YOU-SHAPED AI A practical field guide by Tim Walker https://youshapedai.com Use AI without removing yourself. Bring your judgment. Keep your presence. Own the outcome. PRIVACY AND USE Review your provider’s memory, data-sharing and retention settings before sharing. Memory-off does not necessarily prevent chat history from being saved. Use temporary or incognito mode where available if appropriate; retention may still apply. Do not share passwords, account numbers or unnecessary sensitive information. Review the profile and keep your own copy. Prompts do not guarantee storage, connections, scheduling or agent availability. Replace [brackets] with your details, then paste into the assistant you choose. SHARED STARTERS AND WORKFLOWS THE PERSONAL INTERVIEW ====================== I want you to become a more useful assistant for the way I actually live and work. Start by interviewing me. Do not give me a generic productivity plan. Ask one question at a time and wait for my answer. Use follow-up questions when I am vague. Start with what I want help with, then learn the relevant parts of: - My roles, responsibilities, current priorities, and goals - What a normal week looks like and where work gets stuck - What energizes me, what drains me, and how I make decisions - My strengths, recurring friction, and support needs, in my own words - How I communicate, learn, organize information, and prefer feedback - The tools I already use and my current comfort with AI - The constraints you need to respect, including time, budget, privacy, and decisions I want to keep for myself Ask only what would help you support me. I may skip any question. Do not ask for passwords, account numbers, or unnecessary sensitive information about me or anyone else. Do not diagnose me or infer traits I have not described. After about 10 to 15 questions, tell me whether you know enough or whether one or two important gaps remain. Then draft a Personal Working Profile in plain language with these sections: 1. What matters to me right now 2. My responsibilities and current projects 3. How I work and make decisions 4. What gives me energy and what drains it 5. How to communicate with and support me 6. Useful recurring tasks you could help with 7. Boundaries and things that need my approval 8. Open questions and facts you are not sure about Keep facts, my stated preferences, and your tentative interpretations clearly separate. Skip praise and personality-label filler. Ask me what you got wrong. Revise it until I approve it. After I approve it, ask which parts I want retained. If this app supports memory or saved instructions, explain how those work here and save only the durable context I approve through the available feature. Tell me accurately what was saved and what was not. Never claim that saving a document guarantees memory in a new chat. Also give me a concise reusable version I can keep and paste into another conversation. Exclude anything I have marked private or temporary. Finally, use my approved profile to create a complete, personalized 30-day AI coaching prompt I can copy into a new conversation. Include the relevant approved context inside that prompt so it works on its own. The coach should ask about my daily time budget, use real tasks from my life and work, and adapt to my tools, ability, goals, and constraints. It should give me one day at a time with a practical outcome, a short explanation, an exact action or prompt to try, a quality check, a correction or feedback move, a brief reflection, and something reusable to keep. Review and adapt the plan on days 7, 14, 21, and 30. Use the five levels as a map: useful conversations; reusable context and setups; approved connections; reliable scheduled work; and bounded agent-owned jobs. Do not promise that I will reach Level 5 in a month or require extra tools that do not help. Keep my judgment, privacy, and approval boundaries explicit. The coach must not claim that a prompt creates memory, connections, schedules, or automatic reminders; those features must actually be available and verified. Where reminders are unavailable, tell me to return for the next session. Show me the complete coaching prompt for review before starting Day 1. YOUR 30-DAY COACH ================= Use my approved Personal Working Profile below to coach me for the next 30 days. If the profile is missing, ask me to paste it or run a short interview before planning. [Paste your approved Personal Working Profile here.] Build the plan around my actual tasks, current ability, tools, goals, and constraints. Ask how much time I can realistically spend each day; suggest ten minutes if I am unsure. Start with the most useful next skill for me. Do not turn this into a generic list of 30 AI tricks or a requirement to buy more tools. The five levels are: 1. Curious: useful conversations with an AI assistant. 2. Equipped: reusable prompts, working context, and repeatable setups. 3. Connected: approved access to relevant email, calendars, and files. 4. Automatic: useful work that runs on a reliable schedule. 5. Autonomous Agents: bounded whole jobs carried out and reported back for my approval. Use those levels as a map, not a race. I do not have to reach Level 5 in 30 days. Only introduce connections, schedules, or agents if the tools are available and the setup is genuinely useful to me. Explain any costs, privacy choices, and approval requirements before recommending a change. Don't claim a prompt alone can create a connection, schedule, persistent memory, or autonomous worker. First show me a short plan for the month. Organize it around my real use cases and the skills I need. Then coach one day at a time. Each daily session should contain: - One practical outcome connected to my life or work - A short explanation in everyday language - An exact prompt or action to try - A task using my own material, with private details removed where appropriate - A clear test for whether the result is good enough - One correction or feedback move so I learn to improve the output - A brief reflection and the reusable thing I should keep On days 7, 14, 21, and 30, review what I actually tried, what worked, what failed, and what I want to change. Adapt the remaining plan. Track time saved only when I have measured it; don't invent results. Help me build a small personal prompt library and an updated working profile along the way. Keep human judgment central. I review important facts, sensitive messages, public content, purchases, and consequential decisions. Prefer a draft or a proposed action when permission or information is missing. Begin by confirming my main goal and time budget, then give me Day 1. If you cannot send reminders from this app, say so; tell me to return for the next session rather than promising to contact me. KEEP ME IN THE WORK =================== Help me with [task] in a way that strengthens my contribution. Start by asking for my purpose, current thinking, standards, and anything that requires my judgment or personal presence. Then suggest which parts you can research, organize, draft, design, test, or prepare. Make the decisions that remain mine explicit. Don't invent beliefs, experiences, feelings, promises, or conclusions on my behalf. Challenge weak assumptions and show meaningful alternatives. Work from the relevant source material. Separate facts from inferences. Give me a result I can inspect and improve, with the important uncertainties and checks made visible. Help me understand enough to stand behind what I use. Before any consequential action, confirm the scope and approval it needs. LEVEL 2: A REUSABLE SETUP ========================= Help me turn this repeat task into a reusable setup: [task]. Ask for any essential missing context. Use my living brief and this example of a good result: [example]. Create a prompt with a clear goal, the inputs I need to supply, the output format, boundaries, and a short quality checklist. Run it on this week’s real task. After I review the result, update the prompt to reflect my corrections. LEVEL 3: A SOURCE-LINKED LOOSE-END CHECK ======================================== Using only the sources you can actually access, look for commitments I made about [project] in [time range] that may still be open. For each candidate, show the original source and date, explain what appears unfinished, and check for later evidence that it was completed or changed. Separate confirmed open items from uncertain ones. Suggest a next step for my review. Do not send messages or change records. LEVEL 4: A VERIFIED SCHEDULE ============================ Help me set up a recurring [brief/report/check] using the scheduling features actually available here. It should run [schedule and time zone], use [approved sources], and deliver to [my chosen destination]. A useful result includes [required contents]. First show me the complete instruction and any limitations. Test one run, then verify the saved schedule. Tell me how I can inspect, pause, change, or stop it and how I will know if a run fails. Do not claim it is scheduled until the app confirms it. LEVEL 5: A BOUNDED WHOLE-JOB BRIEF ================================== Outcome: [the finished result I want] Why it matters: [brief context] Sources and access: [approved sources and tools] You may: [specific permitted actions] Stop and ask before: [sends, publication, spending, commitments, sensitive data sharing, or other boundaries] Done means: [checkable acceptance criteria] If blocked: Explain the exact blocker and the smallest decision or access needed. Continue independent work that remains within scope. Do not quietly substitute an unapproved action. Return: The finished artifact or result, links to the evidence, checks you performed, anything uncertain, and the decisions I still need to make. Don’t describe a plan as completed work. A DESIGN BRIEF WITH TASTE ========================= Help me create [artifact] for [audience]. The main thing they should understand or do is [outcome]. It will be used in [format, device, or setting]. Here are my references: [links or examples I can share]. What I like about each is [specific qualities]. Use these to understand my taste; don't copy their artwork, branding, or exact layout. Required content: [content]. Constraints: [size, accessibility, tools, timing]. Things to avoid: [generic visual habits or styles that don't fit]. Ask the few questions that would materially improve the brief. Then propose two genuinely different visual directions. Explain the hierarchy, typography, spacing, color, and interaction choices in plain language. Recommend one for this audience and purpose. After I choose, make a first draft. Critique it against the brief and references before presenting it. Show me the three biggest improvements you would make next. Refine in focused passes using my feedback, preserving the parts that work. Verify the final file or experience in its intended format and tell me what you actually checked. BUILD A SMALL, USEFUL APP ========================= Help me build a small tool for [person] who needs to [one specific job]. I am describing the goal in everyday language; explain technical choices simply and don't assume I am a developer. The essential flow is: [starting point] → [main action] → [useful result]. Start with sample data. Keep the first version small enough to test. Ask essential questions before building and show me a brief plan. For visual direction, use these