
Turn one worthwhile goal into a next action
A practical guide to choosing a clear output, fitting the work to your capacity, checking dependencies and revising from evidence. Includes three complete fictional plans and six independent exercises.
Start small: choose a personal creative, organising or hobby goal. Keep medical, financial, legal and other high-impact decisions outside this general planning method.
Choose a reading route
Foundations: define, choose and execute
- 1. Give your goal a route into the real world
- 2. Separate purpose, outcome, output and process
- 3. Capture the starting point without oversharing
- 4. Expand possible routes before choosing one
- 5. Work backwards, then check forwards
- 6. Fit the plan to capacity and uncertainty
- 7. Make the next action easy to start and safe to stop
- 8. Review evidence and revise deliberately
Complete worked plans
Repair, practise and use independently
1. Give your goal a route into the real world
A goal can matter deeply and still fail to tell you what to do next. “Finish my photo collection”, “write regularly” and “start a reading group” all point towards something worthwhile. None tells you which file to open, what to produce in the next half hour, or how to respond when somebody else is unavailable. The gap between wanting and doing is often a missing translation: from an attractive future to a sequence of actions that a real person can carry out under real constraints.
Super Intelligence can help with that translation when you give it a bounded job. Ask it to clarify the result, expose assumptions, suggest alternative routes, identify dependencies and test whether the actions fit your available time. Then inspect the plan yourself. A fluent plan can be wrong about what matters to you, the effort involved, the information available or the permissions required. The plan becomes useful through checking and execution, not through how confidently it is written.
In the eduKateSG Life series, Super Intelligence, or SI, is a practical editorial umbrella for using contemporary AI tools, information and repeatable workflows deliberately. It does not mean that a present-day assistant has been established as artificial superintelligence. Technical ASI refers to a hypothetical broadly superhuman form of machine intelligence. This guide is about ordinary personal planning with tools available to a reader, with human judgement, consent and verification kept in place.
Choose a low-stakes goal for your first attempt: completing a creative project, organising a personal collection, preparing a hobby gathering or establishing a manageable reading routine. A medical treatment decision, investment strategy, legal deadline or consequential decision about someone else needs domain-specific expertise and stronger safeguards. A well-formatted checklist does not make those choices safe to delegate. This article does not provide medical, financial or legal prescriptions.
The method follows the Life loop: Capture → Clarify → Expand → Structure → Execute → Review → Improve. Capture the wish and the facts you already have. Clarify what completion would mean. Expand the possible routes before committing. Structure a route into dependencies and next actions. Execute a bounded step. Review the resulting evidence. Improve the next version of the plan. These stages are a way to keep reality involved; they are not a requirement to complete seven documents before starting.
Here is the smallest useful version. Write one sentence describing the result you want. Write what is true now. Choose one action you can complete with what you already have. Specify the evidence that would show it is done. Choose when you will inspect the result and decide what follows. If an assistant produces a twenty-step plan but cannot answer those five questions clearly, simplify the plan before adding more detail.
For example, “organise my photos” becomes “create a private twelve-image collection from one outing, with accurate captions, by the end of the next two weekends”. The first action might be “open the outing folder and mark twenty candidates, without moving or deleting any originals”. The completion evidence is a candidate list, not a vague feeling of progress. The next checkpoint asks whether those candidates tell the intended story, before time is spent on layout.
Notice what has changed. The goal now has a boundary, a first action and a test. It still leaves room for taste and discovery. You have not promised that the collection will delight everyone or that each stage will take an exact number of minutes. You have created a route that can be inspected and revised. That is the central purpose of an action plan: enough structure to support the next meaningful move, with enough flexibility to learn from it.
2. Separate purpose, outcome, output and process
Begin by separating four things that often get mixed together. Your purpose explains why the goal is worth pursuing. Your outcome describes the change you hope to see. Your output is a thing you can make or an event you can arrange. Your process is the set of behaviours intended to produce that output and support that outcome. These are working distinctions for planning, not claims that every personal ambition fits neatly into a formal model.
Suppose your purpose is to spend more enjoyable time discussing books. Your desired outcome is a satisfying conversation with interested friends. Your output could be one small reading session with an agreed text and three discussion questions. Your process includes checking interest, proposing options, reading the text and preparing the questions. You can control whether you prepare and send a respectful invitation. You cannot control another person’s enthusiasm or availability.
That distinction protects the plan from making promises about other people. “Have four people attend” is partly outside your control. “Ask four people whether they would be interested, accept no gracefully, and decide by Friday whether a smaller session is worthwhile” is a controllable action sequence. It supports the outcome while acknowledging uncertainty. If nobody is available, the plan can produce a useful decision to postpone rather than a false record of personal failure.
An output also needs a quality test. A document exists as soon as you save a blank file, but that rarely means the goal is complete. “Finish a short story” might mean a complete beginning, middle and ending; a deliberate revision for clarity; and a saved version that you are willing to reread after a break. A word count alone cannot stand in for those qualities. Conversely, an endlessly moving standard such as “make it perfect” prevents a reasonable stopping decision.
Use a completion sentence that names the deliverable, the boundary, the acceptance test and the review point. For instance: “By the fourth Sunday, I want a private 1,200–1,600-word story with a complete ending, one structural revision and a final read-through, after which I will decide whether to share it.” Sharing is a separate choice. The writing plan does not quietly become permission for an assistant to publish the story or send it to a friend.
A process target should also be modest enough to inspect. “Work on it every day” hides what counts as work and whether daily sessions fit your life. “Use two planned forty-minute sessions this week to draft the opening and test an ending” is clearer. You can record whether the sessions happened and whether they produced useful material. If the second session reveals that the ending does not work, that is information about the project rather than proof that the entire week was wasted.
Ask SI to check the distinctions rather than merely polishing the goal. A useful request is: “Identify the purpose, outcome, output and process in this goal. Mark anything that depends on another person’s choice. Show where a count could look good while the real purpose is missed. Do not invent my motivation.” The last instruction matters. An assistant can suggest possible purposes for you to consider, but it cannot know that your hobby is secretly about confidence, career change or family expectations.
