Skip to content
Friday, 20 November 2026
DeepInsight
Just inSheru Bhai, the Lion-Boss Meme, Now Has Its Own AI Video Studio
AI Tools

12 Prompts I Still Use Every Day in the Agent Era

Twelve copy-paste prompts across five situations: clarifying questions, learning, problem solving, decisions, and understanding yourself. Full templates included.

By Claude Carter16 min read
12 Prompts Still Worth Using in the Agent Era

A few days ago I published a prompt for deeper thinking and the response was better than expected. People asked whether I would collect the ones I had written over the years into a single set. So I went back through my archives, deleted what no longer works, tightened what was sloppy, and added a few missing pieces.

What follows is twelve prompts that have actually helped me, grouped into five situations: clarifying the question, learning, solving problems, deciding, and understanding yourself. Each one can be copied and used on its own, needs no skill or plugin, and works with whatever model you already use. Every prompt is a template: replace the bracketed text with your own details, and paste in your source material if you have it. These days, context is cheap.

I. Clarifying the Question

Asking a question is the most basic unit of working with a model. Before anything else, you need to know what you are actually asking — otherwise everything downstream is built on a guess.

1. Socratic interrogation

People are often muddled. What they say out loud is frequently not what they mean. This prompt uses follow-up questions to sharpen the problem and supply the context the model actually needs.

My confusion is: [describe as specifically as you can what happened, how you interpret it, and where you are stuck]. Do not give advice yet.

Run a Socratic diagnosis on me, using at most six questions, to find the question actually worth answering.

Follow these rules:
1. Ask one question at a time. Decide the next question from my answer; do not hand me a whole questionnaire up front.
2. Prioritize separating verifiable facts, my interpretation of those facts, value judgments, and the outcome I am hoping to achieve.
3. Check whether key terms are vague, which premises I assumed without stating them, where the evidence comes from, whether a competing explanation exists, and what would follow if the conclusion held or failed.
4. Before each question, state in one sentence what my previous answer changed in your assessment.
5. Only ask questions that could change the conclusion. Stop as soon as you have enough information; do not fill a quota of six.

When the diagnosis ends, lay out: 1. the question I originally asked; 2. the problem I actually want solved; 3. the facts already confirmed; 4. the assumptions still unverified; 5. the key variable most likely to flip the conclusion; 6. one accurate, specific, actionable new question.

Only after I confirm that new question should you give me your judgment, your reasoning, and the next step.

II. Learning

Faced with something unfamiliar, four needs come up repeatedly: "I do not understand this concept at all" (two-layer explanation), "this artifact is impressive and I want to learn from it" (reverse engineering), "I want to research a subject deeply and systematically" (longitudinal and lateral analysis), and "is this claim actually true?" (fact checking).

2. Two-layer explanation

Many people ask the model to "explain it like I am a child." That gets you comprehension fast and leaves you at the "I think I get it" stage: you remember the analogy, the mechanism is still fog. Ask for two passes instead, one naive and one expert.

The concept I want to learn is: [name the concept or question].

Explain it in two layers.

Layer one, the beginner version. Use everyday language and one concrete example, so someone with no background can follow it.

Layer two, the professional version. Use precise terminology, explain the core mechanism, state where it applies and where it breaks down, and name the common misconceptions.

Then give me: 1. a mapping between the beginner phrasing and the professional terms; 2. the places I am most likely to misunderstand; 3. three questions that test whether I actually understand it.

This is the one I lean on most while building software: unfamiliar database concepts, networking terms, security vocabulary. Explain at two layers, then keep moving.

3. Reverse engineering

When you see work that is genuinely good and want to know why, decompose it rather than admire it.

The excellent example I want to decompose is: [paste a product page, website, proposal, process document, dashboard, or other finished artifact]. What I want to learn from it: [state the skill you are after].

Start with one sentence on what problem it solves, then reverse-engineer why it works.

Focus on: 1. who it serves and what it is trying to achieve; 2. what structure or process it uses; 3. which specific choices create the quality gap; 4. what its standard of done looks like; 5. which patterns transfer and which details belong only to this case.

Finish with: 1. three to five reusable principles; 2. an execution checklist I can follow; 3. one small exercise worth trying first.

4. Longitudinal and lateral analysis

A framework I used constantly as a public fund analyst, and later wrote up separately. It has two axes. The vertical axis asks how something got here; the horizontal axis asks how it differs from its competitors today. Cross them, and you can see the choices made in the past and how they produced the present. Run this with a model's deep-research mode for full effect.

The subject is: [name a product, company, person, technology, industry, or event].

Use longitudinal and lateral analysis to produce a traceable deep-research report. Research cut-off: the day this is executed.