references: [links or examples]. I like [specific hierarchy, typography, spacing, navigation, or interaction qualities]. Translate those qualities into an original design with a clear main action. Avoid decorative filler and stock layouts that don't fit the purpose. Before coding, define what must work and how we will test it. Include the normal flow, empty input, invalid input, long content, and a phone-sized screen. Don't hide unfinished features behind buttons that appear to work. Build the smallest useful version, then test the entire flow. Show me what works, what remains incomplete, and any assumptions. Critique the visual quality as well as the behavior: hierarchy, type, spacing, contrast, consistency, and ease of use. Fix the three highest-impact issues before adding features. Use mock data until I approve any real data connection. Ask before adding costs, accounts, permissions, real personal data, or public deployment. Keep secrets out of the code and the browser. Return a working preview or runnable files, clear usage instructions, and the tests you actually performed. CONNECTED WALKTHROUGH: READ-ONLY EVENT SWEEP ============================================ Using only [approved source], look for still-open commitments about [one event] between [dates]. Check later replies for completion or a changed plan. Link the original source and date. Separate commitments I accepted from requests I have not accepted. Suggest next steps for my review. Do not send messages or change records. Say exactly what you could and could not search. AUTOMATIC WALKTHROUGH: MAKE THE PROVEN CHECK ARRIVE =================================================== Help me schedule the read-only event sweep we have already tested. Run [days and time] in [named time zone], using [approved source], until [end date or ongoing]. Deliver privately to [destination]. Include only [useful contents], with source links. If nothing is open, say so briefly. If access fails or data is stale, report that explicitly. No messages, record changes, or commitments. Show the complete instruction, test it safely, confirm the saved schedule and next run, and show me how to pause or stop it. If unsupported, say so rather than promising delivery. AGENT WALKTHROUGH: FINISHED EVENT PACKAGE ========================================= Prepare the complete follow-up package for [one event] by [date, time, and time zone]. Use only [approved notes and relevant messages]. Done means the current commitments are reconciled, the necessary drafts are written for verified recipients, required documents are ready, and every important fact has a source. You may read, research, draft, and check. Ask before sending, publishing, buying, deleting, changing access, or making commitments. If blocked, explain what is missing and continue independent work within scope. Return the actual package, source links, tests performed, uncertainty, and decisions for me. Do not call a plan or incomplete batch finished. REVIEW A WORKER’S RESULT ======================== Here are the original brief and the worker's result. Check them against each other. Which acceptance tests are actually supported by evidence? What is missing, uncertain, or outside the approved scope? Give me the decisions that require my judgment. Do not assume an action happened because the worker described a plan. Do not contact the worker or take external action unless I explicitly authorize it. FIVE APPLICATION CATEGORIES × FIVE LEVELS The examples below are fictional teaching examples. They show what a useful result could look like; this website has not accessed an account, run a schedule, or completed the work for you. LIFE · LEVEL 1 Plan one thing you've been putting off Use a short interview to make an everyday plan fit your time, energy, and preferences. Time: About 10 minutes to make a plan; do the first small step afterward. Needs: A conversation in ChatGPT or Claude. No connections or paid feature required for the exercise. THE REUSABLE PROMPT Interview me about [an everyday plan or personal project], one question at a time. Learn what would make it useful, my time and budget, what I enjoy, and what usually gets in the way. Then propose a small next step and a realistic plan. Separate facts from assumptions. Let me correct it before treating it as my plan. READY-TO-PASTE FICTIONAL EXAMPLE Help me plan a small task I keep putting off: sorting my kitchen drawers. I have 20 minutes on Saturday, no budget for storage boxes, and I get distracted if I empty everything at once. Ask one question at a time, then suggest a plan that fits. Do not make the project bigger. Let me correct the plan before we settle on it. TRY WITH I keep putting off sorting my kitchen drawers. I have 20 minutes on Saturday, no budget for storage boxes, and I want to find the utensils I actually use. I get distracted when I empty everything at once. A USEFUL RESULT MIGHT LOOK LIKE Choose one drawer. Take a before photo for yourself, then spend 5 minutes removing obvious trash, 10 minutes grouping the utensils you use, and 5 minutes putting things back. Leave the other drawers alone. Stop at 20 minutes even if the whole kitchen is unfinished. CHECK IT - Does it respect the 20-minute limit and zero budget? - Is the first action small enough that you could start now? - Can you find the utensils more easily afterward? That matters more than finishing the conversation. IF IT MISSES, SAY That plan is too big. Keep it to one drawer, 20 minutes, and things I already own. Ask me one question if you still need to know what makes this difficult. WITHOUT ACCESS Use any assistant that can hold a text conversation. A paper note of your constraints is enough context; you do not need to upload a photo. IT WORKED WHEN The plan fits your actual constraints and you take one useful next step. WHAT YOU KEEP A corrected plan and the preferences worth reusing. YOUR PART Decide what a good outcome means for your life and correct the plan until it fits. LIFE · LEVEL 2 Make a weekly planning brief Save the context that helps your week start with less friction. Time: About 15 minutes to create the brief; test it again next week. Needs: An assistant and somewhere you can save text, such as a note. Built-in memory is optional. THE REUSABLE PROMPT Help me make a reusable weekly planning prompt. Ask about the commitments I need to respect, the activities I want room for, my energy patterns, and what I tend to overlook. Use general descriptions rather than private details about other people. Draft a short working brief and a weekly review format. Test it on my next week and revise it from my feedback. TRY WITH My weekly planning rules: keep Wednesday evening free, leave one unscheduled hour on Sunday, choose no more than three priorities, and do errands in one trip. My unfinished tasks this week are a library return, a cabinet cleanup, and choosing a birthday gift. A USEFUL RESULT MIGHT LOOK LIKE A reusable brief begins with your four planning rules, asks for this week's fixed commitments, and returns three priorities, one combined errands trip, and a short 'not this week' list. The next week starts with the same rules and fresh facts. CHECK IT - Are lasting preferences separate from this week's temporary tasks? - Does the plan leave the protected evening and spare hour free? - Can you run the saved prompt next week by changing just the weekly inputs? IF IT MISSES, SAY You scheduled something during the evening I keep free. Add that as a hard boundary in the reusable brief, repair this week, and show me the exact instruction to save. WITHOUT ACCESS Paste the brief at the start of a new conversation. Saving a note is a useful reusable setup even if your app has no memory feature. IT WORKED WHEN Your second weekly plan needs edits rather than a complete restart. WHAT YOU KEEP A personal weekly brief and reusable planning prompt. YOUR PART Choose the priorities and make room for the people and activities that matter to you. LIFE · LEVEL 3 Check one source against your plan Let an approved calendar or reference folder supply context you can verify. Time: About 15 minutes after you review and approve one source connection. Needs: A supported calendar or file connection with read access. Check what your account can access and what permission it requests. THE REUSABLE PROMPT Using only [the calendar or source I have connected and approved], compare [my proposed plan] with the relevant existing commitments. Show the source and date of anything that affects the plan. Say what you cannot see. Suggest adjustments for my review without changing events or making commitments. TRY WITH Sample calendar: 'Community class, Thursday 6–7 p.m.' Proposed plan: 'Leave for the class at 6 p.m.; journey takes 25 minutes.' Ask the assistant to compare the plan with the calendar and show the original event. A USEFUL RESULT MIGHT LOOK LIKE The class starts at 6 p.m., but leaving then would make you 25 minutes late. Suggested departure: 5:25 p.m. for a 10-minute buffer. The answer links the event and makes clear that it has not changed it. CHECK IT - Open the event yourself. Are the date, start time, and time zone right? - Does the suggestion distinguish the supplied journey time from a live travel estimate? - Did it leave your calendar unchanged? IF IT MISSES, SAY Show exactly which calendar and event you used. If you could not read the source, say so. Use only the journey time I supplied and label the buffer as a suggestion. WITHOUT ACCESS Paste a redacted event and your proposed plan. You can practice the comparison now, but you have not yet tested a live connection. IT WORKED WHEN You find and verify one real conflict or useful detail you would have missed. WHAT YOU KEEP A source-linked planning check. YOUR PART Check the actual commitments and choose which tradeoffs you will make. LIFE · LEVEL 4 Make next week's preview arrive Schedule a useful planning check only after trying it by hand. Time: About 20 minutes to test and set up; verify delivery for two weeks. Needs: A scheduling feature that can read the approved source and deliver where you will see it. A recurring chat promise is not a saved schedule. THE REUSABLE PROMPT Help me turn the weekly preview I have already tested into a recurring read-and-report task. Use [approved source], run at [time and time zone], and deliver to [my chosen destination]. Include only [the details I find useful]. Show the instruction before setup and confirm the saved schedule. Explain how to stop it and how missed runs will be visible. If scheduling is unavailable, save it as a reusable routine. TRY WITH Every Sunday at 6 p.m. in my own time zone, show next week's fixed commitments, any overlaps, and two preparations worth doing. Read only my approved calendar. Deliver privately in this app. Do not change events or contact anyone. A USEFUL RESULT MIGHT LOOK LIKE Before saving, the assistant shows the complete instruction, named time zone, source, destination, and next run. A supervised test links the relevant events. After setup, you can open the saved schedule, see its next run, and find its pause or delete control. CHECK IT - Compare the first test with your calendar, including an event near midnight or a time-zone boundary if relevant. - Inspect the saved task yourself: instruction, time zone, next run, source, and private destination. - For two weeks, record expected and actual delivery and confirm what happens when access fails. IF IT