was great 🎉. Several people also asked whether I could gather everything I've written over the years — plus the prompts I find most useful — into a single collection.
So I dug back through my archives and found that, over the years, I really had written quite a lot of them. The only problem: they were scattered across dozens of different articles, and a fair number had grown stale — no longer suited to what today's models can do.
So I reworked the whole thing from scratch. I cut some, refined some, and added some. What remains are the 12 prompts that genuinely work for me — the ones that have truly helped. 💪
They fall into five areas: clarifying questions, learning, problem-solving, decision-making, and knowing yourself.
Every one can be copied and used on its own, requires no skill or plugin, and works with all the AI tools available today 🤖. I've also turned everything into templates — just replace the content inside 【】 with your own information. And if you have raw material like documents or notes, drop those in too. These days, there's no such thing as too much context.
That's the shape of it. Let's get started. 🚀
🧠 Part 1 · Clarifying Questions
Asking questions is probably the most fundamental part of how we interact with AI. This section is dedicated to helping you get your question straight — because only when you know what you're really trying to ask does everything else become possible.
At the foundation, I strongly recommend reading a classic: Asking the Right Questions by Neil Browne. There are also skills for this — the famous "grill me" and others — but they all boil down to the same thing: how to ask more precisely.
Personally, I don't love using skills inside chat. Some prompts keep interrogating you with endless follow-ups and it gets exhausting 😅. So the Socratic method is my most-used approach.
1. Socratic Questioning 🔍
Human beings are, sometimes, quite messy. What we say out loud and what we actually mean are often not the same thing. Socratic questioning fixes that: through follow-up questions, it helps you think more clearly while also pulling out the context the AI actually needs.
My confusion is: 【describe as specifically as possible what happened, how you understand it, and where you're stuck】. Don't give advice yet. Please run a Socratic diagnosis on me — through at most 6 questions, help me find the question that's actually worth answering. Follow these rules: 1. Ask only one question at a time, and decide the next based on my answer — don't hand me a full questionnaire upfront. 2. First distinguish whether what I'm saying is a verifiable fact, an interpretation, a value judgment, or a goal I want to achieve. 3. Check whether key terms are vague, what assumptions I'm taking for granted, where the evidence comes from, whether there's a counter-interpretation, and what it means if the conclusion holds or falls apart. 4. Before each question, state in one sentence what my previous answer updated in your judgment. 5. Only ask questions that could change the conclusion. Stop as soon as you have enough — no need to force all 6. After the diagnosis, summarize: 1. The question I originally asked 2. The problem I'm actually trying to solve 3. The facts already confirmed 4. The assumptions still unverified 5. The key variable most likely to change the conclusion 6. A precise, specific, actionable new question Wait for me to confirm this new question, then give your judgment, reasoning, and next step.
This prompt helps you answer: "What is the essential question I'm actually trying to ask?" Only once the question is clear can the whole process move forward. ✅
📚 Part 2 · Learning
When we face something unfamiliar, we tend to have a few distinct needs. After reviewing my own patterns, I've distilled them into four:
- 😵 "I don't understand this concept at all" → Two-Layer Explanation
- 🤩 "This piece of work is impressive — I want to learn from it" → Reverse Deconstruction
- 🔬 "I want to study something deeply and systematically" → Vertical-Horizontal Analysis
- 🤔 "Is the claim in this material actually true?" → Fact-Checking
2.1 Two-Layer Explanation 🎓
A lot of people tell AI: "Explain it to me like I'm a child." It's easy to understand that way — but it's also easy to get stuck at "I kind of get it" 😶. The analogy sticks, but the real mechanism stays foggy.
So I prefer to ask for two separate explanations — once from a beginner's angle, once from an expert's — which helps me learn far more deeply.
What I want to learn is: 【fill in the concept or question】. Explain it in two layers: Layer 1 — Beginner's version. Use everyday language and a concrete example so that someone with zero background can follow. Layer 2 — Expert version. Use precise terminology and explain the core mechanism, the limits of when it applies, and common misunderstandings. Finally, summarize: 1. A mapping between the beginner phrasing and the professional terms 2. The place where I'm most likely to misunderstand 3. Three questions to check whether I've truly understood
I use this constantly during Vibe Coding sessions when I hit unfamiliar terms — databases, network security, whatever. Learn as you go. 💡
2.2 Reverse Deconstruction 🔧
Sometimes you see a great product or website and want to understand what actually makes it work. That's when you use reverse deconstruction.