Before moving on, answer a final question: if the output is completed exactly as described, would you still value it? If not, the specification is probably a proxy that has drifted away from the purpose. Perhaps you want to enjoy drawing but have defined success as posting a picture every day. Perhaps you want a usable personal archive but have made elaborate folder labels the main objective. Repair the definition before turning it into a schedule.
3. Capture the starting point without oversharing
An assistant cannot build a realistic route from an invented starting point. Give it a short current-state packet: what exists already, what is missing, what resources you can use, what time is genuinely available and what you do not want the project to consume. A paragraph of accurate constraints is usually more valuable than several pages of personal biography. Include the information that changes the plan; leave out material that merely makes the conversation feel more intimate.
For a photo collection, useful facts include the number of candidate images, whether they are already copied somewhere safe, the intended audience and the software you already know. The assistant does not need a complete upload of your private photo library to help structure the project. You can describe the collection using invented filenames and anonymous captions. If identifying images or private information would be shared with a service, make that a deliberate decision based on its settings and your obligations to the people involved.
Separate facts, estimates, preferences and unknowns in the packet. “I have forty candidate files” is a fact you can check. “Captioning each will take two minutes” is an estimate. “I prefer a plain layout” is a preference. “I do not know whether the export keeps captions readable” is an unknown. When these categories are mixed together, a generated plan may treat a guess as a requirement or turn a preference into a rule you never intended to impose.
A useful baseline is an observation that helps you judge change. Before planning an elaborate writing routine, notice how your last two sessions actually went. Did you spend most of the time choosing a topic, searching for a missing file or revising the same paragraph? You do not need a month of tracking to begin. One honest observation can expose the immediate obstruction. The next action can then remove that obstruction rather than asking you to be more disciplined in the abstract.
Also state the limits of the assistant’s knowledge. If it cannot see your calendar, it should not claim that a proposed slot is free. If it has not inspected a file, it should not say the file is complete. If it has access to only part of a collection, it should label the inspection as partial. You can request a plan based on declared assumptions, but the assumptions should remain visible until checked. A confident summary is not a substitute for access or evidence.
NIST’s Generative AI Profile identifies confabulation among generative-AI risks, including confidently presented false content. That is a practical reason to verify factual details that determine a plan, such as a venue’s rules or a tool’s export behaviour, rather than accepting them because an assistant supplied them. The profile is a general risk-management source, not an endorsement or evaluation of the personal planning method in this article. NIST Generative AI Profile
Keep the packet small enough to refresh. A working example might say: “Goal: one private photo booklet. Existing material: forty files from one outing. Available time: three forty-minute sessions this week and two next week. Constraints: no new software, no spending and no deletion of originals. Unknowns: whether twelve images are enough and whether captions are readable in export. Authority: help plan and review anonymous samples; do not upload, share or modify the originals.” This gives the assistant real planning material without granting broad access.
At the end of Capture, you should be able to point to the current state rather than merely describe your ambition again. If you cannot locate the material, the first action is to locate it. If you have not chosen which collection or story matters, the first action is to choose a bounded candidate. A plan should start at your actual starting line, including uncertainty and unfinished preparation, rather than at the point where the project would be easier to describe.
4. Expand possible routes before choosing one
A first draft plan often inherits the first solution mentioned in the prompt. If you ask how to build a complex personal dashboard, you may receive a complex dashboard plan even when a one-page list would solve the problem. Before committing, ask SI for a few genuinely different routes to the same purpose. The aim is not an endless brainstorming session. It is to make one deliberate choice about how much machinery the goal really needs.
For a personal collection, three routes could be a simple folder with a text index, a private slide deck, or a designed booklet. Each supports a different balance of effort, presentation and maintenance. The folder is easy to update but less polished to view. The slide deck provides a visible sequence using familiar tools. The booklet may be satisfying as a finished object but require more layout work. None is automatically superior; the right route depends on the reason you want the collection.
Compare routes against a few criteria you can actually assess: can you use the tools already available, can you inspect the result, can you finish a small version within your capacity, and can you reverse the choice if it proves unhelpful? Avoid assigning precise scores to guesses just to make the comparison look objective. A sentence such as “layout effort is uncertain because I have never used this template” is more honest than giving the route a confident efficiency score of eight.
Ask for a smallest useful version of each route. A reading group can begin as a conversation between two willing adults about one short text. A writing project can begin as one complete scene with an ending rather than a detailed trilogy outline. A personal archive can begin with one clearly labelled collection. The small version should preserve the purpose; it should not be so reduced that success becomes meaningless. One correctly named file does not establish a usable archive unless that was genuinely the whole problem.
An exploratory action is appropriate when the route depends on an unknown. For a booklet, create one sample page and export it before laying out twenty pages. For a reading session, ask whether people want a discussion before designing invitations and schedules. For a writing project, test a short ending before polishing the opening for hours. These actions produce information that changes the plan. Their value lies in the decision they enable, even if the final route is different.
Make the choice explicit. Record: “I am choosing the private slide deck because I know the tool, want an ordered story and do not need print production. I will reconsider if captions cannot remain readable.” This prevents a later assistant from expanding the project into a website simply because websites are possible. The rejected options can stay in one line for reference; they do not need to become parallel tasks. An option becomes work only after you decide it belongs in the plan.
It is also legitimate to decide that the goal should wait. You may discover that the purpose is weak, the available time is already committed or the project depends on somebody who has not agreed. A clear deferral with a revisit condition can be a good planning outcome. “Reconsider after the current story is finished” is different from “keep this active indefinitely”. The former protects attention. The latter creates a permanent sense that you are behind on something you never properly chose.
5. Work backwards, then check forwards
Working backwards starts with evidence of completion. If the goal is a private booklet that opens correctly and tells a coherent story, what must be true immediately before that? The export must exist, its pages must have been checked, the sequence must be settled and the captions must be accurate. Each requirement suggests earlier work. This is more reliable than beginning with a long list of generic verbs such as research, organise, prepare and improve.
A milestone describes a meaningful state, not merely time spent. “Image selection approved” is a milestone if you have a specific selected set and a reason for the sequence. “Spent two hours browsing images” describes effort. Effort is useful to record, but it may not move the project into a new state. When a plan has only effort milestones, it can report progress forever without showing whether the intended result is becoming possible.
