How to Prioritise Tasks with Super Intelligence

1. Choose what deserves your next unit of attention

How to prioritise tasks with Super Intelligence starts with a choice that a long to-do list often conceals: what deserves to be done next, what can wait, and what should leave the list altogether? A useful answer is more than a numbered collection of tasks. It explains the consequence of each choice, respects promises already made, identifies the information that could change the order, and leaves you with something you can actually start. The goal is a defensible selection, rather than the satisfaction of watching an assistant produce a tidy ranking.

In the eduKateSG Life series, Super Intelligence, or SI, is an editorial name for practical AI assistance used with human judgment. It does not mean that current products have been established as artificial superintelligence. In this guide, an assistant helps organise supplied facts, compare alternatives, expose conflicts and test the logic of a proposed priority list. You remain responsible for what matters, which facts are correct and which real-world actions are authorised. A recommendation to cancel, send, purchase or disclose something is not permission to do it.

This article has a specific job. Goal planning turns an ambition into actions. Day planning locates activities within a particular day. Week planning allocates capacity across several days. Task prioritisation decides which competing actions deserve that capacity in the first place. These jobs meet at their edges, but they are not interchangeable. A beautifully timed schedule can still contain the wrong work. A sensible priority can also be impossible to execute until someone provides a missing input. We will separate selection, readiness and scheduling so that each receives the right decision.

Consider two tasks: polishing a personal slide deck for forty minutes and answering a ten-minute question that lets someone else prepare a venue. The deck may be more exciting and more visible. The question may nevertheless deserve earlier attention because a specific person is blocked. Now change the facts: the question is already answered in a shared document and the deck contains an error that will mislead its audience. The order may reverse. Priority belongs to a task in a situation, not to a task title permanently stamped with a colour.

The method used here is an editorial working method, not a validated psychological assessment or a universal productivity formula. Its sequence is simple: protect non-negotiable boundaries, verify commitments, expose dependencies, compare the remaining choices, select work that fits, and record why other work waits. The broader Life loop is Capture → Clarify → Expand → Structure → Execute → Review → Improve. Used lightly, it can fit on a single page. Used carefully, it can reveal that no honest ranking will solve an overloaded set of promises.

Begin with ordinary, reversible activities such as personal learning, hobby projects, routine administration or a volunteer presentation. Do not use this guide to rank urgent symptoms, alter treatment, make investment decisions, interpret legal deadlines or choose consequential outcomes for other people. Those tasks require appropriate authoritative sources and qualified judgment. A personal task board can record the need to contact a qualified person; it cannot replace that person’s assessment. Keeping that boundary clear makes everyday prioritisation both more useful and more trustworthy.

Back to contents · Next: 2. Capture a decision packet, not your entire life

2. Capture a decision packet, not your entire life

A model cannot prioritise a fact you have not supplied, yet supplying everything creates its own problems. Start with the particular decision window: the next ninety minutes, your available project capacity today, or the next three sessions of a hobby. State whether the window includes planning and checking. State what time is already unavailable. You do not need to upload a full calendar, an inbox export or a personal biography to choose between six ordinary tasks. A small, accurate packet is usually easier to inspect than a huge, partly stale one.

For each task, record an identifier, a concrete next action, a completion condition, an effort estimate, any deadline and its source, the consequence of waiting, a dependency, and the person who can decide. The identifier stays stable when wording changes. The completion condition prevents “work on presentation” from silently becoming a three-hour redesign. The deadline source might be your own preference, a message you have checked, a booking confirmation or an agreed handover. These are different kinds of evidence and should not all be labelled “urgent”.

Write observations in plain language. “A neighbour needs the borrowed projector back before 18:00; the return takes twenty minutes including travel” is useful. “Projector: high priority, easy, important” hides the reason for the label. “Complete a twenty-minute drawing exercise; no external deadline; chosen personal learning goal” is useful too. You do not need an external authority to justify worthwhile personal work. You do need to distinguish your chosen goal from an obligation someone else is relying on.