MISSES, SAY The preview did not arrive. Check whether the schedule exists and is enabled, its next run and time zone, source access, and delivery destination. Report the actual failure; do not claim a successful run from the plan alone. WITHOUT ACCESS Save the same prompt and run it yourself each Sunday. That is useful preparation, but the Automatic milestone is earned only when the supported schedule actually runs reliably. IT WORKED WHEN For two weeks, the preview arrives on time without a manual start and you verify it against the source. A manual routine is useful preparation, but does not yet meet the Automatic milestone. WHAT YOU KEEP A verified schedule or a one-step weekly routine. YOUR PART Keep the preview in service of your life, and change or stop it if it creates noise. LIFE · LEVEL 5 Hand over the preparation for a personal project Give a worker a bounded outcome while you keep the decisions. Time: About 15 minutes to brief and review; research time varies. Needs: An agent that can carry out the research steps and return inspectable sources. Begin without private account access. THE REUSABLE PROMPT Prepare everything I need to start [a personal project]. The outcome is [specific useful package]. Use [approved sources] and these constraints: [time, budget, preferences]. Research and draft only. Do not purchase, book, contact anyone, or share personal information. Return the plan, options with source links, a checklist, assumptions, and the decisions for me. Verify that the package meets the finish line before reporting it complete. TRY WITH Prepare a beginner balcony herb-growing plan. There is a 3-foot sunny ledge, three existing pots, and a $40 total spending limit. Research only; do not purchase anything. Ask for the location and growing conditions if needed before recommending plants. A USEFUL RESULT MIGHT LOOK LIKE A finished package contains three suitable herb options, local-season caveats, a shopping list with dated prices and links where available, a simple care schedule, and decisions for you. Any unavailable price is marked unverified rather than invented. CHECK IT - Do the plants fit the light, space, climate, and season you actually described? - Open the product and growing-guidance links and check the total, including likely delivery costs. - Can you choose what to do next without repeating the research, and is it clear nothing was bought? IF IT MISSES, SAY This is a list of ideas, not the finished package. Complete the shopping list, source checks, care plan, and unresolved decisions. Stay within research and preparation; do not buy or contact anyone. WITHOUT ACCESS Use the brief in a normal assistant conversation and manage each step yourself. That teaches the same briefing skill, while leaving the end-to-end agent milestone for a supported setup. IT WORKED WHEN The package is ready to use and your remaining work is choosing, approving, or acting. WHAT YOU KEEP A complete preparation package and a better handoff brief. YOUR PART Make the choices and commitments yourself; use the preparation to be more present. KNOWLEDGE WORK · LEVEL 1 Think through one real question Use AI to clarify your thinking and test an argument against source material. Time: About 10 minutes with a short piece of text you can judge. Needs: A normal assistant conversation and source material you are allowed to share. THE REUSABLE PROMPT Here is a question I am working through: [question]. Here are my current thoughts and source material: [material]. First ask what I am trying to understand or decide. Then identify the strongest argument, the gaps, and a credible alternative interpretation. Distinguish what the source says from your inference. Help me develop my own view rather than supplying a conclusion to accept. READY-TO-PASTE FICTIONAL EXAMPLE Help me think through this fictional note: Twenty people answered our event survey. Fourteen prefer Saturday. All replies came from the existing Saturday group. Does that show Saturday is best for everyone? Explain what the note supports, what it does not support, and one question that would improve our judgment. Keep it brief and do not invent more data. TRY WITH Fictional note: 'Twenty people answered our event survey. Fourteen prefer Saturday. All replies came from the existing Saturday group.' Question: Does this establish that Saturday is the best day for everyone? A USEFUL RESULT MIGHT LOOK LIKE The note supports a preference among these respondents. It does not establish the preference of everyone who might attend, because the sample came from a Saturday group. A useful next question is who was missing from the survey. CHECK IT - Can you point to the sentence supporting each claim? - Did it avoid inventing survey results or treating a small group as everyone? - Can you explain the alternative interpretation in your own words? IF IT MISSES, SAY Separate what the note directly says from what you inferred. Quote the relevant short phrase for each factual claim, then identify what we would need to learn before making the wider conclusion. WITHOUT ACCESS Paste a short public or fictional passage. Research tools and file uploads are not required for this exercise. IT WORKED WHEN You can explain the reasoning, identify uncertainty, and point to the supporting material. WHAT YOU KEEP A clearer question and a short, evidence-aware thinking note. YOUR PART Bring your own question and reasoning, test the evidence, and form the conclusion you can defend. KNOWLEDGE WORK · LEVEL 2 Save your best research brief Make a good research request and quality standard reusable. Time: About 20 minutes; reuse the template on a second question. Needs: An assistant, one example of a useful answer, and a saved text note or project. THE REUSABLE PROMPT Turn this research task into a reusable brief: [task]. Use this strong example: [example]. Include the question, audience, source standards, time scope, output structure, and checks for unsupported claims. Test it on [current topic]. After I mark what is missing or unhelpful, revise the template. TRY WITH Create a reusable brief for choosing a local evening class. I need the date, travel time, total cost, beginner suitability, cancellation terms, and a link to the organizer's current information. Label anything you could not verify. A USEFUL RESULT MIGHT LOOK LIKE The template asks for location, topic, dates, and budget before researching. It returns the same comparison fields each time, a short recommendation tied to your criteria, and an 'unknowns to check' section. CHECK IT - Does the template request the few inputs that change the answer? - Does it prefer the organizer's current page over an undated summary? - Can you use it for another class without rewriting the method? IF IT MISSES, SAY Your answer leaves out total cost and cancellation terms. Add those to the reusable quality checklist and mark missing information as unknown instead of filling it in. WITHOUT ACCESS Supply the relevant source text yourself. If browsing is unavailable, do not treat recalled class details as current. IT WORKED WHEN The next brief starts with the right question, source standard, and structure. WHAT YOU KEEP A research template and a quality checklist. YOUR PART Set the evidence standard and correct the research pattern when it misses. KNOWLEDGE WORK · LEVEL 3 Answer from the actual documents Connect one approved source and trace the answer back to it. Time: About 15 minutes after source access is in place. Needs: A supported connection to a small approved folder or knowledge source and permission to use the documents. THE REUSABLE PROMPT Using [approved folder or knowledge source], answer [specific question]. List the documents you actually used and their dates. Link each important claim to its source. Look for contradictory or newer evidence before concluding. Separate documented facts, interpretation, and gaps. Tell me if access or search coverage is incomplete. TRY WITH Fictional folder: 'Event plan v1' says the session lasts 60 minutes. 'Event plan approved' is newer and says 45 minutes. Ask: What session length should the current handout use? A USEFUL RESULT MIGHT LOOK LIKE Use 45 minutes, based on the newer approved plan, with a link to that passage. The answer also flags the conflicting 60-minute draft so you can decide whether it needs updating. CHECK IT - Do the links open the actual documents and relevant passages? - Did it inspect the newer evidence rather than simply prefer the first search result? - Does it say which sources were searched and what might be missing? IF IT MISSES, SAY You used an older draft. Compare the document dates and approval status, show the conflict, and revise the answer from the current approved source. Do not silently edit either document. WITHOUT ACCESS Paste the two short excerpts with their titles and dates. This practices source comparison, but not connected retrieval. IT WORKED WHEN You can open the cited sources and verify the material conclusions. WHAT YOU KEEP A sourced answer and a map of the relevant documents. YOUR PART Read the important sources and own the interpretation you use. KNOWLEDGE WORK · LEVEL 4 Create a signal-only research digest Schedule a focused source check that earns its place in your week. Time: About 20 minutes to test and schedule; evaluate for two weeks. Needs: A supported scheduler and reliable access to a narrow set of sources. Confirm how the last check is tracked. THE REUSABLE PROMPT Help me test a recurring digest about [topic] from [approved sources]. Cover only changes since [last check], explain why each matters to [my purpose], and link the evidence. Exclude repeats and low-value items. Define what to report when there is nothing useful. Run it once for review, then help schedule it only if the result is worth receiving. TRY WITH Each Friday, check the official pages of three local museums for newly announced free public events next month. Report only new or changed events, with dates, links, and why they fit an adult beginner. If nothing changed, say 'No relevant changes.' A USEFUL RESULT MIGHT LOOK LIKE A useful digest reports one newly announced event with its official source, explains the fit, and gives the date checked. It excludes the two events already included last week. No-change weeks use the agreed short message. CHECK IT - Open the official page and verify the event date and admission conditions. - Does it compare with the prior check, rather than repackage old items as new? - Can you see the saved schedule, delivery destination, and a clear failure notice when a source cannot be read? IF IT MISSES, SAY This repeats last week's item. Compare against the prior digest, state the actual change, and omit unchanged items. If you cannot track prior runs, disclose that limitation and propose a reviewable method. WITHOUT ACCESS Keep a dated note of the last digest and run the check yourself. Do not call it scheduled until the app confirms a supported recurring task. IT WORKED WHEN Two weeks of runs bring useful changes without repeat noise or hidden failures. WHAT YOU KEEP A narrow digest instruction and a verified delivery rhythm. YOUR PART Decide what matters and periodically check that the digest supports your thinking. KNOWLEDGE WORK · LEVEL 5 Delegate an evidence-backed decision memo Ask for a complete research outcome with counterarguments and uncertainty. Time: About 15 minutes to set the brief; allow time for evidence checks and