Translate each milestone into actions using a concrete verb, an object and a done test. “Review images” becomes “compare the twenty candidates against the twelve-image limit and mark the final twelve, keeping the original files unchanged”. “Work on story” becomes “draft the final confrontation in a separate document, stopping when the main character makes a choice”. The action does not have to be tiny, but it should be clear enough that you can tell whether it happened.
Then map dependencies. Some actions require outputs from earlier actions: captions cannot be finalised until the images are selected. Others only appear dependent because of habit: you may test export readability with a temporary sample before choosing the final images. A useful question for SI is: “Which dependencies are logically necessary, which depend on my chosen method, and which are merely suggested ordering?” Removing a false dependency can let you test a risky assumption earlier.
Treat waiting separately from doing. Asking a friend whether they want to join a reading session takes a small amount of active work. Waiting for a reply may take days. The schedule must represent both, but the waiting period is not a task you can complete by concentrating harder. Record the request, the expected response point and a fallback. Do not silently assign the other person a deadline they have never agreed to meet.
After building the backward chain, walk forwards from the current state. Do you have the files required for the first action? Does the next action use an output that the previous action actually creates? Is there a hidden permission step? Does a quality check arrive before the irreversible action? Forward checking catches plans that look coherent at the milestone level but fail when somebody tries to execute them.
For example, “collect photographs → design booklet → share booklet” misses selection, factual caption checking, export checking and a separate sharing decision. A repaired route could be: inspect copies, choose a small set, verify captions, test a sample layout, complete the layout, inspect the export, then decide whether sharing is appropriate. You may combine steps during execution, but keeping the tests visible reduces the chance that a polished layout masks an incorrect caption or an unintended disclosure.
Finish with one ready action rather than twenty equally urgent ones. An action is ready when its inputs, permissions and required resources are available. If the first step is blocked, identify the action that can remove the block, or choose an independent ready step. A plan with every item marked “start now” is not prioritised. It has simply moved the entire project into the present.
6. Fit the plan to capacity and uncertainty
A plan needs a capacity budget as well as a task list. Start with the time you are genuinely willing to allocate after existing commitments, rather than every theoretically empty minute. A personal project competes with rest, travel, household responsibilities and other chosen activities. Treating those as errors in the calendar produces an unrealistic plan. The purpose of planning is to fit meaningful work into a life, not to erase the rest of that life.
Separate active effort from elapsed time. You might need thirty minutes to prepare a sample page, a day before you can revisit it with fresh eyes, and ten minutes to make a decision. The total active effort is forty minutes. The elapsed time is longer. An assistant that packs the two sessions together because the arithmetic fits has missed the reason for the gap. Conversely, waiting for a file export does not necessarily occupy your attention for the entire interval.
Use ranges for unfamiliar work, and label where they came from. “This could take twenty to forty minutes because I have completed a similar selection before” is more informative than “twenty-three minutes”. If there is no basis, call it an initial estimate and test a small unit. After captioning three images, you may learn whether the work involves simple labelling or checking old notes. Update the remaining estimate using that observation; do not preserve the first guess merely because the plan looks tidy.
Leave some unallocated capacity when uncertainty is meaningful. There is no universal buffer percentage that makes every personal plan safe. Choose a reserve that reflects the unfamiliarity, consequences and flexibility of this project. If your two-hour session is the only opportunity before an important external deadline, a completely full plan is brittle. If the project is a flexible hobby, a smaller first version or a later finish may be more sensible than aggressive time compression.
A simple arithmetic check prevents many avoidable problems. If you have four forty-five-minute sessions, your gross capacity is 180 minutes. If you deliberately reserve thirty minutes for repairs, the planned work budget is 150 minutes. A plan containing 175 minutes of required work does not fit merely because each individual task seems short. You must reduce scope, add genuinely available capacity or move the review point. Asking SI to make the schedule “more efficient” cannot create the missing twenty-five minutes.
Ask the assistant to show the total and assumptions in plain language. A useful instruction is: “Add the active minutes, include setup and review, and keep waiting time separate. Compare the total with my available sessions. If it does not fit, show three scope choices instead of silently shortening every estimate.” You remain responsible for choosing which compromise matches the purpose. The assistant can reveal a trade-off without deciding that your rest or another commitment should be sacrificed.
Keep a minimum version and an optional extension distinct. The minimum version must still meet the stated purpose and quality threshold. For a twelve-image booklet, the minimum might be twelve readable pages with accurate captions. Decorative maps, extra essays and a printed cover can be optional. If those extras become necessary only because the assistant suggested them, remove them from the required route. Otherwise the plan can expand faster than you can execute it.
Finally, test the schedule against context, not just totals. A forty-minute action requiring your laptop cannot be placed in a fifteen-minute gap on a bus. A task requiring a quiet room may not fit a shared evening space. Include setup, shutdown and the ability to resume. Sometimes the best repair is to make the action smaller and more portable; sometimes it is to accept that the action needs a different window. Both are better than labelling the person unreliable for missing an impossible slot.
7. Make the next action easy to start and safe to stop
A ready action should include a start cue. Instead of “write sometime on Saturday”, choose a realistic moment such as “after breakfast on Saturday, open the scene document at the marked paragraph”. This removes one small decision at the point of starting. Keep the cue under your control where possible. “When everyone else becomes quiet” is a hope; “when I reach the reserved desk session” is a clearer condition if that session is actually available.
Researchers Peter Gollwitzer and Paschal Sheeran describe implementation intentions as if–then plans linking a situation with a goal-directed response. Their overview distinguishes intending to reach a goal from specifying when and how to act. That supports using a concrete cue-and-action statement, without implying that a prompt or an if–then sentence guarantees success for any particular reader. The examples and workflow here are our own practical applications. Implementation Intentions, hosted by the US National Cancer Institute
Write the action so that stopping also preserves progress. “Select candidates until the timer ends, then save the marked list and write the next comparison” is safer than “keep working until it feels finished”. A short shutdown note can say what changed, what remains uncertain and what to open next. It reduces the cost of returning to a project after an interruption. It also gives SI current evidence if you later ask for help revising the plan.