Keep private details outside the packet unless they are genuinely needed and appropriate to share with the service you are using. Replace names with roles when that preserves meaning. A task can say “reply to the organiser” without including the organiser’s address. A planning exercise can say “fixed private appointment” without describing its purpose. Do not paste passwords, payment details or confidential documents into a prioritisation prompt. Where context is sensitive, make the selection offline or use a short, non-sensitive summary that you have checked yourself.

Unknown is a valid field value. If a task has no verified duration, write a range or state that the estimate is untested. If you remember a deadline vaguely, mark it for verification. If another person might do the work, record “delegation not agreed”. An assistant should not fill these gaps with plausible-looking answers. The first useful result may be a short question list, particularly when one missing fact would change the top choice. It is better to spend two minutes checking an actual deadline than twenty minutes optimising a mistaken one.

Capture unfinished work without automatically promoting it. Yesterday’s task has an age, but age is not a sufficient reason to outrank today’s genuine commitment. Record whether the task is still wanted, whether its purpose has changed and whether anyone is waiting. A stale “research new notebook app” item can be removed if your existing system works. A delayed promise to return borrowed equipment deserves a different treatment. A clean packet preserves that distinction instead of presenting every open loop as equally deserving of guilt or attention.

Back to contents · Next: 3. Clarify importance, urgency, readiness and ownership

3. Clarify importance, urgency, readiness and ownership

Importance asks what useful result the task contributes to. Urgency asks what changes if it waits beyond a particular point. Readiness asks whether you can do it now with available inputs and permissions. Ownership asks who is responsible for deciding or completing it. These four questions should be answered separately before you combine them into a choice. A task may matter greatly and be unready. Another may be ready and quick but contribute almost nothing. A third may be urgent for someone else without being your responsibility to perform.

Replace an unsupported urgency claim with a consequence statement. “Urgent: choose a font” becomes “The event brief can be sent in its existing readable font; changing it today does not alter the organiser’s ability to use it.” That statement lowers its position for a reason. “Routine: confirm attendance” becomes “The organiser cannot finalise the room arrangement until I answer before noon.” That statement raises its position without needing a dramatic label. This is how clarification corrects a list distorted by visual appeal, habit or the loudness of a notification.

Deadlines require at least three distinctions. A fixed external deadline comes from a real constraint or agreement. A negotiated deadline can change if the relevant person agrees. A self-chosen target expresses your intention and can be revised by you. None is meaningless. A self-chosen practice session may be central to your long-term goal. But changing it does not carry the same relationship consequence as silently missing an agreed handover. Ask SI to preserve deadline type and source in its answer instead of collapsing everything into one due-date column.

Readiness also has several forms. An informational blocker means you lack a fact. A material blocker means the equipment or file is unavailable. A permission blocker means you do not have authority to use, share or change something. A coordination blocker means another person has not agreed to a role or time. The repair differs in each case. Checking a file location will not repair a lack of consent. A confident assistant cannot remove a permission boundary by estimating that someone would probably say yes.

Ownership prevents an innocent-looking ranking from reallocating other people’s labour. “Ask Arun whether he can collect the projector” is your action. “Arun collects the projector at 17:00” is not a confirmed task until Arun agrees. Similarly, “prepare a reply for review” does not mean “send the reply”. Ask for a plan that identifies proposed handoffs separately from agreed ones. When someone else’s effort is uncertain, retain a feasible fallback rather than treating the proposed delegation as free capacity you already possess.

Finally, clarify whether an item is a task or an outcome. “Become a better photographer” cannot compete fairly with “return the projector”. Translate the goal into a concrete unit such as “complete one twenty-minute composition exercise and save the result”. That gives it a cost and an observable finish. If you cannot define a next action without deciding the goal itself, use the goal-planning guide first. Prioritisation works best when the candidate actions are clear enough to compare, while their underlying purposes remain visible.