your review. Needs: An agent with access to the research sources and a clearly bounded question. THE REUSABLE PROMPT Produce a decision memo on [question] for [audience]. The decision criteria are [criteria]. Use [approved sources or research scope]. Compare the strongest options, investigate contradictory evidence, and state uncertainty. Return a concise recommendation with sources, a comparison, open questions, and a list of checks performed. Do not make the decision, contact people, or commit resources. Done means I can judge the options without reconstructing your research. TRY WITH Prepare a decision memo comparing three possible venues for a free reading group. Requirements: 12 seats, step-free access, a quiet weekday evening, and no booking commitment yet. Use official venue pages; flag anything needing human confirmation. A USEFUL RESULT MIGHT LOOK LIKE The memo compares the options against all four criteria, recommends one conditionally, explains the strongest reason to choose an alternative, and separates verified facts from questions such as current room availability. It has not contacted or booked a venue. CHECK IT - Does every material fact link to evidence you can inspect? - Are missing accessibility or availability details left unresolved rather than assumed? - Can you make the decision from the memo, including the trade-offs and remaining questions? IF IT MISSES, SAY You recommended a venue without checking a required criterion. Complete the comparison or mark that criterion unverified, then revise the recommendation and show the strongest counterargument. WITHOUT ACCESS Work through the memo in a normal conversation using pages you provide. You manage the research steps until a supported agent can own the whole bounded job. IT WORKED WHEN The memo supports a real judgment and every important factual claim has inspectable evidence. WHAT YOU KEEP A decision-ready memo and a repeatable job brief. YOUR PART Make the decision and be able to explain the evidence and uncertainty behind it. DESIGN · LEVEL 1 Teach AI what good looks like Describe your taste with references before asking for a design. Time: About 15 minutes to choose a direction. Needs: An assistant and two public references or examples you have permission to share. Descriptions work if image viewing is unavailable. THE REUSABLE PROMPT Help me define a visual direction for [artifact and audience]. Here are two or three references I can share: [links]. I like [specific qualities] in each. Ask about the purpose and the main action. Explain what makes the references work in terms of hierarchy, type, spacing, color, and imagery. Propose two original directions and recommend one. Do not copy the references' artwork or identity. TRY WITH Design a simple notice for a neighborhood book swap. The key action is to bring a book to the community hall on Saturday at 10 a.m. I like the generous space in reference A and the large, readable title in reference B. I dislike crowded flyers. A USEFUL RESULT MIGHT LOOK LIKE Direction one is a calm, text-led notice with a large date and room to breathe. Direction two is a playful grid of book shapes with the same clear date and action. The recommendation explains which serves quick reading from a noticeboard better. CHECK IT - Can a reader tell what, when, where, and what to do within a few seconds? - Can you name a specific reason for the type size, spacing, and color choices? - Are the references used as principles rather than copied artwork? IF IT MISSES, SAY The result looks generic and crowded. Keep the event facts, remove decorative filler, and offer two genuinely different layouts. Explain how each makes the main action easier to notice. WITHOUT ACCESS Ask for a written visual brief and a simple layout sketch described in words. Image generation is optional; choosing a thoughtful direction is the skill. IT WORKED WHEN You choose a direction for specific reasons and can explain why it fits the audience. WHAT YOU KEEP A small reference board and an intentional visual brief. YOUR PART Notice and name what you like; your deliberate choice starts the design. DESIGN · LEVEL 2 Build a small reusable style kit Turn one good design into repeatable decisions and a critique checklist. Time: About 25 minutes to extract and test a small kit. Needs: One design you can inspect, an assistant, and a tool capable of editing the chosen format if you want a finished visual. THE REUSABLE PROMPT Use my approved visual brief and this example: [example]. Create a small style kit for [kind of work]: type hierarchy, spacing, colors, image treatment, and layout rules. Explain each choice simply. Apply it to one new piece, critique the result against the reference, and refine the three biggest weaknesses. Keep the kit focused on decisions I will actually reuse. TRY WITH Use the approved book-swap notice as a starting point. Make a small kit for future community notices: title, date, body text, colors, margins, and where the main action belongs. Apply it to a different fictional event. A USEFUL RESULT MIGHT LOOK LIKE The kit records a clear title/date/body hierarchy, one dark text color, one accent, generous margins, and a short readability check. The next notice feels related but changes its layout to fit the new amount of text. CHECK IT - Is the kit small enough to use, with actual choices rather than vague advice? - Does the second notice remain legible when the title or venue name is long? - Are color and spacing rules consistent without forcing every piece into the same layout? IF IT MISSES, SAY The second piece copies the first layout even though the content is longer. Keep the hierarchy and visual character, but redesign the spacing around this content and test it at its intended size. WITHOUT ACCESS Save the kit as a text checklist and apply it manually in a design tool you already know. A style kit does not require a new subscription. IT WORKED WHEN A second piece feels coherent and purposeful without copying the first layout. WHAT YOU KEEP A practical style kit, template, and critique checklist. YOUR PART Own the visual standard and critique whether the reusable system serves the purpose. DESIGN · LEVEL 3 Work from the approved asset library Use real brand assets and current content with traceable sources. Time: About 20 minutes with a small approved asset folder. Needs: A supported file connection, permission to use the assets, and enough access to inspect versions and licensing notes. THE REUSABLE PROMPT Using only [approved asset folder], prepare a design brief for [artifact]. Identify the current logo, approved colors, type choices, images, and required copy you actually found. Flag missing files, unclear rights, conflicting versions, or unverified claims. Link each selected asset to its source. Then propose a layout that fits my visual references. Do not invent assets or present placeholders as final. TRY WITH Fictional folder: a current logo, an older logo, a short color guide, two licensed photographs, and event copy. Ask for the assets suitable for an A4 poster and the missing information that would block printing. A USEFUL RESULT MIGHT LOOK LIKE The brief links the current logo and chosen photograph, records the approved colors, and flags a missing image-resolution or usage-rights detail. It does not substitute an invented logo or pretend the missing detail has been checked. CHECK IT - Does each chosen asset link to the correct file and version? - Are dimensions, rights, and required copy actually verified or clearly unknown? - Can you distinguish final assets from placeholders before making a design? IF IT MISSES, SAY You selected an old logo. Check the folder's current-version guidance, replace it with the approved file, and show the source. Leave any unresolved rights or resolution issue visibly open. WITHOUT ACCESS Upload or paste a small set of assets only if you are allowed to share them. You can learn asset checking without granting access to an entire drive. IT WORKED WHEN Every selected asset and required statement has a verified source, and gaps are visible. WHAT YOU KEEP A source-backed design brief and an approved asset shortlist. YOUR PART Choose appropriate assets and verify that the content and rights fit the intended use. DESIGN · LEVEL 4 Make a recurring visual quality check Schedule a review of new design work once the checklist has proved useful. Time: About 20 minutes for a supervised check; validate repeated runs over two weeks. Needs: A supported scheduler, access to the design files, and tools that can actually inspect the relevant visual or export format. THE REUSABLE PROMPT Help me test a recurring review of new items in [approved design folder]. Check only against [my style guide and brief]: hierarchy, readability, spacing, contrast, consistency, correct assets, and export readiness. Return the three highest-impact issues with the file and location, plus what already works. Do not edit or publish. After a supervised trial, schedule only if this source and check are supported. TRY WITH Every Friday, review new draft notices in one approved folder. Use our kit to check text hierarchy, event facts, contrast, image quality, and print size. Report the three most important problems. Do not edit, publish, or email anything. A USEFUL RESULT MIGHT LOOK LIKE The report identifies the file and page where a date is too small, explains why it is a problem, and suggests a specific fix. It says when a file could not be opened and avoids claiming a visual inspection of unreadable formats. CHECK IT - Seed one harmless known issue in a sample file and see whether the check catches it. - Does every reported issue name a file/location and support a concrete repair? - Verify the saved schedule, the files actually inspected, and visible reporting of skipped or failed files. IF IT MISSES, SAY You reported a visual check without opening the export. State which files and pages you actually inspected. If your tools cannot render this format, mark that check incomplete and ask for a supported export. WITHOUT ACCESS Run the same checklist by hand whenever you make a piece. That builds a trusted review routine before you consider scheduling it. IT WORKED WHEN The review reliably finds actionable issues in real work without inventing defects or silently skipping files. WHAT YOU KEEP A tested critique checklist and, if supported, a verified review schedule. YOUR PART Keep developing your eye; a checklist supports your critique without deciding taste for you. DESIGN · LEVEL 5 Hand over a complete design package Delegate a usable set of assets with a brief, taste standard, and review gate. Time: About 20 minutes to brief and choose a direction; production and review time vary. Needs: An agent with tools that can create and open the requested files. Confirm deliverable formats before starting. THE REUSABLE PROMPT Create [a bounded design package] for [audience and purpose] using [approved content and assets]. Use these references to understand my taste: [references and reasons]. Propose two directions and wait for my choice. Build the selected direction, critique it, and refine the highest-impact issues. Deliver [required formats and sizes]. Open the exports and verify text, dimensions, readability, and consistency. Report anything incomplete. Do not publish or buy assets. TRY WITH Create a fictional book-swap package: one A4 print notice and one square social image. Use approved event copy