Use a fallback that preserves the purpose rather than pretending the original action happened. If a forty-minute session disappears, you might spend ten minutes locating the exact source material and identifying the next choice. Record that as preparation, not as a completed draft. If the fallback occurs repeatedly, review the plan. A fallback is a recovery route for disruption; it should not become a device for reporting full completion while the meaningful work never happens.
Distinguish assistance from authority during execution. The assistant may draft a checklist, identify a contradiction or suggest a smaller action. That does not grant it permission to message friends, buy materials, upload private files or alter a shared document. Put any consequential or externally visible step behind a deliberate decision. A planning assistant that can use tools still needs the appropriate access, explicit scope and verification before it can act on your behalf.
Maintain a short status vocabulary: ready, in progress, waiting, blocked, done, deferred and cancelled. “Waiting” means someone or something else must respond. “Blocked” means a missing input or decision prevents the action. “Done” means the agreed test passed. These states are useful only when they reflect reality. If every task is labelled in progress for weeks, the list no longer tells you where to intervene.
Do not confuse a draft with a finished action. A generated invitation is not a sent invitation. A suggested time is not an agreed meeting. A saved export is not a checked export. A status update should name the evidence: “draft prepared for review”, “message sent to the approved recipients”, “two people confirmed”, or “file opened and all twelve captions inspected”. That language lets the next step start from facts rather than from a story of progress.
A good execution session ends with one of three useful results: a completed output, a newly answered question, or a clearly identified blocker. All three can advance the project. A session that produces only more optional ideas may be enjoyable, but it should not automatically be counted as execution. When planning starts replacing action, close the chat and do the smallest ready step that your current information already supports.
8. Review evidence and revise deliberately
A checkpoint is a decision point with evidence, not a ceremonial moment to ask whether you feel productive. Specify what you will inspect and what decision it will support. At a booklet checkpoint, inspect a sample export for readability and decide whether the layout is usable. At a reading-session checkpoint, inspect actual replies and decide whether to proceed, simplify or postpone. At a writing checkpoint, inspect whether the draft has an ending and decide what kind of revision it needs.
Separate process evidence from outcome evidence. Three writing sessions are evidence that the process happened. A complete draft is evidence that an output exists. A story you still want to revise after rereading may be evidence that the output serves your purpose. None automatically proves the others. This distinction helps you diagnose whether the problem lies in starting, choosing useful actions, judging quality or choosing the goal itself.
Use an evidence-first review note: what was planned, what actually happened, what changed, what remains uncertain, and what decision follows. Keep the note short enough to maintain. For example: “Planned twelve captions in forty minutes. Completed eight in forty minutes because location notes were missing. The remaining four need source checking. Reduce decorative work and use the reserved session for checking; do not invent the locations.” That is more useful than “I need better focus next week”.
When the plan misses a checkpoint, diagnose before escalating effort. Was the action unclear? Was the estimate weak? Was a dependency overlooked? Did the available time change? Did the project stop being worth doing? Different causes require different repairs. More reminders will not fix an unavailable file. A more detailed checklist will not create somebody else’s consent. A tighter schedule will not resolve a purpose you no longer believe in.
Change one or two important variables at a time when possible. If you simultaneously change the goal, tool, schedule, scope and quality test, it becomes difficult to learn why the next attempt went differently. You do not need a scientific experiment for every hobby. You do need enough continuity to notice whether the repair addressed the actual obstruction. A useful revision note states the reason: “Version two reduces the image count because caption verification is taking longer, while preserving the one-outing story.”
Sometimes the right revision is a cancellation. A project can be abandoned thoughtfully because it no longer serves the purpose, requires disproportionate effort or depends on conditions that have not materialised. Close it cleanly: save useful material, remove obsolete reminders, note any commitments that need a respectful update, and decide whether there is a smaller alternative worth keeping. Cancellation is not the same as leaving a plan silently active and feeling responsible for it indefinitely.
When a goal is complete, perform a completion review once. Verify the acceptance test, store the result where you can find it and identify any maintenance it actually needs. A private booklet may require no ongoing project at all after a backup. A recurring reading group may need a fresh agreement for its next session. Do not let a finite goal turn into an endless system unless the continuing benefit justifies the maintenance.
The Improve stage should produce a reusable lesson rather than a character judgement. “Test exports before full layout” is reusable. “Ask for interest before choosing a date” is reusable. “I am bad at finishing things” is broad and unhelpful. Record the lesson at the level where you can change an action. The next plan should inherit better evidence, not a harsher description of yourself.
9. Worked plan: a private twelve-image photo booklet
This case is entirely fictional. Mira is an adult making a private digital booklet from one local outing. The numbers, files, observations and results are invented to demonstrate the method; they are not a report of an AI product test. She has forty photographs copied into a working folder, a presentation tool she already knows and a wish to create something she can enjoy without searching through the whole collection. She does not want new software, printing costs or public sharing.
Her purpose is to preserve a small, coherent memory of the outing. Her desired outcome is to be able to revisit it easily. Her output is a twelve-image digital booklet with short factual captions. Her initial acceptance test is: the file opens, contains twelve selected images in a deliberate sequence, has captions checked against her own notes, uses readable text, and contains no unwanted identifying detail. She will make a separate decision later about showing it to anyone else.
Mira’s current-state packet identifies forty candidates, one notes file and five forty-minute sessions over two weeks. Gross capacity is 200 minutes. She initially reserves forty minutes for uncertainty and repairs, leaving 160 minutes of planned work. Her working estimates are twenty minutes to inspect and shortlist, twenty to select and order, thirty to write and check captions, twenty to test a sample layout, forty to assemble, and thirty to inspect and correct the export. Those six estimates total 160 minutes.
The plan also records assumptions. The forty files are accessible copies, not the only originals. The notes contain enough information for accurate captions. The familiar tool can export a readable file. Twelve images are enough to tell the story. The first assumption is checked before editing. The third is checked with a sample before full layout. The second and fourth remain open questions at the selection checkpoint. This makes the plan testable without requiring every uncertainty to be resolved at the beginning.