Back to contents · Next: 4. Protect boundaries before comparing benefits

4. Protect boundaries before comparing benefits

Some choices should be screened before you rank benefits. If an action requires permission you do not have, it is not eligible merely because it is quick. If it would expose private information unnecessarily, look for a safer version. If a plan assumes you will ignore a fixed responsibility, the conflict must be resolved rather than hidden in a score. These are eligibility checks. They prevent an attractive benefit from compensating for a condition that should not be traded away in the first place.

In everyday planning, start by identifying already-confirmed commitments and genuine constraints. A borrowed item may need to be returned. An agreed short response may let another person proceed. A file may need checking before it can be shared. The correct question is not always “Which activity produces the largest personal benefit?” Sometimes it is “Which feasible action protects a promise I have already made?” That does not mean every request must be accepted. It means existing commitments deserve deliberate treatment before new optional work consumes their capacity.

Keep recovery and ordinary life visible when defining available capacity. If you have ninety usable minutes after meals, travel and fixed commitments, give the assistant ninety minutes. Do not give it the whole evening and expect it to infer what must remain protected. The article does not prescribe one universal rest allowance. It asks you to represent the capacity you can realistically use. If a task requires sustained concentration and you cannot presently provide it, selecting a smaller safe task or stopping may be the more accurate choice.

When two genuine obligations do not fit, escalation is part of prioritisation. You may need to ask for a revised handover, reduce scope with agreement or arrange an accepted alternative. SI can help you describe the conflict clearly: what is due, what is possible, what options exist and which decision is needed. It should not manufacture agreement, conceal the collision or imply that effort alone can satisfy incompatible constraints. A plan that names one unavoidable trade-off is more reliable than a plan that quietly spends the same thirty minutes twice.

Use the same discipline for safety-sensitive or high-impact items. A reminder to seek professional advice may be eligible; an AI-generated instruction to alter a medical, legal or financial arrangement needs a different process. If an urgent safety issue arises, ordinary productivity ranking is the wrong framework. Follow appropriate emergency or professional guidance for the situation. An assistant’s fluency does not make its ordering authoritative, and a personal scoring sheet should never decide whose wellbeing or rights matter more.

The practical output of this stage is small: a set of eligible tasks, a set of protected commitments, and a set of blocked or out-of-scope items with explicit next steps. You may discover that the highest-priority action is to verify a claim or seek a decision rather than perform the original task. That is not a failure to get things done. It is a way to avoid completing work that should never have been started, or starting work under assumptions that cannot support it.

Back to contents · Next: 5. Expand the options before ranking the list

5. Expand the options before ranking the list

A list often looks overloaded because each item is presented in its largest imagined form. Before asking which task wins, ask whether the options can change. Could a forty-five-minute rewrite become a fifteen-minute factual correction? Could a complete room reorganisation become clearing one surface needed for tomorrow? Could a long research session become one question to a reliable source? These are alternatives to compare, not automatic shortcuts. The smaller version must still satisfy the real purpose and any agreement attached to the task.

Distinguish reducing scope from reducing quality. Removing decorative slides may preserve a useful presentation. Skipping verification of the date or venue may make the same presentation unsafe to rely on. A good reduction preserves the acceptance condition and removes optional enhancement. Ask SI to identify the minimum useful output, the checks that must remain, and the features that can wait. Then verify that the receiver can actually use the smaller output. Only the relevant person can approve a change to a promise that affects them.

Consider not doing the task at all. A subscription catalogue may be interesting but unrelated to your current learning goal. A duplicate folder cleanup may be unnecessary because the original organisation is already clear. A task created by an old plan may no longer serve its purpose. Deleting it from your personal list is different from deleting files or withdrawing a promise; those actions have their own permissions and consequences. Sometimes the right prioritisation outcome is to retire an intention while leaving records and other people’s expectations untouched.