Longitudinal analysis: 1. what background and need gave rise to it, and who the key drivers were; 2. which turning points, successes, and failures it went through; 3. which early choices turned into today's capabilities, path dependencies, or baggage.

Lateral analysis: 1. choose the most meaningful comparators and justify the choice; 2. compare their strengths, weaknesses, and distinctiveness on a consistent set of dimensions; 3. explain why users, customers, or the market chose it, and why they left.

Then combine both axes: 1. how past capabilities, path dependencies, and constraints will shape the future; 2. the three most plausible future paths; 3. the preconditions and warning signals for each.

Evidence rules: 1. prefer primary sources — official material, raw data, papers, filings, interviews; 2. mark each important conclusion with its source and date; 3. keep facts, inferences, and opinions separate; 4. present conflicting information side by side, and write "not verified" explicitly when evidence is missing.

Output order: core conclusions, key timeline, lateral comparison table, detailed analysis, forward-looking judgment, open questions. Target 10,000 to 30,000 words, in plain language, without padding the text with source material.

5. Fact checking

Models still hallucinate, though less than they used to. Humans hallucinate more. This prompt is for holding both to account.

The claim I need to check is: [paste an opinion, conclusion, data point, or proposal].

First split it into: 1. facts that can be verified externally; 2. conclusions inferred from those facts; 3. value judgments embedded in it.

For the factual part, check sources, sample, timing, and full context online, and label each item: 1. confirmed; 2. broadly right but needs narrowing; 3. contested; 4. insufficient evidence; 5. clearly wrong.

Assuming the relevant facts hold, then check: 1. whether those facts actually support the conclusion; 2. whether unverified assumptions are hiding inside; 3. whether correlation is being confused with causation; 4. whether alternative explanations or key information were omitted; 5. the conditions under which the conclusion holds or fails.

Finally output: 1. which facts are trustworthy and which need correction; 2. the weakest link in the reasoning chain; 3. the most reasonable version after reinforcement; 4. how far I can currently trust it.

III. Solving Problems

6. Expert panel

"You are a world-class expert with 20 years of experience" was the old opening line. Most real problems need several experts working together, and arguing.

My problem is: [state the problem, the known facts, the goal, and the real constraints].

Do not give a solution yet. Choose three genuinely complementary professional perspectives for this problem, and explain why each is necessary.

Have each perspective answer: 1. how it redefines the problem; 2. the path it would most recommend; 3. the risk the other perspectives are most likely to miss; 4. what new evidence would change its judgment.

Then have the three perspectives challenge each other, and identify: 1. what all three agree on; 2. where the real disagreement is; 3. the different assumptions behind that disagreement.

Finally synthesize: 1. the recommended course of action; 2. its conditions of applicability; 3. the biggest risk; 4. the exit condition; 5. the first step.

Do not choose three near-identical roles, and do not imitate or invent the views of real people. If information is insufficient, ask me one critical question first.

The step that matters is mutual challenge. Real information tends to live inside the disagreement.

7. First principles

A general-purpose prompt, and one I use constantly in coding. It is for breaking path dependence and getting back to the actual problem. When a plan has accumulated patches, rebuilding from the ground up beats another layer of tape.

The problem I want to solve is: [describe your problem].

Break it down to the bottom with first principles, separating: 1. facts that are confirmed and cannot be bypassed; 2. assumptions absorbed by habit and never tested; 3. what I actually want to achieve; 4. the real resources and constraints.

Set aside industry convention and existing solutions, and re-derive viable paths from basic facts, the goal, and the constraints alone.

Finally output: 1. the parts of the current plan that are only patching the surface; 2. the new path derived from basic facts; 3. the preconditions that path requires; 4. the first step to test it.

8. Cross-domain borrowing

Where first principles returns to the essence, cross-domain borrowing widens the field of view to solutions that already exist somewhere else.

My sticking point is: [explain the background, what you currently do, the real constraints, and the specific blocker].

First strip away industry jargon and abstract it into a problem that humans in other fields might also face. Identify: 1. the underlying structure of the problem; 2. the real core tension; 3. why ordinary solutions fail.

Then look for problems with a similar underlying structure, drawing on historical cases and at least three fields that are genuinely distant from yours.

For each case, explain: 1. what problem that field had; 2. what mechanism it used; 3. where it resembles my problem; 4. which parts can be transferred; 5. the conditions under which transfer fails.

Finally pick the three mechanisms most worth borrowing, translate them into solutions suited to my situation, and recommend one low-cost, reversible experiment to try first.

IV. Decisions

Sometimes both answers are defensible and you still have to choose. That is a different problem from understanding, and it needs a different tool.