Her first action is to open the working folder, confirm that the originals remain elsewhere, and mark twenty candidates without deleting anything. The done test is a saved list of twenty filenames. SI helps by suggesting selection questions: does this image add a different moment, is it legible at the intended size, and does it require a caption Mira cannot verify? The assistant does not need the actual private photographs to help formulate those questions.
At the next step, Mira reduces the set to twelve and creates a provisional sequence: arrival, three scenes along the route, details, a pause, and a closing view. She notices that two similar images add little. Replacing one with a quieter detail makes the sequence more varied. This is a human editorial choice, not a measurable fact that the model can settle. The milestone is reached when the twelve filenames and order are saved, not when the assistant says the selection is beautiful.
Caption work reveals a problem. Eight captions can be checked against the notes; four contain location guesses. Mira marks the uncertain names rather than treating them as confirmed. She spends the planned thirty minutes and completes eight captions. The remaining four need twenty more minutes of checking or simplification. The useful revision is to draw twenty minutes from the forty-minute reserve. She does not compress the final inspection or ask SI to supply plausible place names.
For one image, the exact location remains uncertain. The original draft caption says “the old market entrance”, but the notes do not establish that. Mira changes it to a visible description, “a shaded entrance beside the walking route”. The caption now says only what she can support. This illustrates a repair that preserves the project without manufacturing certainty. A personal memory can include “I do not remember exactly where this was” if that is the honest account.
The sample-layout test uses one image, one short caption and the intended page size. Mira exports it and opens the export on the device she expects to use. The text is too small, so she increases the caption size before building the remaining pages. The twenty-minute sample allocation includes that adjustment. This early test avoids discovering the same problem only after all twelve pages are assembled. The evidence is the inspected sample, not the template’s promise that it is suitable.
She then assembles the booklet in forty minutes. The final thirty-minute check confirms image count, order, captions, readability and absence of unwanted names. She finds one duplicated page, replaces it with the intended image and reopens the corrected export. The final count is twelve images. Planned work consumed 160 minutes plus the twenty-minute caption repair, for 180 minutes total. Twenty minutes of the original 200-minute capacity remain unused.
The work fits the five actual slots because these tasks can pause and their estimates include opening files, saving progress and leaving a restart note. Session one holds inspection for twenty minutes and selection for twenty. Session two holds caption work for thirty minutes and ten minutes of the repair. Session three completes the remaining ten repair minutes, tests the sample for twenty and starts assembly for ten. Session four finishes assembly for thirty and begins inspection for ten. Session five finishes inspection for twenty, leaving twenty unused. This sequence preserves the dependency checks rather than merely making the totals match.
Her completion record names the saved final file and the checks performed. It does not claim that the booklet is backed up unless she has verified the backup, or that someone else has received it when she has not shared it. Her final lesson is specific: check caption evidence before decorative layout and test the export with one page. The goal closes as a finite project. She can decide later whether another outing deserves its own booklet without automatically creating a new obligation.
If the caption repair had required sixty extra minutes instead of twenty, the plan would no longer fit the original capacity. Mira would need a conscious choice: simplify uncertain captions, reduce scope while preserving the purpose, add an available session, or move the finish. The assistant could compare those routes. It could not honestly mark the original schedule achievable by changing the estimates after the fact.
10. Worked plan: one small reading conversation
This second case is fictional and concerns adults. Arun wants to have one enjoyable conversation about a short piece of writing with two friends. He has no agreement from either person yet. He can spend up to two hours preparing over the next fortnight, wants to use a text everyone can lawfully access, and does not want a subscription, venue booking or standing commitment. His initial statement, “start a book club”, is larger than the immediate purpose requires.
Clarification produces a smaller output: one optional forty-five-minute conversation about an agreed short text, with a simple set of discussion questions. The desired outcome is a worthwhile exchange, but attendance and enjoyment remain partly outside Arun’s control. His controllable commitment is to check interest respectfully, offer a manageable choice, prepare if there is agreement and cancel or postpone clearly if the conditions are not met.
The action plan begins with permission-sensitive distinctions. SI can help draft a neutral interest-check message. Arun reviews it and decides whether to send it to the two intended friends. Preparing the draft does not send it. If a tool is used to send on his behalf, he must authorise the recipients and content within the tool’s rules. The assistant should not add other people from his contacts, infer that a friend has agreed from an old conversation, or create recurring invitations.
The first message does not announce a fixed event. It asks whether either friend would like a short discussion in the next fortnight and offers two broad windows. It also makes declining easy. Arun’s first milestone is not “two attendees confirmed”; it is “interest checked and actual responses recorded”. He decides to review replies after three days. That is his review point, not a deadline imposed on the friends unless they have agreed to one.
His original active-work budget is fifteen minutes for the interest check and response record, twenty for choosing a text after interest is established, forty for reading, twenty for preparing questions, ten for confirming logistics, and fifteen for a brief post-session note. These preparation and review tasks total 120 minutes. The forty-five-minute conversation is a separate activity that must also fit everyone’s agreed availability; it is not hidden inside the two-hour preparation budget.
The dependency chain is clear. Interest comes before extensive preparation. Text choice comes before reading and questions. An agreed time and access method come before treating the event as confirmed. Some work can happen in parallel: Arun can check whether a candidate text is accessible while waiting for replies. Any early access check counts inside the existing twenty-minute text-selection allocation, leaving correspondingly less of that allocation after replies arrive. He limits the check so that an unconfirmed idea does not consume the whole preparation budget.
At the first checkpoint, one friend is interested but unavailable in either window; the other is interested in one window. Arun does not interpret the unavailable friend’s response as rejection or ask SI to devise pressure tactics. He considers three options: a two-person conversation, another mutually available time, or postponement. His purpose allows a two-person conversation. He confirms that choice with the available friend and lets the other know that joining is optional if circumstances change.
They choose a short text after both confirm they can access it. The plan records the specific edition or link so they are not reading different versions unknowingly. SI may suggest discussion categories, but any factual claim about the text must be checked against the actual text. A generated summary should not replace Arun’s reading if the point of the activity is to engage with the writing himself. The assistant’s role is preparation support, not pretending to have supplied his experience.