The excellent example I want to deconstruct is: 【paste a product page, webpage, proposal, process, dashboard, or any finished work】. What I want to learn is: 【fill in what you hope to take away from it】. First, explain in one sentence what problem it solves, then reverse-engineer why it works. Focus on: 1. Who it serves and what its goal is 2. What structure or process it uses 3. Which key choices create the gap in quality 4. What its definition of "done" is 5. Which patterns are transferable, and which details only fit this specific case Finally, give me: 1. Three to five reusable patterns 2. An operational checklist I can follow 3. One small exercise worth trying first
2.3 Vertical-Horizontal Analysis 📊
Back when I worked in public funds, this was a classic framework I used constantly to analyze companies and new concepts. The core is just two axes.
The vertical axis looks at how something became what it is today. The horizontal axis looks at how it compares against its competitors. Crossing the two reveals the historical choices — and how they evolved into the present. Pair this with your AI's deep-research feature for full power. ⚡
Research subject: 【fill in a product, company, person, technology, industry, or event】. Please use the vertical-horizontal analysis to produce a traceable deep-research report. Cutoff date is today. Vertical analysis: 1. Under what background and need was it born, and who were the key drivers? 2. What important turning points, successes, and failures did it go through? 3. Which early choices became today's capabilities, path dependencies, or liabilities? Horizontal analysis: 1. Pick the objects most worth comparing, and explain why you chose them. 2. Compare strengths, weaknesses, and uniqueness across a unified set of dimensions. 3. Explain why users, customers, or the market chose it — and why they abandoned it. Combine the two axes and judge: 1. How will the capabilities, path dependencies, and constraints formed in the past affect the future? 2. What are the three most likely future paths? 3. What are the preconditions and warning signals for each path? Evidence rules: 1. Prioritize primary sources — official materials, raw data, papers, financial reports, and interviews. 2. Cite the source and date near every important conclusion. 3. Keep facts, inferences, and opinions separate. 4. Present conflicting information side by side; when evidence is missing, write "not yet verified." Output in this order: core conclusion → key timeline → horizontal comparison table → detailed analysis → future judgment → open questions. Target length: 10,000–30,000 words, in plain language.
If you only have half an hour to build a basic mental model of something unfamiliar, this framework alone is more than enough. ⏱️
2.4 Fact-Checking 🕵️
We all know AI hallucinates — less than it used to, but still 😬. And human hallucination is often even worse these days. There's a saying: AI has helped a lot of people climb to the peak of the Dunning-Kruger curve. So whether it's human or AI content, following Descartes' instinct, we should doubt everything as much as possible.
The claim I want to verify is: 【paste an opinion, conclusion, data point, or proposal】. First break it into: 1. Facts that can be externally verified 2. Conclusions drawn from those facts 3. The value judgments embedded within For the factual part, search online to verify the source, sample, timeframe, and full context, and label each as: 1. Confirmed 2. Basically holds, but needs narrowing 3. Disputed 4. Insufficient evidence 5. Clearly wrong Assuming the relevant facts hold, continue checking: 1. Do these facts actually support the current conclusion? 2. Are there unverified assumptions hidden inside? 3. Is correlation being confused with causation? 4. Are other explanations or key information being omitted? 5. Under what conditions does the conclusion hold or fail? Finally, output: 1. Which facts are trustworthy and which need correcting 2. The most critical flaw in the chain of reasoning 3. The most reasonable strengthened version 4. How much I can currently believe this
Works not just for facts — you can also use it to audit opinions, proposals, and arguments of all kinds. 🎯
🛠️ Part 3 · Problem-Solving
By now we know how to ask and how to learn. Now it's time to actually solve problems. I've broken this into three approaches: Expert Panel, First Principles, and Cross-Domain Borrowing.
3.1 Expert Panel 👥
It used to be trendy to open a prompt with "You are a world-class expert with 20 years of experience." But many problems actually depend on multiple experts working together 🤝. So I prefer to have the AI assemble a genuinely complementary panel — and then let them challenge each other.
My problem is: 【fill in the problem, known facts, goal, and real-world constraints】. Don't jump to a solution yet. First, select three genuinely complementary professional perspectives for this problem, and explain why each one is necessary. Have each perspective answer: 1. How does it reframe the problem? 2. What solution path does it most recommend? 3. What risk is most likely to be overlooked by the other perspectives? 4. What new evidence would make it change its judgment? Then let the three perspectives challenge each other and find: 1. The facts they all agree on 2. The genuine disagreements 3. The different assumptions behind those disagreements Finally, synthesize and output: 1. The most recommended overall solution 2. The conditions under which it applies 3. The biggest risk 4. The exit condition 5. The first step Don't pick three near-identical personas, and don't imitate or fabricate the views of real people. When information is insufficient, ask me only the single most critical question first.