Consider sequencing alternatives as well as task alternatives. A small permission request may unlock a later creative task, so doing it early creates a chance of receiving an answer. But sending the request does not mean the answer will arrive, and the waiting period is not active labour. Build a usable fallback that does not depend on a fast response. This separates the value of opening an option from the certainty of completing the dependent task. Both matter, but they should not be counted as if they were the same result.

Ask for two genuinely different selections rather than ten cosmetically different rankings. One might protect existing promises and include a small learning activity. Another might protect the same promises and spend the remaining capacity reducing a known technical risk. Compare what each selection achieves and what it leaves exposed. Do not require the assistant to invent a dramatic third path if the facts support only two sensible choices. A short set of meaningfully different options is easier to evaluate than an artificial catalogue of “balanced”, “ambitious” and “optimal” variants.

Expansion has a stopping point. Once you have a feasible option and understand its principal trade-off, more alternative generation may add little. Give the planning step a proportionate budget. For a low-stakes choice between two similar household tasks, a brief human decision may be enough. Save deeper analysis for a conflict involving a real promise, a scarce opportunity or an uncertain dependency. SI should make choices clearer at an appropriate cost, rather than turn every small decision into a research project of its own.

Back to contents · Next: 6. Structure a defensible selection without false precision

6. Structure a defensible selection without false precision

After the eligibility checks, compare tasks using reasons that belong to the actual situation. Useful questions include: what consequence will occur if this waits, who is blocked, how strongly does it serve a chosen goal, what uncertainty does it remove, and what effort will it consume? The answer need not become a universal numerical score. A task that prevents a missed commitment may simply come before a cosmetic improvement. The explanation is more important than inventing a decimal that makes the comparison look scientific.

A practical selection can use four states: do next, do after a named condition, defer until a named review, and remove from the current list. These are states of work, not labels of personal worth. Add an owner and a reason to each. “Defer portrait edit until Saturday’s creative session because no one is waiting and today’s agreed brief comes first” is a complete decision. “Low priority” is incomplete because it does not say whether the work is unwanted, blocked, forgotten or deliberately postponed.

When several eligible tasks compete, compare a pair at a time. Would completing A before B cause B to miss a verified threshold? Does A unblock B or someone else’s work? If both can wait safely, which serves the goal you actually chose? If they contribute similarly, which has a clearer completion condition or lower commitment cost? Stop comparing when the decision is clear enough. A partial order is often sufficient: first confirm the headcount, then complete either of two independent tasks before the brief is sent.

Use effort estimates as constraints and sensitivity tests, not promises. Suppose a task is estimated at twenty to forty minutes. Ask whether the selection still fits at forty. If it only fits at twenty, identify what will be removed if it runs longer. Where possible, define a limited first step that produces evidence before consuming the whole budget. “Spend ten minutes checking whether the source files open” can be more useful than an invented precise estimate for a repair nobody has inspected.

Scores can help when you genuinely need to make repeated comparable decisions, but they bring choices of scale and weight. If you use a score, define each criterion before applying it, avoid counting the same benefit twice and inspect whether a small change reverses the ranking. A score of 8.2 is not evidence that an action is objectively superior to one scoring 8.1. When reasonable assumptions produce different winners, report the tie or ask the human to decide. The separate decision-matrix guide explores that tool in greater depth.

Write the final reasoning in a sentence a future version of you can understand. “I selected the attendance reply because the organiser is blocked until noon; I deferred browsing because it has no verified deadline and does not support this week’s chosen practice goal.” This sentence is a compact audit trail. If the organiser later says the reply is no longer needed, the reason disappears and the priority can change. A transparent reason makes revision easier than a mysterious rank that must be trusted or discarded wholesale.

Back to contents · Next: 7. Worked packet: twelve tasks competing for three hours

7. Worked packet: twelve tasks competing for three hours

The following case is entirely fictional. Mei is an adult preparing a small volunteer photography talk while maintaining her own hobby practice and routine files. The numbers are teaching inputs, not measured claims about productivity or predictions about any reader. She has 180 usable minutes remaining across several free periods today, after her fixed commitments and ordinary needs. Her aim is to honour today’s confirmed promises, preserve useful work and keep a small amount of personal learning if capacity permits. She is choosing work, not building a minute-by-minute timetable.

