Pick a problem
Pick something you're carrying right now. It could be a personal issue in your life, or a professional roadblock you've been trying to get past for months. Try to make it something that matters so you can really feel the impact this problem solving technique can have on your life in all areas.
If no problem comes to mind, let your agent help you find one. Open a chat in the Project/Notebook holding your Agent Profile, or attach your current Profile to a new chat. Use the prompt below to find a problem you can take to the next step.
Using my Agent Profile and my answers, help me choose one real problem worth working through. For now, help me identify and describe the problem. Do not investigate causes, suggest solutions, or make a plan. Ask me up to three short questions, one at a time, about what is taking too much time, repeatedly going wrong, or making my work harder. With each question: - Offer two or three short suggested answers based on my role, responsibilities, priorities, or projects in my Agent Profile. - Write these options directly in your response as a numbered list. - Present them as possibilities for me to confirm, not problems you assume I have. - Include “Something else / none of these.” - Let me choose a number, combine options, or answer in my own words. Wait for each answer before continuing. Use my answer to shape the next question, and stop asking as soon as we have a clear problem I want to work on. If we still have several possibilities, suggest up to three specific problems grounded in my profile and answers. Describe each in one sentence about what is going wrong, rather than a task to complete or something to create. Ask which one I want to work on, and wait for my choice. Once I choose, write a one- or two-sentence problem description in my voice. Describe the pain and include its impact only if I have told you what that is. Do not add an assumed cause, a proposed solution, or facts I have not provided. Return that description in one copyable code block. Include only the words I should paste in place of [describe your problem here: the pain, not the task]. Then stop and wait for my next prompt. If I have already named a clear problem I want to use, skip the questions and options and give me that description directly. If details should not be shared, let me describe the general pattern instead.
In this guide, you will learn about a new way to prompt your agent called a problem prompt. These prompts help turn your agent into a thinking partner that can help improve your critical thinking and ability to solve problems. You aren't asking your agent to think for you, your agent reflects your own thoughts back at you from different perspectives and can spark ideas you wouldn't have otherwise thought of.
Task Prompts vs Problem Prompts
A task prompt like "Draft an email to Bob about the upcoming site opening" tells the Agent exactly what to produce, so it jumps straight to generating generic email drafting and the session becomes an exercise in formatting. That's fine when you already know what you need and just want it made.
A problem prompt starts with the situation instead: "We open a new site in about ten weeks. Every opening so far has been invented from scratch, each one went wrong somewhere different, and I want this to be the last one we improvise." Because it describes the pain rather than prescribing the output, it gives the Agent something to explore. It can ask questions, surface what went wrong before, and shape a solution that fits.
Rule of thumb: use the task style when the deliverable is already clear and defined; use the problem style when you're problem-solving or brainstorming and want the Agent to help figure out what the right deliverable even is.
Start the conversation
Open a new chat in the Project/Notebook holding your Agent Profile, or attach your Profile to a new chat in your approved tool. Copy the kickoff prompt and replace the bracketed line with your problem in a sentence or two: what is happening and why it matters, not the fix.
Work a problem in four phasesI have a problem I want to work through with you, step by step. If my Agent Profile is available, use it as background. The problem, in a sentence or two: [describe your problem here: the pain, not the task] We will go in four phases. You pace us: when a phase feels done, briefly say where we are and ask whether I am ready for the next phase or want to stay longer. Move on only after I say so. Phase 1: Interview. Ask me up to five questions to understand the problem. One at a time, short, each one earning its place. Offer answer options when they would help me answer faster. Otherwise ask the question plainly. Do not suggest solutions yet, and no commentary between questions. Then play the problem back: state it in your own words, including anything you see underneath what I first said, note what you are still unsure about, and ask me to confirm or correct it. My confirmation is your green light for Phase 2. Phase 2: Brainstorm. Give me three or four genuinely different ways to approach it, not variations on one idea. For each: what it looks like, its main advantage, its main drawback. Phase 3: Decide. Compare the options against what came out of the interview. Recommend one and say why. The pick is mine. After I make a provisional pick, give the strongest objection to it and name the assumption most worth checking, even if you recommended it. Wait for my confirmed choice before planning. Phase 4: Game plan. Turn my pick into a plan I can start this week: the first step, its proposed owner, the next few steps, and how I will check whether it helped. Mark unconfirmed owners and dates. Do not invent commitments. Carry forward any important assumption we still need to check. When needed, make a small information check or reversible trial the first action, and say what result would make me keep or reconsider the choice. Keep the choice provisional until then. Rules for all phases: if you would need information I should not share, let me answer with the general pattern instead. If my answers contradict each other, say so. The important calls stay with me. Start Phase 1 now.