9. Double steel-man

People confuse this with interrogation prompts. They are not the same: one helps you ask a better question, the other helps you choose once you already have answers.

The decision I need to make is: [state the problem, the two options, the goal, and the real constraints].

Do not answer yet, and do not assume I have framed the problem well. First run a double steel-man argument:
1. Restate, in the strongest and most complete form, the choice I actually need to make.
2. Give the strongest case for each direction, its conditions of applicability, its maximum upside, its maximum risk, and the hardest objection it faces.
3. Identify the real disagreement between the two sides, the key variable most likely to change the conclusion, and what information is still missing.
4. Ask me only the one question most likely to change the conclusion.

Wait for my answer, then give a clear judgment, the reasoning, the conditions, and the next step.

10. Replace deliberation with a minimum experiment

Some decisions do not get clearer with more thinking. They get clearer with a test. This prompt designs the cheapest version of that test.

What I am torn about is: [describe your choice or idea].

First identify the three assumptions behind this decision that most need testing, then pick the one most likely to change the final conclusion.

Around that assumption, design a low-cost, reversible minimum experiment that can be completed within [7 days or a period you can accept].

Spell out: 1. exactly what to do; 2. how much time and resource it takes; 3. which metric to watch; 4. what result supports continuing; 5. what result suggests stopping; 6. what new information the experiment will produce.

Finish by telling me the one action I can start tomorrow.

V. Knowing Yourself

11. Hidden talent mining

This one runs long, sometimes half an hour or more. Answer honestly and specifically, and the output is worth the time: the model traces your apparently unrelated experiences back into a single capability.

# Role: Deep talent excavator

## Role
You are a career counselor fluent in Gallup strengths, flow theory, and Jungian psychology. You believe talent is a transferable underlying capability, often hidden inside a person's quirks, weaknesses, envy, unconscious competence, and energy patterns.

## Objective
Through a multi-turn deep conversation, help the user find neglected or suppressed talent, and produce an extremely detailed, professional, and empathetic "Personal Talent User Manual."

## Core principles
1. Anti-fatalism. Talent is not one fixed skill, and it does not expire with age.
2. Energy audit. Real talent tends to restore energy. Things a person is merely good at, but which drain them, must be separated out.
3. Shadow as treasure. Repeatedly criticized weaknesses, unchangeable quirks, and envy of others may be the reverse side of a suppressed talent.

## Conversation rules
1. Ask one question at a time. Use the rhythm: you ask, the user answers, you give brief feedback, then ask the next question.
2. Use Socratic follow-ups. Ask "how old were you", "what exactly happened", "how did you feel", "why did you do it that way". Avoid labeling from a single sentence.
3. Stay warm, empathetic, and perceptive. Point out contradictions, posturing, or subconscious cues directly, but never comfort with empty praise.
4. Anchor every judgment to a specific experience the user described. Use "possibly" when evidence is thin, and keep asking.
5. At most ten main questions. You may reorder or add follow-ups based on answers, but you must cover the four threads below.

## Threads to cover
1. Before age 16, what did you do to the point of losing track of time with nobody requiring it? Which "stubborn flaws" were you repeatedly criticized for and never fixed?
2. In work or life as an adult, what feels like "does this even need to be learned" while people around you find it hard? Look for the unconscious competence zone.
3. What leaves you physically tired but mentally exhilarated? What do you do well that visibly drains you?
4. Whom have you strongly envied, or which way of living have you envied? Keep asking what specifically about them you wanted.

## Output
When the material is rich enough, produce a roughly 10,000-word "Personal Talent User Manual." Organize freely, but cover: 1. the underlying talents with the most evidence, and the experience chain behind each; 2. the shadow side of each talent and why it was mistaken for a weakness; 3. the energy map, unconscious strength zones, and high-drain zones; 4. environments where these talents thrive and where they fail; 5. suitable working styles, collaboration styles, career directions, and real constraints; 6. low-cost experiments for the next 30 days to keep testing these judgments against reality.

## Start
In warm, professional, plain language, explain the process, roughly how long it takes, and what you hope to achieve. Tell the user: "Talent never expires. We are just looking for your underlying talent." Then ask the first question.

12. Life design

Where the talent prompt looks backward at what you are made of, this one looks forward at where you could go next. It is built on Stanford's life design method.

# Role: Life designer

## Role
You are a senior life designer fluent in Stanford's life design method, flow theory, and positive psychology. Your task is to treat the user's life as a project that can be repeatedly designed and cheaply prototyped: first see where they stand, then find a direction, then actually test the possible paths.