Mei reserves fifteen of the 180 minutes for reading the task packet, making the decision and recording it. She reserves another fifteen minutes for unexpected overrun. This leaves 150 minutes for selected actions. Each action estimate below includes its ordinary setup, the stated completion check, saving and any travel explicitly mentioned. No action overlaps another, and no background computer process is counted as additional usable human time. The reserve is a local assumption for this example, not a recommended percentage for every reader or every day.

Task A is to confirm her own attendance and the agreed headcount with the adult volunteer organiser. It takes ten minutes, including checking the existing invitation and sending the reply herself. The organiser’s checked message asks for the reply before noon because room preparation depends on it. Task B is to return a borrowed projector to a neighbour by the agreed 18:00 deadline. It takes twenty minutes including packing, travel, handover and return. Mei is responsible for both tasks; no other person has agreed to take them over.

Task C is to ask a photographer whether Mei may use one image in the talk. It takes ten minutes to identify the exact image and send a clear request herself. Permission has not been granted. Task D is to finish and send the agreed one-page event brief to the organiser before 17:00. It takes forty-five minutes including checking the event facts, exporting, opening the exported file and sending it herself. The organiser has explicitly agreed that the brief may use a plain placeholder if image permission is unavailable. Therefore D depends on correct event facts, but not on a positive answer to C.

Task E is to copy the current talk files to an existing local backup drive and open the copied brief and slide file to verify that they are usable. It takes twenty minutes. The current working copy is the only checked copy, the drive is available, and no new account or cloud sharing is involved. Task F is one twenty-five-minute rehearsal of the opening and closing of the talk, including a short correction note. Task G is one twenty-minute composition exercise in Mei’s personal photography course, followed by saving the result. Neither F nor G has an external deadline today; Mei values both, with rehearsal more directly connected to the forthcoming talk.

The remaining items are optional. H is a fifteen-minute desk tidy with no blocked activity depending on it. I is twenty-five minutes browsing a newsletter archive with no specific question. J is fifty minutes refining a personal portrait photograph that nobody is waiting for. K is thirty minutes comparing new courses even though Mei has an unfinished course she still wants to use. L is forty minutes reorganising an old remote folder; no current file is missing, and changing the folder would require a separate check of its sharing arrangements. These tasks are possible interests, not hidden promises.

The full candidate effort is 310 minutes: A through G total 150, and H through L total 160. Planning and reserve add another thirty minutes if all candidates are mistakenly accepted, creating a 340-minute demand against 180 minutes available. The point is not to cram the excess into the evening. The point is to choose a justified 150-minute action set and label the other 160 minutes honestly. Readers can reproduce the arithmetic directly from the supplied estimates before considering any AI-generated recommendation.

Back to contents · Next: 8. Worked decision: select, defer and keep a fallback

8. Worked decision: select, defer and keep a fallback

Mei’s initial prompt can be brief but specific: “Use only this fictional task packet. Separate verified facts from assumptions. Protect the confirmed commitments and existing permission boundaries. Choose up to 150 minutes of actions, leaving my fifteen-minute planning allowance and fifteen-minute reserve untouched. Show selected and deferred tasks, dependencies, a feasible order and the fact that would most likely change your recommendation. Do not send messages, modify files or assume another person has agreed to help.” This asks for a reviewable decision rather than unrestricted execution.

The first selection contains A, B and D because they protect verified agreements. Their combined effort is seventy-five minutes: ten plus twenty plus forty-five. C is selected early for ten minutes because it opens the possibility of using the requested image while leaving the agreed placeholder available. E adds twenty minutes to protect and check the current files. F adds twenty-five minutes of rehearsal. G adds twenty minutes of personal learning. The complete selection is therefore 10 + 20 + 10 + 45 + 20 + 25 + 20 = 150 minutes.