The most crucial step: making them challenge each other. The real insight is almost always found inside the disagreement. 💥
3.2 First Principles ⚙️
A universal prompt — I use it constantly during Vibe Coding. It's best for breaking path dependency and getting back to the essence of a problem. When your solution has been patched so many times it's barely recognizable, first principles lets you tear it down and rebuild from the ground up. 🏗️
The problem I want to solve is: 【fill in your problem】. Use first principles to strip it back to its foundation, and distinguish: 1. The basic facts that are confirmed and cannot be bypassed 2. The assumptions I've accepted out of habit but never verified 3. The goal I actually want to achieve 4. The resources and constraints in reality Set aside industry norms and ready-made solutions. Starting only from the basic facts, the goal, and the constraints, re-derive viable paths. Finally, output: 1. The parts of the original solution that only patch the surface 2. The new path re-derived from first principles 3. The premise on which this path holds 4. The first step to verify it
Especially powerful when restructuring an organization, designing product architecture, or debugging complex systems.
3.3 Cross-Domain Borrowing 🌐
First principles takes you back to the essence. Cross-domain borrowing expands your view outward — helping you find solutions from other fields that map onto your problem. The perspective is more divergent. 🔭
My confusion is: 【explain the background, current approach, real-world constraints, and the specific sticking point】. First strip away the industry jargon and abstract it into a problem humans might face in any other field. Then find: 1. The underlying structure of the problem 2. The real core contradiction 3. The reason ordinary solutions fail Then draw on historical cases and at least three fields that are far apart from each other. For each case, explain: 1. What problem that field faced 2. What mechanism it used to solve it 3. How it resembles my problem 4. Which parts can transfer 5. Under what conditions it would fail Finally, pick the three mechanisms most worth borrowing, translate them into solutions that fit my current situation, and recommend one low-cost, reversible experiment to try first.
If your problem feels unsolvable even after going back to first principles, look sideways 👀. There's a good chance another field solved your exact problem a decade ago — and that's often where the breakthrough comes from. 💡
⚖️ Part 4 · Decision-Making
Everything above helps you find ideas, solutions, and more information. But sometimes you'll still find two answers equally convincing — so which one do you choose? 🤷 That's when you actually have to decide. For decisions, I have two prompts I rely on.
4.1 Two-Way Steelman Argument 🥊
Some people say this is just like "grill me." I'd say the underlying logic is different: one helps you ask better questions; this one is for when you already have two answers and need to choose between them. If you're hesitating, give it a try.
The decision I need to make is: 【lay out the problem, the two options, the goal, and the real-world constraints】. Don't rush to answer, and don't assume I've already thought it through. First run a two-way steelman argument: 1. Restate, in the most complete and forceful way, the choice I actually need to make. 2. Separately give the strongest reasons supporting each direction — the conditions where each applies, the biggest upside, the biggest risk, and the hardest objection to answer. 3. Identify the genuine disagreement between the two sides, the key variable most likely to change the conclusion, and what information is still needed. 4. Ask me only one question — the one most likely to change the conclusion. After I answer, give a clear judgment, reasoning, conditions of application, and next step.
4.2 Replace Daydreaming with a Minimum Experiment 🧪
In the real world, some decisions won't get any clearer no matter how much you debate them on paper 📄. At that point, you need to take the first step — try it, and see what the real-world feedback looks like. This prompt helps you design exactly that.
What I'm torn about is: 【fill in your choice or idea】. First identify the three assumptions most in need of verification behind this decision, then pick the one most likely to change the final conclusion. Around that assumption, help me design a minimum experiment that is low-cost, reversible, and can be completed within 【7 days, or a period you can accept】. Make it clear: 1. What specifically to do 2. How much time and resources it requires 3. What metrics to observe 4. What result means "keep going" 5. What result means "stop" 6. What new information I'll gain after the experiment Finally, tell me the first action I can start tomorrow.
🪞 Part 5 · Know Yourself
When it comes to exploring one's own life, I have two prompts I'm genuinely proud of 🌟. They've spread across communities and been deeply useful to me personally. The foundation of a meaningful life, I think, is the most important thing: knowing yourself.
5.1 Unearthing Hidden Talents 💎
This one suits people who still hold expectations for the world and want to find their natural gifts — and equally those who feel like they just don't have any talent and have started to doubt themselves 😔. The AI will take you through deep questioning, connect the experiences in your past that seemed unrelated, and piece together a personal talent manual just for you.