and an original visual direction inspired by two permitted references. Show two directions, wait for my choice, then produce the files. Do not post them. A USEFUL RESULT MIGHT LOOK LIKE The returned package contains the two requested exports, editable source where supported, and a check report covering dimensions, spelling, required event facts, readability, and consistent visual choices. Any unsupported output format is disclosed before it is promised. CHECK IT - Open both exports and inspect them at their real viewing sizes. - Are the details identical across formats and is the main action easy to find? - Are the files actually usable rather than screenshots of unfinished editor views? IF IT MISSES, SAY The files exist, but the date is hard to read in the square image. Preserve the direction, fix the hierarchy and spacing for that format, then reopen and verify both exports before calling the package complete. WITHOUT ACCESS Have the assistant prepare the complete brief, exact copy, and layout directions, then use a design tool you already have. Label that as a prepared brief rather than a finished visual package. IT WORKED WHEN You can open and use the deliverables at the intended sizes, and the design choices serve the brief. WHAT YOU KEEP A reviewed design package, editable source where available, and export checks. YOUR PART Choose the direction, critique the result, and approve work that serves the audience. ADMINISTRATION · LEVEL 1 Move one stalled follow-up Let AI draft the finish while you supply context and judgment. Time: About 5–10 minutes, including your review. Needs: A normal assistant conversation. No inbox connection is necessary. THE REUSABLE PROMPT Here is a follow-up I have been avoiding: [context with unnecessary private details removed]. Ask only what you need to understand the recipient, purpose, and desired next step. Draft it in a clear, natural voice using this example of my writing: [example]. Do not invent a commitment or deadline. Tell me what needs checking. I will review and send it myself. READY-TO-PASTE FICTIONAL EXAMPLE Help me write a short follow-up. I promised to send our gardening club meeting notes, but I have not finished them. I need to ask whether people want the planting plan or the volunteer list first. Do not invent a deadline, an excuse, or an attachment. Give me a clear draft under 70 words and tell me what I should check before sending it myself. TRY WITH I promised to send the notes from our gardening club meeting, but I have not finished them. I need a brief message asking which of two topics should be covered first. Do not invent a completion date or say the notes are attached. A USEFUL RESULT MIGHT LOOK LIKE A short draft explains that the notes are still being prepared and asks which topic would be most useful first. It does not invent an attachment, an excuse, or a promise. You make the final wording your own before sending. CHECK IT - Is every factual statement true? - Does it avoid commitments, emotions, excuses, and attachments you did not supply? - Is the recipient's next action clear and easy to answer? IF IT MISSES, SAY You promised delivery on Friday, but I did not agree to that. Remove the deadline and apology you invented. Keep the message warm, direct, and under 70 words. WITHOUT ACCESS Describe the situation without names or sensitive details, then send the reviewed draft yourself in your usual app. IT WORKED WHEN One real follow-up moves after your review, rather than staying at a half-finished draft. WHAT YOU KEEP A finished draft and one correction that improves the next one. YOUR PART Check the facts and voice, and make only commitments you intend to keep. ADMINISTRATION · LEVEL 2 Turn three bullets into a reliable update Save a repeatable structure and examples of your real voice. Time: About 20 minutes to set up; try again on the next update. Needs: A few examples of writing you are allowed to share and somewhere to save the finished setup. THE REUSABLE PROMPT Help me build a reusable setup for [weekly update, recap, or follow-up]. Here are two examples I actually wrote: [examples]. Learn the structure and voice without inventing my feelings. The inputs should be three to five factual bullets. Produce the format, a prompt, and a short check for accuracy, commitments, and missing context. Test it now, then revise it from my edits. TRY WITH Fictional weekly bullets: room reserved; printed signs still needed; two volunteers confirmed; need a decision on the start time. Preferred format: completed, next, and one question. No upbeat filler and no invented dates. A USEFUL RESULT MIGHT LOOK LIKE The reusable prompt turns factual bullets into the three agreed sections, highlights the start-time decision, and labels any missing owner or date as a question. It uses the structure of your examples without copying unrelated facts. CHECK IT - Can you trace every statement back to an input bullet? - Did it keep the decision request distinct from general updates? - Does a second run need small edits rather than a new explanation of your style? IF IT MISSES, SAY You turned 'signs still needed' into 'signs will be ready tomorrow.' Restore the uncertainty, ask for the owner or timing if needed, and add 'never convert an open task into a promise' to the saved checklist. WITHOUT ACCESS Keep the prompt and examples in a private note and paste them each time. Automatic memory is helpful only if available and correctly configured. IT WORKED WHEN A repeat update comes back close enough that you are making judgment edits. WHAT YOU KEEP A saved communication setup with authentic examples. YOUR PART Correct the style and meaning so the draft represents what you actually want to say. ADMINISTRATION · LEVEL 3 Find a real loose end Check actual source records and later replies before declaring something open. Time: About 15 minutes after reviewing access to one source. Needs: A supported inbox, message, or document connection with a narrow read scope. THE REUSABLE PROMPT Search [approved source] for commitments about [scope and time range] that may still need my attention. For each candidate, show the original source, date, and requested next step. Check later replies or records for completion, cancellation, or a changed plan. Separate confirmed open items from uncertain ones. Prepare suggested next steps without sending messages or changing records. TRY WITH Fictional thread: Monday, 'I'll send the volunteer rota.' Tuesday, 'Here is the rota.' Wednesday, 'Could you also confirm the room key pickup?' Ask which item may still be open and why. A USEFUL RESULT MIGHT LOOK LIKE The rota is closed by Tuesday's reply. The room-key question is a candidate open item, supported by Wednesday's message, pending a check for later replies. It is not automatically an overdue promise by you. CHECK IT - Open the original request and any later reply before acting. - Does it distinguish a request made to you from a commitment you actually accepted? - Does the search cover the right source and time range without sending or changing anything? IF IT MISSES, SAY You called the rota overdue even though it was sent. Read later replies and reconcile the status. For the key pickup, say whether I accepted the task or was only asked about it. WITHOUT ACCESS Paste a short redacted thread to learn how to verify open items. A pasted thread is practice, not proof your whole inbox has been searched. IT WORKED WHEN You verify one genuine open item you had lost track of. WHAT YOU KEEP A short source-linked list with clear next actions. YOUR PART Verify what is still open and decide what deserves action. ADMINISTRATION · LEVEL 4 Schedule the sweep you already trust Make one proven open-loop check arrive on a predictable rhythm. Time: About 20 minutes to test and save; check reliable operation over two weeks. Needs: A scheduler that can read the approved source and deliver a private report. Access, recurrence, and delivery must each be confirmed. THE REUSABLE PROMPT Turn my tested open-loop sweep into a recurring read-and-draft task. Run [schedule and time zone] against [approved sources] and deliver to [destination]. Include only open items that matter, source links, and draft next steps. Nothing is sent or changed without my approval. Show me the saved instruction and confirm the schedule. Define how missed runs and missing access will be reported. TRY WITH At 8 a.m. every weekday in my named time zone, check the approved project folder and relevant messages for still-open commitments due this week. Return at most five important items with sources and draft next steps. No messages are sent and no records are changed. A USEFUL RESULT MIGHT LOOK LIKE A saved task has the exact schedule and source limits. Each report separates confirmed commitments from uncertain requests and includes evidence checked for later completion. A failed read produces a visible access warning rather than an empty 'all clear.' CHECK IT - Run the instruction once while watching and compare every item with the source. - Open the saved task and verify its timing, time zone, instruction, and destination. - Check two weeks of expected versus actual runs and find the pause/stop control before relying on it. IF IT MISSES, SAY An empty report is not enough. Tell me whether you searched successfully, what you searched, whether access failed, and whether later messages were checked. Keep uncertain status separate from 'nothing open.' WITHOUT ACCESS Save the proven sweep and run it yourself at a regular time. Do not treat the routine as Automatic until the app confirms and delivers the recurring task. IT WORKED WHEN The sweep runs for two weeks without manual starts, missed delivery, or silent failures. WHAT YOU KEEP A verified schedule and a short review routine. YOUR PART Review the important findings and keep authority over sends and commitments. ADMINISTRATION · LEVEL 5 Create a bounded approval desk Have a worker prepare a batch of completed drafts and decisions. Time: About 15 minutes to brief a small batch; reserve time to review each proposed action. Needs: An agent with approved read access and the ability to return a finished, reviewable work package. Sending permission is not needed for preparation. THE REUSABLE PROMPT Own the preparation of [bounded administrative job] for [scope]. Use [approved sources] and [my saved standards]. Gather the current facts, check whether each item is still open, prepare the needed drafts or documents, and bring me one review batch. For each item include the source, proposed action, and any uncertainty. Never send, publish, spend, delete, or make commitments without my explicit approval. Reconcile the finished batch against the source before reporting. TRY WITH Prepare follow-up for one volunteer event. Read only the approved event notes and relevant messages. Reconcile unanswered questions, draft the needed replies, and return one review batch. Do not send, book, publish, spend, delete, or make commitments. A USEFUL RESULT MIGHT LOOK LIKE Each batch item shows its source, current status, exact draft, verified intended recipient, any decision or commitment in the wording, and what remains uncertain. Already resolved items are removed. Your review ends in approve, change, or reject. CHECK IT - Are all drafts based on current facts and addressed to the right person? - Can you see every consequential decision, promise, or disclosure before approving it? - Did the agent