Arun prepares three open questions: which passage stood out, where did the argument or story change direction, and what would each reader ask the author? He also prepares one concrete passage to discuss if conversation slows. He avoids a long quiz or a demanding reading schedule because neither was part of the agreement. The quality test is that the questions invite discussion and refer accurately to the selected text, rather than that there are many of them.
A day before the proposed conversation, the friend’s availability changes. Arun uses his fallback: ask whether a shorter twenty-five-minute conversation would still be welcome, or postpone without further preparation. The friend chooses the shorter session. Arun keeps two questions and drops the optional third. He records the change rather than claiming the original forty-five-minute event occurred. Because there is no paid booking or external commitment, the adjustment is straightforward, but consent still matters.
The fictional conversation happens and Arun writes a short note afterwards: they both wanted more time on one passage, the second question was too broad, and neither has committed to another session. His improvement is to use one precise passage question next time. His outcome evidence is limited to his own experience and what his friend actually said. SI should not label the friend’s feelings, satisfaction or future enthusiasm from a sparse recap.
The project closes with a choice, not an automatic series. Arun can ask later whether another conversation would be welcome. If nobody had wanted to take part, a good result would have been an honest decision to stop, not a manufactured event or a more persuasive invitation. This case shows why action planning must preserve other people’s agency: a personal goal can guide your effort without turning other people into resources the plan is entitled to consume.
11. Worked plan: finish a short story without outsourcing the voice
This fictional case concerns Lena, an adult writing a short story for her own enjoyment. She has an opening of about 400 words, three possible endings and a habit of repeatedly editing the opening. She wants a complete private draft within four weeks. She can offer two forty-five-minute sessions each week, for eight sessions and 360 minutes total. She does not intend to publish during this cycle and wants the story’s decisions and final wording to remain her own.
Her purpose is to experience completing and revising a story. Her outcome is a draft she can read as a whole and decide whether to develop. Her output target is 1,200–1,600 words with a complete beginning, middle and ending, one structural revision and one clarity pass. Word count provides a scope boundary; it does not certify artistic quality. The acceptance test also asks whether the main character makes a consequential choice and whether the ending follows from events in the draft.
Lena asks SI to help examine the three endings without writing the story for her. She supplies brief invented summaries rather than the entire private draft: one ending resolves the conflict through a choice, one introduces a coincidence, and one leaves the conflict unresolved. The assistant compares what each ending would require earlier in the story. Lena chooses the first because it matches the kind of story she wants to write, not because the assistant presents a ranking as objective literary truth.
Her first week’s two sessions have distinct jobs. In session one, she writes a rough version of the ending and a five-line connection from the existing opening. In session two, she drafts the missing middle without line-editing the opening. The done test for the week is a beginning-to-ending sequence that can be read continuously, even if it is rough. The plan explicitly permits awkward wording because the checkpoint concerns structure, not polished sentences.
The second week is a structural review and revision. Lena reads the whole draft, marks where she cannot explain the character’s choice, and chooses the most important gap. SI can ask questions such as “what does the character know at this point?” or “which earlier event supports this choice?” It should not invent a hidden theme Lena supposedly intended. She revises the causal connection in the first session and rereads the new version in the second.
The third week contains a clarity pass and a scope check. Lena looks for unclear references, unnecessary repetition and paragraphs that do not contribute to the intended effect. If the draft exceeds 1,600 words, she decides what to cut while preserving the story. If it is shorter than 1,200, she asks whether something essential is missing; she does not add filler merely to reach the lower number. A conscious scope revision can be legitimate if the original range no longer fits the completed story.
The fourth week’s first session is a final read-through and completion decision. The last session is a reserve for a specific repair, or remains unused if the draft passes the chosen tests. Each session is forty-five minutes, so the first seven sessions account for 315 minutes and the reserve is forty-five minutes. The plan does not require filling all 360 minutes. Finishing means meeting a reasonable acceptance test, not consuming the entire time allocation.
In the fictional execution, the first ending takes longer than expected. Session one produces only the ending and no connection notes. Lena records that the missing bridge still needs work. In session two, she spends fifteen minutes on the bridge and thirty on the middle. The draft is incomplete at the first checkpoint. Instead of rewriting the entire four-week calendar, she uses the first session of week two to finish the middle and moves structural review to the second session.
This consumes flexibility. A structural repair now takes the first session of week three, shifting the clarity pass to its second session. The scope check is combined with the first part of week four’s first session, followed by the final read-through. If that combination proves too tight, the reserved eighth session remains available. The change log shows which checks moved and why. No acceptance test is silently deleted to preserve a decorative appearance of being on schedule.
During review, SI suggests a dramatic new subplot. Lena saves the idea outside the active draft and declines to add it during this cycle. The suggestion may be interesting, but it expands the goal and adds dependencies. A planning assistant should help protect the chosen scope as well as generate possibilities. Otherwise every review can reopen the entire creative project and prevent completion.
At the final checkpoint, the story has 1,430 words, a complete sequence and an ending supported by the character’s earlier choice. Lena notices one confusing transition, repairs it in the reserved session and saves the final private version. These numbers are fictional checking examples, not a claim that this plan produces a publishable story. Her own reading supplies the completion decision. The assistant cannot certify that all future readers will understand or enjoy it.
The reusable lesson is that testing an ending early helped expose the missing middle, while the reserve protected the review steps when drafting took longer. Lena also learns that a forty-five-minute session can support one substantial writing job, but not always two. Her next plan can use that observation. She finishes with a manuscript and a better estimate of her process, rather than merely a larger collection of prompts about writing.
12. Repair common planning failures precisely
When an action plan stalls, resist the urge to add a more elaborate system immediately. First name the failure at the smallest useful level. A missing input, an unsuitable action, a weak estimate and a goal you no longer want are different problems. The following repairs are practical diagnostics. They are not diagnoses of personality, attention, motivation or mental health, and an assistant should not use them to label the person behind the plan.
If the plan has many verbs but no outputs, add done tests. “Research, organise, improve and review” may all be reasonable activities, yet they do not tell you when to stop. Repair one item at a time: “read two relevant instructions and record the export settings that affect this file”. The test should name the result you need, not demand exhaustive knowledge. Research becomes useful when it answers the question blocking the next decision.