The reasoning matters more than the alphabetical list. A needs attention before noon because someone is waiting. C is useful early because another person’s reply cannot be produced on demand. D can be completed with the placeholder if permission remains unknown. E belongs after the current brief and slide changes have been saved, so that the checked backup contains the intended version. B must fit before 18:00, but it does not require Mei to interrupt every other activity immediately. F and G are selected benefits after the confirmed obligations and file protection fit.

A feasible ordering is A, C, D, E, B, F and G, provided Mei’s actual free periods allow A before noon, D before 17:00 and B before 18:00. That proviso is important: a priority order does not prove a timetable. If the only travel opportunity is earlier, B can move forward without changing why it was selected. If the actual free periods cannot satisfy those three deadlines, Mei must use the day-planning process or renegotiate an affected commitment. The assistant should flag missing clock-time information instead of pretending aggregate minutes establish every deadline.

H is deferred because clearing the desk does not enable any selected task. I is removed from today’s list because browsing has no decision question. J is deferred to a future creative session. K is removed from the current selection because comparison shopping for another course is less aligned with Mei’s stated intention than using her existing course. L is deferred pending an actual need and a sharing check. None of these decisions says the activities are intrinsically bad. They simply do not justify displacing the chosen work under the supplied constraints.

Before using the result, Mei performs four checks. She verifies the arithmetic independently. She checks that all twelve identifiers appear exactly once in selected, deferred or removed states. She checks that no permission has been inferred from silence. She confirms that the plan uses her actual goals rather than a generic preference for finishing the maximum number of tasks. Completing seven tiny items would not necessarily be better than completing one substantial commitment. In this packet, seven actions fit because their combined purposes and costs fit, not because seven is a target.

The execution card can now be short: “Protect A, B and D. Ask C early, but use the approved placeholder if no permission arrives. Back up the saved current files with E. Use remaining chosen capacity for F and G. Keep H, J and L deferred; leave I and K off today’s list. Check the clock-time fit before acting.” That card is enough to resume without rereading a long chat. The detailed packet stays available to explain the decisions if circumstances change.

Back to contents · Next: 9. Worked repair: forty-five minutes disappear

9. Worked repair: forty-five minutes disappear

Suppose Mei completes A and C, using twenty action minutes, and then learns that an unrelated fixed commitment will consume forty-five of her remaining usable minutes. This is another fictional input, not a surprise that the assistant should have predicted. Her total usable capacity is now 135 minutes rather than 180. The fifteen-minute planning allowance has already been used and remains counted. She keeps the fifteen-minute reserve. Her revised action budget is therefore 105 minutes, of which twenty have already been spent, leaving eighty-five action minutes.

The repair starts from current state, not from a fresh copy of the original list. A is complete with a sent reply. C is complete as a request, while image permission is still unknown. B, D, E, F and G remain unstarted. H, J and L remain deferred. I and K remain outside today’s selection. That distinction prevents the assistant from counting A and C as both completed and still available to remove. Time already used cannot be reclaimed by changing the label of the task that used it.

B and D still protect verified commitments, so together they require sixty-five of the remaining eighty-five minutes. E takes the remaining twenty. F and G are deferred. The revised selected-action total is A ten, C ten, B twenty, D forty-five and E twenty: 105 minutes. Adding fifteen minutes of planning and fifteen reserve yields exactly 135. The forty-five-minute reduction is matched by deferring F and G, whose twenty-five and twenty minutes total forty-five. The calculation is visible enough to catch a hidden promise that capacity has somehow expanded.

There is still a decision within the repair. Mei could propose replacing E with a twenty-minute fragment of rehearsal, but that would leave the only checked copy of current work unprotected. Given the packet’s stated condition, she keeps E. This is a reasoned choice, not a general rule that file backup always outranks practice. If the packet instead said that a current, opened and verified backup already existed, E might no longer be necessary. Priority changes when relevant evidence changes. It should not change merely because a new chat response sounds more confident.