finish the whole agreed batch and clearly identify anything blocked? IF IT MISSES, SAY This is a to-do list for me rather than a review-ready batch. Prepare the actual drafts, verify recipients and current status, and put the decisions I need to make beside each item. Do not send them. WITHOUT ACCESS Give a normal assistant a redacted set of notes and ask for the same review format. You will carry material between tools and manage the steps yourself. IT WORKED WHEN The batch is complete, traceable, and ready for you to approve, correct, or reject. WHAT YOU KEEP A review-ready work package and improved instructions. YOUR PART Read the batch, resolve uncertainty, and approve each consequential action deliberately. DEVELOPMENT · LEVEL 1 Build a small tool, one step at a time Describe one useful job, choose a simple plan, and personally try each small step. Time: About 20–30 minutes for a very small prototype, depending on the tools available. Needs: An assistant surface that can build and preview a small interactive page, or a coding tool with an explained setup. A plain chat may only return instructions or code. THE REUSABLE PROMPT Help me build a small tool for [person] to [one useful job], one step at a time. Explain technical choices in everyday language. First ask essential questions, propose one simple user flow and say how I can tell whether it works. Pause so I can choose or correct the plan. Then help me make the smallest preview using sample data and no accounts. Use these visual references for principles, not copies: [references]. Give me one clear thing to try myself, and wait for what I observe. After I test it, help me make one focused correction to the behavior or design. Keep me involved in choosing, trying and revising rather than handing back a whole unseen project. Ask before costs, accounts, new permissions, real data or public deployment. TRY WITH Help me make a packing checklist for a day trip. Start by proposing a tiny plan for adding an item and ticking it when packed; wait for me to choose. Then help me try that flow with fictional items and no account or personal data. I will tell you what happens before we add a reset button or make a design change. A USEFUL RESULT MIGHT LOOK LIKE You approve an add-and-tick plan, open a small preview, add “water bottle” and tick it yourself. You notice that a long item is hard to read and ask for one layout fix. Then you test again. The assistant explains whether the list survives a refresh. CHECK IT - Did you choose or correct the plan before the build started? - Try adding and ticking one item yourself. Describe one confusing behavior or visual choice and ask for a focused fix. - Try the revised flow, an empty item and a long item. Check keyboard use and whether refresh behavior matches what the assistant claimed. IF IT MISSES, SAY Pause. I want to learn this one step at a time. The Add button does nothing when I try it. Explain the smallest fix, help me apply it, and let me test again before we add anything else. WITHOUT ACCESS Ask for a plain-language brief and a screen-by-screen sketch first. If your app cannot run a preview, ask what supported tool would be needed; do not install something or buy a plan merely because code was produced. IT WORKED WHEN You have personally tried the small flow, made one useful correction, and can explain what it does and does not do. WHAT YOU KEEP A small prototype, the brief you chose, and one lesson from testing it. YOUR PART Choose the next step, try the tool yourself, and turn what you notice into a concrete correction. DEVELOPMENT · LEVEL 2 Save a project recipe that includes taste Reuse requirements, visual decisions, and tests instead of starting with a vague request. Time: About 20 minutes to capture and test the recipe. Needs: A working small prototype, access to its source or editable project, and a way to keep the instructions. THE REUSABLE PROMPT Turn this successful prototype into a reusable project recipe: [prototype]. Record the audience, job, main flow, sample data, visual references, design rules, and acceptance tests. Explain the setup in plain language. Include a critique step for hierarchy, type, spacing, contrast, and ease of use. Use the recipe for one small change and verify that the original flow still works. TRY WITH Use the packing checklist from the first exercise. Save its purpose, simple visual direction, no-account boundary, persistence behavior, and tests. Then add a 'remaining items' count without changing the rest of the experience. A USEFUL RESULT MIGHT LOOK LIKE The project recipe tells the next assistant what the tool does and how to judge it. The count changes when an item is added or ticked, stays readable on a small screen, and does not break the original controls. CHECK IT - Can another conversation understand the project from the recipe and supplied source? - Do the original add–tick–reset actions still work? - Does the new count match the visible unticked items after each action? IF IT MISSES, SAY The new feature broke reset. Restore the original behavior, add a regression check for reset, and update the recipe so future changes must pass both old and new tests. WITHOUT ACCESS Save the recipe, screenshots, and the checks in a text document. An editable project makes reuse easier, but the learning is in preserving intent and evidence. IT WORKED WHEN A second version keeps the visual quality and behavior while requiring less re-explanation. WHAT YOU KEEP A project brief, visual guide, and practical test checklist. YOUR PART Choose the product and visual standards, and check that the second version still helps. DEVELOPMENT · LEVEL 3 Connect one real source carefully Replace mock data only after the prototype and access boundaries are clear. Time: About 30 minutes or more, depending on the source and permission setup. Needs: A working prototype and a supported way to read one approved source. Real credentials must use the tool's secure setup, never a public page or ordinary code. THE REUSABLE PROMPT Help me connect this working prototype to [one approved source]. First explain what data it needs, what access is required, and what will leave the source. Keep the smallest useful scope and ask before credentials, permissions, costs, or real personal data. Keep secrets out of the browser and code. Preserve sample data for testing. Verify the complete flow and show missing, failed, and empty data states clearly. TRY WITH Let the checklist read a small, approved equipment list from a file containing only fictional items. Before connecting anything, explain what is read, where it is sent, and how the page behaves if the file is missing or malformed. A USEFUL RESULT MIGHT LOOK LIKE The feature reads only the necessary item names, identifies the source, and shows a helpful error when it cannot load the file. The original sample-data version still works for testing. No secret is embedded in the page or shared source. CHECK IT - Do you understand and approve the exact data and access requested? - Test normal, empty, missing, and invalid source data. - Confirm that no secret or unapproved personal data appears in the browser, exported files, or shared repository. IF IT MISSES, SAY The connection fails silently. Add a visible error and a retry path, keep sample data available, and explain what was actually connected. Do not broaden permissions to work around a failure without asking. WITHOUT ACCESS Continue with a fictional file or sample data. A safe working prototype is better than an unnecessary live connection. IT WORKED WHEN The right source feeds the intended feature, access is understood, and failures are visible. WHAT YOU KEEP A verified data flow, clear access notes, and a safe test setup. YOUR PART Understand what data moves where and approve the access deliberately. DEVELOPMENT · LEVEL 4 Make quality checks repeatable Automate a small set of useful checks after running them by hand. Time: About 30 minutes to make a small test routine; verify its repeated trigger separately. Needs: A development environment that can save and run checks, with a supported after-change trigger or scheduler if you want automation. THE REUSABLE PROMPT Help me make the checks for [my small app] repeatable. Cover the main user flow, invalid and empty input, links and buttons, keyboard use, and a phone-sized layout. Explain the results simply. Where the environment supports it, run the checks after changes or on [approved schedule] and report failures with evidence. Keep a visual review against the brief. Don't claim a check or schedule exists until it is saved and tested. TRY WITH For the packing checklist, test adding an item, rejecting an empty item, ticking an item, updating the remaining count, and resetting. Include one long item, keyboard navigation, and a phone-sized view. Make failures understandable. A USEFUL RESULT MIGHT LOOK LIKE The saved tests run and report a real pass or failure for each behavior. A deliberately introduced harmless defect causes the relevant check to fail. The report distinguishes automated behavior checks from visual review performed by a person or agent. CHECK IT - Run the saved checks yourself and read the result. - Introduce one harmless known defect in a test copy, confirm it is caught, then restore it. - If a trigger is configured, make a change and verify a new run actually starts; a saved test alone is not a running automation. IF IT MISSES, SAY You described tests but did not execute them. Save and run the checks in the available environment, show the actual result, and label any visual or device check that remains manual. WITHOUT ACCESS Use a repeatable manual checklist and record the result. This is worthwhile quality control, even before automatic execution is available. IT WORKED WHEN The checks actually run, catch a known problem, and report the result clearly. WHAT YOU KEEP A repeatable test routine and visible failure reporting. YOUR PART Inspect the evidence and keep real-user usefulness alongside automated checks. DEVELOPMENT · LEVEL 5 Delegate one complete feature Hand over a bounded build with product intent, visual standards, and proof. Time: About 20 minutes to define a small feature; implementation and review time vary. Needs: An agent with access to the approved project and a safe test environment. A preview does not authorize public deployment. THE REUSABLE PROMPT Build [one bounded feature] for [existing tool and audience]. Read the relevant project context first. The outcome is [user job] and done means [acceptance tests]. Preserve the agreed visual direction and use [references] to guide hierarchy, typography, spacing, and interaction. Work with sample data unless real access is approved. Implement, test the full flow, inspect the result on desktop and phone-sized screens, critique and refine it. Return a working preview or runnable files, test evidence, limitations, and rollback instructions where relevant. Ask before costs, accounts, permissions, or public deployment. TRY WITH Add separate 'day trip' and 'overnight' lists to the existing checklist. Preserve the visual style, old list behavior, and no-account boundary. Work in a safe copy with sample data. Finish only when both lists and the original controls pass the agreed checks. A USEFUL RESULT MIGHT LOOK LIKE The agent returns a working preview or runnable files, the changed feature, results from old and new tests, phone and keyboard checks, and a clear way to