If the plan is too ambitious, identify the purpose-preserving cut. Removing quality checks is often the wrong cut because it makes the result less dependable. Instead, reduce the number of items, the presentation complexity, the audience or the optional extensions. A private six-image collection may serve the purpose better than a twelve-image collection with invented captions. Make the scope change explicit so that the final result is evaluated against the revised agreement rather than a quietly abandoned original.
If you keep starting but not finishing, inspect work in progress. You may have four active drafts and no protected finishing step. Choose one candidate to carry through an acceptance test while the others wait. Ask SI which unfinished item is closest to a useful complete state and what prevents completion. Do not equate “closest” with “most important”; the choice still depends on the purpose and obligations attached to each project.
If another person is the bottleneck, separate your preparation from their decision. You can send an agreed request, make your own deadline visible and choose a respectful fallback. You cannot make a response happen by checking the same inbox every few minutes. A useful status is “waiting for confirmation; if none by the review point, postpone the optional activity”. It should not imply that silence means consent or that the other person owes participation in your personal goal.
If the model keeps expanding the plan, constrain its role. Say: “Do not add new deliverables. Preserve the chosen output and quality checks. Offer only changes needed to address the observed blocker.” This does not mean the model must ignore a serious issue; it means it should distinguish necessary repairs from attractive extras. An assistant can be creatively helpful at the Expand stage and deliberately conservative during Execute.
If tracking takes longer than the work, reduce the record. A small personal project may need only the current milestone, next action, blocker and review point. A dashboard with dozens of fields can become a second project. Keep a field only if it supports a decision you actually make. It is useful to know that caption verification is blocked; it may be unnecessary to assign each caption a colour, score, priority rank and recurring reminder.
If the same interruption keeps occurring, revise the assumed context. An evening slot that is consistently unavailable is no longer a realistic default. Move the action, shorten it, make preparation portable or reduce the active goals. A fallback used once is recovery. A fallback used every week is evidence that the original plan describes a life you are not currently living. Planning improves when that evidence changes the design.
13. Use SI at the right level and keep a portable plan
The Life series uses five levels: Ask, Think, Build, Integrate and Operate. They describe increasing structure in how you use assistance, not a competition to automate more of your life. For a small goal, Ask may be sufficient: get help turning a vague intention into one clear next action. Move to another level only when it solves a real problem and the additional maintenance remains worthwhile.
At Think, use SI to challenge the route. Ask which assumptions matter, what depends on other people and which alternative preserves the purpose with less effort. At Build, create a reusable action card or checklist. At Integrate, place the plan in the calendar or task system you already use, within the permissions you choose. At Operate, use a repeatable review process that updates the plan from evidence. More connected tools do not automatically make the plan wiser or more reliable.
A portable action card should survive outside the conversation. Include the goal, purpose, current state, chosen output, acceptance tests, constraints, next milestone, next ready action, dependencies, assumptions, review point and stop conditions. You can keep these in a simple note. The point is to avoid having the only usable version buried among brainstorming messages, superseded estimates and abandoned alternatives. A new assistant session should be able to read the card and understand what is current.
Here is a reusable planning request: “Help me turn this low-stakes personal goal into an action plan. First distinguish what I want to change from what I can directly do. Use only the facts in my current-state packet. Mark estimates and unknowns. Offer a small number of routes, then help me choose a minimum useful output. Build milestones with done tests, show dependencies, add the active effort, and finish with one ready action. Ask before assuming permission for external actions.”
For an overloaded draft, use a repair request: “This plan has more work than my available capacity. Preserve the purpose and the quality checks. Show the arithmetic, then propose a smaller scope, a later review point and an alternative method. State what each option gives up. Do not assume I can cancel existing commitments or buy extra tools.” The useful output is a set of honest choices, not a cosmetically compressed timetable.
For a checkpoint, provide actual evidence: “Planned: twelve captions in thirty minutes. Actual: eight checked captions in thirty minutes, with four locations uncertain. Remaining capacity: 130 minutes including a forty-minute reserve. Help me identify the next safe action. Do not invent missing locations or mark the booklet complete.” Specific observations make the review more useful than asking, “Why am I not making progress?”
For a restart after a break, use a continuity request: “Read this current action card and latest evidence note. Identify what is complete, what is waiting, what assumptions may have expired and the smallest ready next step. Do not restart completed work or reactivate cancelled extras without a reason.” This protects the project from being re-planned from scratch whenever the conversation changes. You should still verify the summary against the actual files and commitments.
The portable card also protects your independence. Once the next step is clear, you should be able to take it without keeping a model open. If every small move requires a new conversation, the system may be adding friction. Use the assistant where comparison, diagnosis or structure helps, then let the plan support ordinary action. A successful planning workflow can become quieter as you learn the project and its constraints.
14. Practise independently, then check your answers
Try these exercises before reading the suggested answers. They use fictional, low-stakes situations and are designed to test whether you can distinguish a useful plan from an impressive-looking list. There may be several good answers. Judge yours by whether it preserves the purpose, respects control and consent, fits the stated capacity and gives the next action a visible completion test.
Exercise one: Jo says, “I want everyone in my hobby group to love my photo presentation.” Jo has ten pictures, twenty minutes to prepare and no feedback yet. Identify a controllable output, one outcome that remains outside Jo’s control, and a useful first action. Avoid changing the goal into an elaborate public campaign or asking the assistant to predict how each person will react.
Suggested answer: a controllable output is a short, checked sequence of ten pictures with a clear opening and readable captions. Other people’s enjoyment remains outside Jo’s control. A useful first action is to choose an order for the ten images and explain in one sentence what the sequence is meant to show. The done test is a saved order and purpose sentence. With only twenty minutes, Jo should avoid adding unnecessary design features before checking the basic presentation.
Exercise two: A plan has forty minutes of selection, fifty minutes of captioning, thirty minutes of layout testing and sixty minutes of assembly. The person has three one-hour sessions and wants a thirty-minute reserve. Does the plan fit? Name a repair that does not depend on imaginary efficiency improvements. Keep the reserve separate from required work when doing the calculation.