This one needs patience ⏳ — it can easily take half an hour or more. But the more honestly and specifically you answer, the more useful the result will be. Trust me on that.
# Role: Deep Talent Excavator ## Role You are a senior career counselor familiar with the Gallup StrengthsFinder system, flow theory, and Jungian psychology. You believe talent is a transferable foundational ability that often hides in a person's quirks, flaws, envy, unconscious competence zones, and energy patterns. ## Goal Through multiple rounds of deep dialogue, help the user find talents that have been overlooked or suppressed, and ultimately produce an extremely detailed, professional, and empathetic "Personal Talent Manual." ## Core Principles 1. Anti-fatalism. Talent is not a fixed skill, and it doesn't expire with age. 2. Energy audit. Real talent tends to recharge a person. Things a person is simply good at but finds deeply draining need to be separated out. 3. The shadow is treasure. Flaws repeatedly criticized since childhood, quirks that are hard to change, and envy of others may be the flip side of suppressed talent. ## Dialogue Rules 1. Ask only one question at a time. Follow the rhythm: you ask → user answers → you briefly acknowledge → ask the next question. 2. Use Socratic follow-ups. Ask more about "how old were you then," "what exactly happened," "how did you feel," "why did you do it that way" — avoid labeling someone from a single sentence. 3. Stay warm, empathetic, and sharp. When you spot contradictions, pretense, or subconscious clues, point them out directly — but don't comfort with empty praise. 4. Every judgment must map to a concrete experience the user described. When evidence is insufficient, explicitly use "possibly" and keep asking. 5. A maximum of 10 main questions in total. You may reorder or add follow-ups based on answers, but must cover the four main threads below. ## Required Threads 1. Before age 16: what things would you do obsessively even when no one asked? What recurring "stubborn flaws" were you repeatedly criticized for and could never fix? 2. In adult work and life: what things make you think "do I even need to learn this?" while people around you generally find difficult? Locate the unconscious competence zone. 3. What things leave you physically tired but mentally euphoric afterward? What things do you do well but find clearly draining? 4. Who have you strongly envied, or what kind of life have you admired? Keep probing into what it is about that person or life they truly desire. ## Output When the information is rich enough, output a "Personal Talent Manual" of around 10,000 words. Structure it freely based on the answers, but it must cover: 1. The foundational talents best supported by evidence, and the chain of experiences behind each 2. The shadow side of each talent, and why it was mistaken for a flaw in the past 3. The user's energy map, unconscious strength zones, and high-drain zones 4. The environments where these talents are most likely to shine and most likely to fail 5. The work styles, collaboration styles, career directions, and real-world limitations that suit them 6. Low-cost experiments to try over the next 30 days, using real-world feedback to keep validating these judgments ## Begin Warmly, professionally, and plainly explain the upcoming process, roughly how long it will take, and what you hope to achieve. Tell them: "Talent never expires — we're just going to find your foundational talents." Then begin with the first question.
5.2 Life Design 🗺️
Unearthing hidden talents answers "what do I actually have?" — it looks mostly backward. Life Design answers the next question: "where can I go from here?" It looks to the future. 🔮
This prompt is based on Stanford's life-design methodology. I think it's one of the best tools available for thinking seriously about your future.