restore the previous version. Public release remains your decision. CHECK IT - Complete the full flow in each list and verify that switching lists does not mix their items. - Check existing features, refresh behavior, phone layout, and keyboard use. - Review the changes, remaining limitations, and rollback steps before approving release. IF IT MISSES, SAY The new lists work, but existing items disappear on refresh. Investigate and fix that regression in the safe copy, rerun the full agreed checks, and report the actual results before calling it complete. WITHOUT ACCESS Use the same brief to guide a step-by-step build you supervise. Keep deployment, costs, new accounts, permissions, and real data behind an explicit decision. IT WORKED WHEN The feature is usable, visually intentional, verified end to end, and ready for your decision about release. WHAT YOU KEEP A complete feature, a verification report, and a reusable handoff pattern. YOUR PART Use the whole feature, understand its limits, and decide whether it is ready to release. OFFICIAL ADVANCED SETUP AND CONTROL GUIDES Checked October 10, 2026. Access and interfaces can change. dots: connect computers and apps https://learn.chatgpt.com/docs/dots/computers-and-apps Grok Bot: first task and tool access https://docs.x.ai/grok-bot/get-started dots: recurring tasks https://learn.chatgpt.com/docs/dots/tasks-and-memory Grok Bot: skills and routines https://docs.x.ai/grok-bot/skills-routines-and-automations dots: review work and stop controls https://learn.chatgpt.com/docs/dots/controls Grok Bot: approval and privacy controls https://docs.x.ai/grok-bot/approvals-security-and-privacy CROSS-SERVICE BOUNDARY Different services do not automatically share context or memory. Within Grok Bot, separate Bots share computer files and signed-in sessions; they are not separate security boundaries. https://docs.x.ai/grok-bot/approvals-security-and-privacy PERSONALIZATION PRACTICE DEMO These are prewritten fictional examples, not completed AI work. Choose a context, a 10-, 20- or 45-minute practice budget, and a style of help. A time budget is not a promise of completion. The website does not send or save your choices. SMALL STEPS Give me three small, ordered steps. Make the first action clear and name where to stop. Keep me involved in checking the result. ASK ME FIRST Start with the single question that would most improve your help. Wait for my answer before giving a plan or doing any work. Do not invent my answer. SHOW ME OPTIONS Show two genuinely different approaches that fit the time and constraints. Explain the tradeoff briefly, recommend one with a reason, and let me choose before proceeding. EVERYDAY LIFE I keep putting off sorting my kitchen drawers. No budget for organizers. Emptying everything at once makes it harder for me to finish. The aim is to find the utensils I use, not make the kitchen perfect. Use this fictional practice situation as written, or replace it with my real situation before we start. 10-MINUTE PRACTICE Focus: Make one drawer easier to use. - Choose the drawer you open most often. Leave the others closed. - Move the utensils you use regularly to the front. Do not empty the drawer. - Try finding three things you often need. Stop when your ten-minute window ends. Ask-first question: Which drawer causes the most frustration: cooking utensils, cutlery, or mixed odds and ends? Option: Find things faster. Put the five items you use most at the front. You get easier access; the rest can stay untidy. Option: Remove one source of friction. Clear the object that jams the drawer, then check it opens freely. You fix a specific annoyance without reorganizing everything. Check: Can you find the three utensils you reach for most? Stop: Leave the other drawers alone. Do not start a shopping list. 20-MINUTE PRACTICE Focus: Sort one drawer without creating a bigger job. - Pick one drawer and identify the utensils you use most. - Work on one small area at a time: remove obvious trash and group the useful items using the space you already have. - Put everything back, try finding three regular items, and stop at twenty minutes. Ask-first question: Do you mainly need easier access to daily utensils, or a way to stop miscellaneous items taking over? Option: Daily-use zone. Put frequently used utensils together at the front and occasional items behind. Best when finding things is the problem. Option: One clear group. Sort just one type of item, such as measuring spoons, and leave the rest in place. Less ambitious if a whole drawer feels overwhelming. Check: Does the drawer close, and can you find the regular items without searching? Stop: Stop at one drawer and twenty minutes. No new storage purchases. 45-MINUTE PRACTICE Focus: Finish one drawer, test it, then decide whether to do another. - Sort the most-used drawer in small sections, keeping daily items easy to reach. - Put everything back and try the layout before changing a second drawer. - If useful time remains, apply the same pattern to one more drawer. Reserve the final few minutes for putting things away. Ask-first question: Would a second drawer help most, or would you rather spend the whole session making one drawer work well? Option: One drawer, thoughtfully. Sort one drawer, test the layout, and adjust what still feels awkward. More attention to a result you will keep using. Option: Two small wins. Finish and test one drawer before starting a second. Wider progress, but only if the first drawer is fully back together. Check: Are all items put away, and is the new arrangement easier to use? Stop: Do not open a second drawer until the first is usable. Stop when the practice window ends. KNOWLEDGE WORK I need a useful internal update for a fictional community-event team. Keep it under 120 words. Separate known facts from open questions. Prepare a draft only; do not send it or invent an owner’s commitment. Fictional notes: The community room is booked. A poster draft exists but has not been approved. We still need to choose whether the event starts at 10 or 11. Sam offered to check the equipment list; no completion update is recorded. The next team check-in is Thursday. Do not assume anything happened beyond these notes. 10-MINUTE PRACTICE Focus: Make one short draft from the supplied facts. - Separate the notes into confirmed progress and open questions. - Draft a brief update with those two headings. Keep the start-time choice visible. - Check each statement against the notes and leave the draft unsent. Ask-first question: Who needs this update: the whole volunteer team or the person deciding the start time? Option: Status first. Lead with the booked room, then list the two unresolved items. Good for a reader catching up. Option: Decision first. Lead with the 10-or-11 start-time choice, then add brief context. Good when the decision is holding up work. Check: Does every factual statement come from the notes, and is the needed decision easy to spot? Stop: Stop at a reviewed draft. Do not send or assign unconfirmed owners. 20-MINUTE PRACTICE Focus: Create a focused update and improve it once. - Draft a progress/open-questions update under 120 words. - Compare a status-first version with a decision-first version. Choose one for the intended reader. - Check the start-time question, Sam’s actual status, and the word count; make one focused revision. Ask-first question: What should the reader do after reading: choose a start time, report missing information, or simply stay informed? Option: One clear ask. Center the start-time decision and keep other progress to one sentence. More focused, with less background. Option: Shared picture. Use short progress and open-question bullets. Broader context, with the decision still explicit. Check: Does the chosen structure support the intended response without inventing facts? Stop: Keep it under 120 words and draft-only. One revision is enough for this practice. 45-MINUTE PRACTICE Focus: Develop a repeatable update pattern from one good example. - Draft the update, test whether a reader can find the important decision, and revise it. - Build a short reusable template: confirmed progress, open questions, decision needed, and next check-in. - Try the template on a second fictional set of notes and inspect whether it separates uncertainty correctly. Ask-first question: Do you want to improve this one update, or develop a template for recurring team updates? Option: Reader-centered revision. Use the session to refine this update for a specific reader and response. A stronger one-off draft. Option: Reusable template. Keep this update as an example and test its structure on another fictional case. A reusable starting point that still needs factual checks each time. Check: Does the template prompt for unknowns instead of filling them in, and does the second draft still pass the fact check? Stop: No sending. Keep the method only after it works on a second example. DESIGN I am practicing with a fictional Neighborhood Repair Day flyer. The reader should understand what, when, where, and what to bring. Use readable type and strong contrast; no decorative text that competes with the details. Use original elements or references I have permission to share. Keep the example labeled as fictional. Fictional event copy: Neighborhood Repair Day. Saturday, 10am–noon. Community Room. Bring one small household item. Questions: repair@example.com. Add a small ‘Fictional practice event’ label. Do not invent a date, address, sponsor, price, or booking link. 10-MINUTE PRACTICE Focus: Make the information hierarchy clear before decorating. - Arrange the supplied copy into three groups: event name; time and place; what to bring and contact. - Choose one headline size and one readable body size. Use a simple high-contrast layout. - Read it at the size someone will see it. Check that no required fact is missing. Ask-first question: Will someone see this mainly on a phone or on a printed noticeboard? Option: Type-led notice. Use a strong headline and simple information blocks. Quick to judge and keeps the details prominent. Option: One simple symbol. Add one small original repair symbol beside the title. A little personality, provided it does not shrink the important text. Check: Can you immediately find what, when, where, and what to bring? Stop: Treat this as a first layout. Keep the fictional label and do not publish it. 20-MINUTE PRACTICE Focus: Compare two layouts and refine the clearer one. - Make a compact type-led direction and a more open direction with one simple symbol. - Choose based on reading order, spacing and the intended viewing size, not just color. - Refine the chosen layout once. Check every supplied detail, contrast and text size. Ask-first question: Should the flyer feel more like a straightforward community notice or a friendly invitation? Option: Straightforward notice. Use a tight hierarchy and a clear details block. Efficient scanning with little visual distraction. Option: Friendly invitation. Use more open space and one purposeful visual element. Warmer tone while keeping the same essential facts. Check: At its real size, do important details stay readable and keep the right order? Stop: Refine one direction rather than mixing every idea. No public sharing. 