Use your Agent Profile as background about you. For a personal problem, describe the part of the routine you control.
Send the prompt and stay in this one chat for all four phases.
Answer the questions and correct the playback
Answer one question at a time from what you know. Where you do not know, say so rather than guess.
Once you have answered all the questions, it will then play the problem back in its own words. Tell it what it got wrong or correct its assumptions until it re-states your problem in a way that you agree with.
This is the most important part of the process, if the agent doesn't have a good understanding of your problem, it will suggest solutions that miss the mark.
Once you are satisfied with how your agent is stating the problem, tell it to move on to phase 2.
Compare the different approaches
Phase 2 gives you three or four approaches or solutions to the problem. Read them for real difference: two options that both amount to another checklist are one option, so ask for one that works differently.
Read each advantage with a skeptical eye. Is the agent being too optimistic or do you agree with its assumptions?
Read each drawback against your own limits, the time you have, the people involved and what you can decide without approval. If there is a solution that requires too much time or resources, tell the agent what your constraints are and it will suggest alternative solutions.
When the options are distinct and you know what each costs, ask for Phase 3 and its recommendation.
Decide on a solution
Review the agents recommendation. Do you agree with what its suggesting?
If you agree, tell the agent and expect it to come back with the strongest objection against your decision and the most dangerous assumption being made. An agent that acts as a yes man will lead you to poor decision making, an agent that acts as a devils advocate and challenges even its own assumptions will lead to better and more confident decisions.
If you disagree, tell the agent what your preferred option is and why. You will get the same pushback described above. Work through the objection, then decide whether to keep or change your choice. The decision stays with you.
If you don't yet know whether an important assumption is true, keep your choice provisional. Ask what small information check or reversible trial would help you find out. Make that your first action, and say what result would make you keep or reconsider the choice.
Confirm your choice, or the small check you want to try first, and ask for Phase 4.
Turn the choice into a first action and a check
Phase 4 gives you a first step, a proposed owner, the next steps and a check. Read it as a proposal, and correct anything the agent got wrong or that you can't actually do.
If your choice still depends on an unconfirmed assumption, check that the plan says how you'll investigate it before committing to the full change.
Make sure that the success check is something concrete that you can observe within the next week or two. The goal is for you to have a clear metric for success that indicates you're on the path to solving the problem.
Copy the reviewed plan into a document or your task list and save it with a title you will find again. Reopen it and check that the first action, the ownership status and the review point read clearly.
Keep the method for the next problem
Download the prompt from step 2 and save it to the My Agent Files folder. We recommend creating a subfolder called Prompts so that you can start building up a library of prompts for later use.
Next time you have a problem that's eating at you, or you don't feel confident in a decision, use this prompt to help you think through it. Start with the new problem’s facts and limits, then work through the four phases again.
Confirm your result
Confirm the outcomes below to finish this guide.
Final checks
You have a problem statement about your own situation that you agree with, and a chosen approach whose main drawback you can explain.
You have a clear first step towards your chosen solution and a concrete checkpoint within the next two weeks.
Provide Feedback
Thanks. That went straight to the people who write these guides.
Week 1 field guides
Keep exploring when you’re ready
Available Tuesday, October 6, 2026