Suggested answer: the required work totals 180 minutes. Gross capacity is also 180 minutes, but reserving thirty leaves a planned-work budget of 150. The plan is thirty minutes over that budget. A valid repair is to reduce the number of images enough to remove at least thirty minutes of work, if the smaller output still serves the purpose; alternatively add a genuinely available session or move the finish. Simply renaming the reserve “spare” does not change the arithmetic.
Exercise three: An assistant proposes, “Create a reading group, pick a date, invite six friends, book a venue and send weekly reminders.” The user only asked for help considering one informal conversation and has not authorised messages or spending. Identify three problems and rewrite the next step. Do not assume the people named in a contact list are available or interested.
Suggested answer: the proposal expands a one-off idea into a recurring group, assumes permission to contact people, and introduces a paid or binding venue action without approval. It also chooses a date before learning availability. A better next step is to draft a brief interest-check message for the user’s review, or first help the user choose which friends they intend to ask. The done test is an approved draft or a clarified recipient choice, not a claim that an event has been arranged.
Exercise four: A writer completes four scheduled sessions but still has no ending. The assistant reports “100% success because all sessions were completed.” What is wrong with the review, and what evidence should be checked next? Distinguish process completion from the output without dismissing the effort that actually happened.
Suggested answer: the review measures adherence to the sessions but not the goal of a complete story. The four sessions are real process evidence. The missing ending means the output acceptance test has not passed. The next review should inspect what the sessions produced, what question prevents the ending and whether the next action should test one possible resolution. More sessions may be needed, but simply repeating the same opening edits would not address the identified gap.
Exercise five: Sam’s collection has twelve captions. Nine are verified, two are uncertain and one is missing. The deadline is tomorrow and there are twenty minutes available. The assistant offers to invent plausible captions so the file looks finished. Write a safer plan and a completion rule. Assume the project is private and flexible, with no external obligation to finish tomorrow.
Suggested answer: keep the nine verified captions, use the twenty minutes to check or simplify the remaining three, and use accurate visible descriptions where specific facts cannot be established. If an essential caption remains unresolved, postpone or explicitly reduce the scope. Completion requires that the chosen captions be supportable; it does not require pretending every location is known. The flexible deadline makes honest revision preferable to manufacturing facts for appearance.
Exercise six: A task is marked “waiting” because a friend has not replied. The user has already sent one respectful invitation and set a private review point for Friday. On Friday there is still no reply. What can the plan conclude, and what can it not conclude? Give a fallback that preserves the other person’s choice.
Suggested answer: the plan can conclude that participation is unconfirmed at the review point. It cannot conclude that the friend is unwilling, rude, available or consenting. For an optional activity, the user can postpone, proceed with people who explicitly agreed, or choose a solo alternative. Any additional message should be a deliberate, proportionate choice, not an automatic escalation designed to force a response.
To transfer the method to your own goal, write a first action card without SI. Then ask the assistant to audit it for missing evidence, hidden dependencies, capacity mismatches and assumed permissions. Compare its suggestions with your original purpose before accepting them. You have understood the method when you can explain why each required action belongs, how you will know it is done and what would make you revise or stop.
15. Questions that matter when using the plan
How much detail is enough?
Enough detail lets you start the next ready action, inspect its result and identify the following decision. Distant steps can remain broad until their assumptions are checked. If you cannot begin, add the missing input, action or done test. If you can begin but keep expanding the document, stop planning and execute. A personal action plan is a tool for doing and learning, not an assignment to produce the most comprehensive schedule.
Should every goal have a deadline?
A fixed deadline is useful when a real event or agreement requires one. For a flexible hobby, a review date can be more honest: inspect progress on Sunday and decide the next scope. Distinguish an external commitment from a self-chosen target. If you miss a private target, revise it consciously. If others depend on a commitment, consider the communication and consequences rather than silently moving the date in your own notes.
What if I do not know the right outcome yet?
Use an exploratory goal. For example, produce two small sample layouts and choose whether either is worth developing. The output is the comparison and decision, not a finished booklet. Keep the exploration bounded so it does not become endless research. Name what you expect to learn, how you will inspect it and which decision follows. Discovery can be legitimate progress when the plan defines the uncertainty it is trying to reduce.
Can SI estimate every task for me?
It can suggest initial estimates and expose missing steps, but your own observations provide better context for your pace, tools and constraints. Label unfamiliar estimates as provisional. Test a small unit and update the remainder. Avoid interpreting a precise number as an assurance. An assistant that has not watched the work or inspected the inputs may be describing a typical-looking task rather than your actual one.
What if I want to change the goal halfway through?
Check whether the change reflects new evidence, a real change in purpose or an attractive distraction. Any of those can be discussed honestly, but they have different implications. Save what remains useful, identify commitments that need updating and decide whether to close the original goal or revise it. Do not pretend the new goal was always the acceptance test. A clear change record protects learning and makes the next action easier to understand.
How do I know SI is helping rather than adding work?
Compare the total workflow with a simpler alternative. Include time spent explaining, correcting, checking and maintaining the plan. Look for a clearer next action, an important assumption discovered, a capacity mismatch repaired or a completed output that met its tests. A longer plan and more chat activity are not sufficient evidence. If a short handwritten list works better for this goal, use it and return to SI only when a specific planning difficulty arises.
What should I do immediately after reading this guide?
Choose one personal goal that is small enough to test safely. State the purpose and the minimum useful output. Record your starting point, available capacity and one important unknown. Define a ready action with a done test and a review point. Carry out that action before building a larger system. The next version of your plan should be informed by something that actually happened.
Sources and further routes
The workflow, distinctions, examples, fictional cases and exercises in this article are original practical guidance. They are not presented as a validated intervention or a guarantee of goal attainment. The two external sources support the specific nearby discussions of if–then planning and generative-AI risk; they do not certify the complete method or any product’s performance.
- Gollwitzer and Sheeran: Implementation Intentions, hosted by the US National Cancer Institute
- NIST: Artificial Intelligence Risk Management Framework, Generative Artificial Intelligence Profile
- Return to How to Leverage Your Life with Super Intelligence
- Build Your Personal Super Intelligence Operating System
- Review Whether Super Intelligence Is Actually Improving Your Life