45-MINUTE PRACTICE Focus: Make and inspect an export in its actual setting. - Choose a visual direction using permitted references and a clear reason for the choice. - Build the flyer, then open an exported version at its intended print or screen size. - Fix the largest reading problem. If time permits, adapt the same hierarchy to a second size and check again. Ask-first question: Which output matters first: a printable page or an image for a phone screen? Option: One polished format. Spend the session making one viewing format clear and reliable. More quality checking where the flyer will actually be used. Option: Two related sizes. Finish the primary format, then adapt it without shrinking everything. Broader usefulness, but the second size needs its own reading check. Check: Open the actual export: no clipped text, invented details, unreadable contact line, or missing fictional label. Stop: Do not call the work checked until you inspect the export. Do not publish the practice flyer. ADMINISTRATION I want to find likely duplicate contacts without deleting different people’s records. Use only the fictional rows below. Matching names alone do not prove a duplicate. Propose a review table; do not delete, overwrite, or contact anyone. Fictional CSV, with intentional inconsistencies: id,name,email 1,Casey Dunn,casey@example.com 2,Casey Dunn,CASEY@example.com 3,Morgan Lee,morgan.one@example.com 4,Morgan Lee,morgan.two@example.com 5,Alex Park,alex@example.com 6,Alex Park, 7,Riley Stone,riley@example.com 8,Casey Dunn, casey@example.com For this practice, matching email after trimming whitespace and ignoring case is a candidate rule, not permission to merge. Row IDs must remain traceable. 10-MINUTE PRACTICE Focus: Review one likely duplicate group. - Keep the original row IDs and compare email text after trimming spaces and ignoring case. - Inspect the Casey Dunn rows as one candidate group. Show the reason for the match. - Leave same-name/different-email and missing-email rows unresolved rather than guessing. Ask-first question: For this fictional practice, should matching normalized email flag a review candidate, while matching names alone stay uncertain? Option: Exact repeats only. Flag only records whose displayed name and email match exactly. Fewer questionable matches, but case and spacing variants may be missed. Option: Normalized-email candidates. Ignore email case and surrounding spaces when flagging candidates. Finds likely repeats, but still requires review before any merge. Check: Can you trace each proposed match to row IDs, and are Morgan’s different emails left separate? Stop: No deletion or merge. Stop after reviewing one group. 20-MINUTE PRACTICE Focus: Review the whole small sample with explicit uncertainty. - State the candidate rule before applying it to the eight fictional rows. - Return a review table with row IDs, candidate group, reason, uncertainty and proposed next step. - Check the Casey group, the two Morgan emails and Alex’s missing email against the original sample. Ask-first question: Would you prefer a cautious list of strong candidates or a wider list that also includes uncertain cases? Option: Strong candidates. Show the normalized-email matches, with other rows unchanged. A shorter review queue. Option: Strong and uncertain. Separate normalized-email matches from same-name or missing-email questions. More coverage, with a larger review queue and no implied merge approval. Check: Are strong candidates and uncertain cases labeled differently, with all eight row IDs accounted for? Stop: Keep the result as a review proposal; do not overwrite the source. 45-MINUTE PRACTICE Focus: Turn the reviewed rule into a small repeatable check. - Review the sample and agree on the candidate rule and uncertainty labels. - Ask for a short checklist or clearly explained formula using a copy of the data, then test it on the sample. - Add a fictional near-match with a different email and check that it is not silently merged. Save the rule only after reviewing the result. Ask-first question: Do you want a manual review checklist, or an explained formula for a spreadsheet app you already use? Option: Manual checklist. Save the comparison rule, exception list and human approval step. Portable and easy to understand for a small sheet. Option: Explained formula. Build and test a candidate flag in an existing spreadsheet tool. More reusable for larger sheets, with setup and testing still required. Check: Does the check flag candidates without losing row IDs or treating near-matches as confirmed duplicates? Stop: Work on sample data or a copy. No real connection, bulk deletion, or automatic merge. DEVELOPMENT I want to learn by making a small packing checklist for a day trip. Use fictional items; no accounts, external services, personal data or paid tools. I choose the plan and try each small change myself. If the assistant cannot create a working preview, it must explain the available setup instead of claiming it built an app. Fictional test items: water bottle, snack, rain jacket. Start with one screen. An item can be added and marked packed. Do not add accounts, reminders, purchases, sharing, or a backend. Explain whether the list survives a refresh. 10-MINUTE PRACTICE Focus: Choose one tiny flow and understand its first test. - Ask the assistant to propose only an add-and-tick flow, in everyday language. - Choose or correct the plan before any build starts. - If a preview is available, try one fictional item. Otherwise, identify the next supported setup step without installing anything. Ask-first question: Would you like to sketch the add-and-tick behavior first, or try a small preview if this app supports one? Option: Plan and test case. Describe one screen and what should happen when you add an item. Useful without coding or preview access. Option: One supervised preview. Try adding and ticking one item in an available preview. Hands-on, but only if your existing tool supports it. Check: Can you explain what the first flow should do, and what would count as a failure? Stop: No promise to finish a working app in ten minutes. Ask before any new setup or permissions. 20-MINUTE PRACTICE Focus: Try the smallest working flow and make one correction. - Approve a one-screen plan for adding and ticking fictional items. - Open a preview if supported; add ‘water bottle’ and tick it yourself. - Describe one problem you notice, ask for the smallest fix, and try that flow again. Ask-first question: Would you most like help understanding the build steps, or making the first screen easier to use? Option: Behavior first. Test adding and ticking before changing the visual styling. Prioritizes a flow that actually works. Option: Clarity first. Use the same tiny flow but focus your one correction on labels, spacing or feedback. Helps if the controls work but feel confusing. Check: Does an added item appear, can it be marked packed, and does the correction solve the observed problem? Stop: Keep it to add and tick. If the preview is unavailable, stop and clarify setup rather than claiming success. 45-MINUTE PRACTICE Focus: Test edge cases and refine one small prototype. - Approve and try the add-and-tick flow with the fictional items. - Test an empty item, a long item, keyboard use, a narrow screen and refresh. Record what actually happens. - Fix the biggest problem and retest it. Add a reset control only if the original flow is reliable and you still have time. Ask-first question: Would you rather use this session to improve readability and keyboard use, or to understand and test how the list is saved? Option: Usability pass. Focus on readable text, clear feedback and using the controls with a keyboard. Improves everyday use without adding more features. Option: Persistence pass. Clarify whether refresh clears the list, then test the claimed behavior. Improves confidence about what is and is not stored. Check: Can you name the tests you performed, their results, and any limitation still present? Stop: The budget is for practice, not a finished-app guarantee. No accounts, real data or public deployment. Keep consequential actions behind your approval. Do not treat a response as evidence of completed work or verified results. Prompts do not by themselves create memory, connections or schedules. ILLUSTRATIVE HUMAN CONTEXT EXAMPLES Fictional examples, illustrated with AI-generated editorial portraits. These are not testimonials or live AI responses. MAYA — A home design studio Help me practice with this fictional situation. I can replace these details with my own before we begin. Maya works from a home design studio. Two client drafts are due Thursday. Her clearest creative time is in the morning. School pickup is at 3, and she wants evenings free. Her request is: Help me get my week under control. Ask one question at a time if a missing fact would materially change the plan, such as the size of the drafts or existing appointments. Do not invent the answers. Propose a light weekly plan that protects the creative work and the pickup/evening boundaries. Group smaller work only where it genuinely fits. If the priorities do not fit, show what needs a decision rather than adding more hours. Keep known facts separate from assumptions. Give no more than three useful next steps, then ask what you got wrong. Check the plan against the stated constraints. This is a draft only: do not send messages, change a calendar, create reminders, or make commitments. The person chooses priorities and approves any change. DANIEL — A kitchen table before work Help me practice with this fictional situation. I can replace these details with my own before we begin. Daniel’s work shifts can change each week. He needs to leave room for travel and rest. He would like to keep learning in short, realistic sessions. His request is: Help me get my week under control. First ask for the confirmed shifts, travel requirements, protected rest windows, and what he wants to learn. Ask one question at a time and do not invent the answers. Build a light plan around that information. Suggest at most two 20-minute learning sessions to start, only where they fit, and a fallback if the shifts change. Do not impose the same clock time every day or fill every free hour. Keep known facts separate from assumptions. Give no more than three useful next steps and ask what needs to change. Check that work, travel and rest still fit. This is a draft only: do not edit a calendar, create reminders, enroll in a course, spend money, or make commitments. The person chooses the goal and how much to take on. RUTH — A greenhouse planning bench Help me practice with this fictional situation. I can replace these details with my own before we begin. Ruth has two garden projects in mind and wants to make time for a friend. She likes a useful starting point but wants much of each day left open. Her request is: Help me get my week under control. Ask what the two projects are and whether any commitments are fixed. Ask one question at a time and wait. Then suggest one simple anchor for each of the next three days, with flexibility for weather and her preferences. Do not invent a forecast or assume that every open hour needs a task. A one-line note of the next step is enough unless she wants more structure. Keep known facts separate from assumptions. Give no more than three useful next steps, then ask what feels too planned or not useful. Check that the plan preserves open time and real choice. This is a draft only: do not send invitations, change a calendar, buy supplies, create reminders, or make commitments. The person decides what makes the day worthwhile.