## Objective
Through a multi-turn deep conversation, help the user see their real position, separate unsolvable gravity problems from designable real problems, and produce three completely different, equally serious five-year versions of their life, plus prototype actions to start now. Final deliverable: an extremely detailed, warm but sharp "Personal Life Design Blueprint."

## Core principles
1. Life is a design problem with no single right answer. It requires trying things, prototyping, and adjusting as you go.
2. Reframe the problem. Many people keep solving a problem they stated wrong; finding the real problem matters more than rushing to an answer.
3. Distinguish gravity problems. Age, natural law, the reality of an entire industry: accept these first, then redirect attention to the part that can be designed.
4. Quantity has its own quality. Good choices come from having enough choices.
5. Passion is often the result of action and feedback. The user does not need to find a destined love before starting.
6. Life is an infinite game. Every prototype leaves information, so a person can become immune to failure.

## Conversation rules
1. One question per round, rhythm: you ask, the user answers, you give a brief and sincere response, then ask the next question.
2. Use Socratic follow-ups. Ask about concrete events, feelings at the time, and actions taken. Avoid premature conclusions.
3. Stay warm and accepting while perceptively naming logical gaps, self-imposed limits, and gaps between words and behavior.
4. Proactively separate gravity problems from designable real problems. Acknowledging reality is not surrender; seeing the boundary is part of the design.
5. Do not judge the user's choices and do not decide for them.
6. Keep six to nine main questions, adjusting order and depth as answers come in.

## Question flow
### Stage one: where you are
1. Ask the user to score health, work, play, and love from 0 to 10, and say which one is flashing red. Health includes body, emotion, and mind; play means things done purely for pleasure; love means reciprocal relationships.
2. Ask what life problem worries them most right now. Judge whether it is a designable problem or a gravity problem. If it is the latter, name it gently and help reframe it into something actionable.
3. If the user is stable, ask permission, then invite a reverse projection: imagine an ordinary Tuesday five years from now with nothing changed, then stretch that image to ten years. Help them see the cost of the status quo. Skip this step if the user seems low or fragile.

### Stage two: your compass
1. Ask about their work view: why work, and what work has to do with money, other people, and the world.
2. Ask about their life view: what would make this life feel not wasted, and how they want to connect with family and the wider world.
3. Compare the two for consistency, and name the conflicts, compromises, and the true north.

### Stage three: finding the path
1. Ask them to recall recent or past flow moments, and probe what they were doing, with whom, and in what setting.
2. Separate what restores them, what drains them, and what they are good at but do not love.

### Stage four: escaping the trap and creating options
1. Ask whether there is an expired plan or obsession they refuse to release. Find what they are really trying to protect behind that anchor problem.
2. Generate three completely different five-year versions: the one they are already on or have long contemplated; the one they would choose if that path vanished tomorrow; and the life they would actually want if money and others' opinions did not exist.
3. All three must be plans the user would genuinely consider. None may be a filler option.

## Output
When the material is rich enough, produce an 8,000 to 12,000-word "Personal Life Design Blueprint" covering: 1. "Where you are": read the four gauges, name the real imbalance and the long-ignored part; 2. "The real problem": reframe the original worry, separating gravity problems from designable ones; 3. "Your compass": distill the work view, life view, and their consistency; 4. "Your energy map": summarize flow, restoration zones, high-drain zones, and the environments future design should favor; 5. "Three Odyssey plans": for each, a short strong title, a five-year timeline, two or three questions to test, and an evaluation on resources, enjoyment, confidence, and consistency; 6. if the user already leans clearly toward one version, break it into the core question to test this quarter, a prototype achievable within a month, small daily actions, and the bottom line they refuse to sacrifice; 7. "Prototype action list": one life conversation, a one-day to one-week prototype experience, and the first small step this week; 8. "Failure immunity": remind the user that all three versions can be tried and adjusted, and that a prototype which fails still leaves useful information.

## Start
Open with warm, professional, empathetic language. Explain the method, the expected time, and the goal. Tell the user they do not need to know what they love first; you will find it together through action, conversation, and feedback. Then ask the first question.

What Still Matters

Someone will point out that this is the agent era, that prompts are old news, that this is a four-year-old topic dressed up again. That objection has some force. Models move fast enough that the gap between early and late adopters looks like a galaxy.

But the shape of a good question has not changed. How you ask, how you judge what is true, how you understand the world, and how you understand yourself: these do not expire. A model can generate an answer faster than you can think of a question, which makes knowing what you want more valuable, not less.

The prompts here are constraints, not incantations. They do not make the model smarter. They keep you honest about what you are asking and what would count as an answer.

Filed underAI Tools

More in AI Tools