# Role: Life Designer ## Role You are a senior life designer familiar with the Stanford life-design method, flow theory, and positive psychology. Your job is to accompany the user in treating their present life as a project that can be repeatedly designed and cheaply tested — first see where they are, then find the direction, and finally actually try out the possible paths. ## Goal Through multiple rounds of deep dialogue, help the user see their true current position, separate the "gravity problems" that can't be solved from the real problems that can be designed, and ultimately generate three completely different, equally serious five-year life versions, plus prototype actions they can start right away. The final product is an extremely detailed, warm yet sharp "Personal Life Design Blueprint." ## Core Principles 1. Life is a design problem with no single correct answer. It needs lots of trying, prototyping, and adjusting as you go. 2. Reframe the problem. Many people keep solving the wrong problem; finding the real problem matters more than rushing to an answer. 3. Distinguish gravity problems. Things like age, natural laws, and the reality of an entire industry can't be directly changed — accept them first, then turn attention to the designable parts. 4. Quantity contains quality. Good choices come from having enough choices. 5. Passion is often the result of action and feedback. The user doesn't need to find their destined passion before they're allowed to start. 6. Life is an infinite game. Any prototype leaves information behind, so people can be immune to failure. ## Dialogue Rules 1. Ask only one question per round, following the rhythm: you ask → user answers → you briefly and sincerely acknowledge → ask the next question. 2. Use Socratic follow-ups — ask more about specific events, feelings at the time, and actions; avoid jumping to conclusions. 3. Stay warm and accepting, while sharply pointing out logical gaps, self-imposed limits, and the gap between language and actual behavior. 4. Actively distinguish gravity problems from designable real problems. Accepting reality is not the same as giving up; seeing the boundaries clearly is itself part of designing. 5. Don't judge the user's choices, and don't make decisions on their behalf. 6. Keep the main questions to 6–9 in total, with flexible reordering and follow-up depth based on answers. ## Question Flow ### Phase 1: You Are Here 1. Ask the user to score health, work, play, and love on a 0–10 scale, and explain which one is flashing red. (Health covers body, emotion, and mind; play means things done purely for joy; love emphasizes two-way relationships.) 2. Ask what life problem they're most anxious about and most want to solve. Judge whether it's a designable real problem or an unchangeable gravity problem. If the latter, point it out gently and guide them to reframe it into an actionable problem. 3. If the user is stable, ask for consent first, then invite them to do a reverse projection: imagine an ordinary Tuesday five years from now if nothing changes, then pull that picture to ten years out. Help them see the cost of staying the same. Skip this step if the user seems at a low point or emotionally fragile. ### Phase 2: Your Compass 1. Ask about their workview: why they work, and how work relates to money, others, and the world. 2. Ask about their lifeview: what would make them feel this life wasn't lived in vain, and how they want to connect with family and the larger world. 3. Compare workview and lifeview for consistency, and point out conflicts, compromises, and their true north. ### Phase 3: Wayfinding 1. Ask them to recall recent or past flow moments, probing what exactly they were doing, with whom, and in what environment. 2. Distinguish the things that recharge them, the things that drain them, and the things they're good at but don't love. ### Phase 4: Getting Unstuck and Creating Possibilities 1. Ask whether there's a fixation or plan that has long since stopped working but they won't let go of. Find what they're really trying to hold onto behind that anchor. 2. Accompany them in generating three completely different five-year life versions: — Version 1: the path they're already on or have been weighing for a long time. — Version 2: the path they'd choose if Version 1 vanished tomorrow. — Version 3: the life they'd truly want if money and others' judgments didn't matter. 3. All three versions must be Plan-A options they genuinely consider — none can be a filler backup. ## Output When the material is rich enough, output an 8,000–12,000-word "Personal Life Design Blueprint" that naturally covers: 1. "You Are Here": interpreting the four dashboards and pointing out the real imbalances and long-ignored parts. 2. "The Real Problem": reframing the user's original worry, separating gravity problems from designable problems. 3. "Your Compass": distilling workview, lifeview, and the coherence between them. 4. "Your Energy Map": summarizing flow, recharge zones, high-drain zones, and the environments future design should favor. 5. "Three Odyssey Plans": each with a short, powerful title, a five-year timeline, two to three questions to be verified, plus assessments on resources, enjoyment, confidence, and coherence. 6. If the user already clearly leans toward one version, break it into the core question to verify this quarter, a prototype they can make within a month, small moves they can advance daily, and the bottom line they'd never sacrifice. 7. "Prototype Action List": design one life conversation, a one-day-to-one-week prototype experience, and the first small step they can take this week. 8. "Immunity to Failure": remind the user that all three versions can be tested and adjusted. Even if a prototype doesn't work, it leaves useful information for the next step. ## Begin Open warmly, professionally, and empathetically. Explain the basic idea of this method, roughly how long it will take, and what you hope to achieve. Tell the user they don't need to know what they love first — we'll find it gradually through action, conversation, and feedback. Then begin with the first question.
Both of these prompts involve extended back-and-forth. When you use them, commit to seeing them through. 🏁
💬 One Last Thing
Above are the 12 prompts I've reworked — the ones I genuinely find most useful.
Some people might say: we're already in the age of agents and you're still reheating leftovers? Four years in and you're still sharing prompts? 😂
Sometimes I'm conflicted too. AI is moving so fast that the people at the frontier and the people just getting started are separated by what feels like a galaxy 🌌. What should I share so that people at every level find at least something useful in it? It's hard.
But no matter how things change, I still believe prompts are the cornerstone of how we use AI. How a person asks questions, how they judge truth from falsehood, how they understand the world, how they know themselves — these things never go out of date. ❤️
The question of life can only be answered by you, yourself.
As AI comes to resemble gods in the firmament, I think what it means to be human is to stand before that power and still know what we want — and where we want to go. 🌟
And that, perhaps, is the final meaning of the prompt.