If permission arrives before Mei finishes D, she may include the image only if the remaining checks and insertion still fit D’s forty-five-minute allowance. Permission does not oblige her to expand the task. If inserting the image would require ten extra minutes, she can keep the agreed placeholder or deliberately use part of the reserve, recording the consequence. She should not hide the extra work inside the original estimate. If permission never arrives, the placeholder remains the authorised route; silence is not approval.

At closure, the fictional actual record says A took ten minutes, C ten, D forty-seven, E eighteen and B twenty. The actions still total 105 minutes because D’s two-minute overrun is offset by E’s two-minute underrun. The fifteen-minute reserve remains unused. Mei checks the sent brief opens, records that the projector was handed back and verifies the copied files. F and G remain explicitly deferred. The outcome is not “everything completed”. It is “the revised commitments were met within the revised budget, with a clear record of what did not happen”.

The improvement follows from this record. Mei does not conclude that every future brief will take exactly forty-seven minutes or every backup eighteen. One fictional observation cannot justify a universal estimate. Instead she notes which parts took longer, whether her next brief is similar and whether the same checking steps will be needed. The method succeeds by making the evidence usable for a later choice. It avoids both false certainty and the discouraging claim that a changed plan was necessarily a bad plan.

Back to contents · Next: 10. Choose information-gathering tasks when uncertainty matters

10. Choose information-gathering tasks when uncertainty matters

Sometimes the best next action is a small check that changes which larger task should be attempted. Imagine that Noor, an adult hobbyist, has sixty usable minutes to prepare files for a community display. A full export might take forty minutes if the files are compatible, but could fail if a required font is missing. A five-minute test export can establish whether that problem exists. This test has value because it changes the next choice. It is not valuable merely because checking feels careful or produces another item to tick.

Make the decision branches explicit before doing the check. If the test opens correctly, Noor can use forty minutes for the full export and fifteen minutes for review, totalling sixty minutes including the five-minute test. If the test fails, a fallback using the approved existing PDF takes thirty minutes plus ten minutes of review, leaving fifteen minutes unused. The unused time is not a defect. It means Noor did not spend forty minutes pursuing a path that the test showed was unsuitable. Both branches should preserve whatever permissions apply to sharing the display files.

Contrast that with a twenty-minute search for “the best export settings” without knowing the specific file requirements. It might produce useful information, but it might also generate a long comparison that changes no action. Ask what decision the research will affect, which result would alter the choice and how much time the uncertainty justifies. SI can help formulate the narrow question, but a model-generated answer about the actual file is not a substitute for opening the file or checking the destination’s real requirements.

Do not confuse uncertainty with low priority. An important task with an unknown prerequisite may need an early diagnostic step. Equally, do not let uncertainty make every speculative possibility urgent. If a problem has no evidence and the cost of checking it exceeds the plausible benefit for this ordinary decision, leave it as an assumption to revisit if a trigger occurs. The point is proportionality. A ten-minute decision does not always need a twenty-field risk register, while a consequential irreversible action needs more than a casual guess.

A useful prompt is: “For each uncertain task, identify the smallest check that could change the selected action. State the possible results and what I would do after each. Do not invent probabilities. If the check would not change my choice, say so.” The instruction against invented probabilities matters. An assistant may present “80 percent likely” because numerical language seems helpful, even when the packet contains no basis for that estimate. A qualitative uncertainty statement can be more honest and easier to use.

This chapter extends the Life loop without making it complicated. Capture the uncertain fact, clarify why it matters, expand the branches, structure a bounded test, execute the test, review the actual result and improve the task packet. The test is finished when it supports a decision, not when all possible uncertainty has disappeared. That stopping condition protects you from research that becomes a way to avoid choosing. Your next action should become clearer because the check occurred, otherwise the check needs a better purpose.

Back to contents · Next: 11. Keep important work from disappearing behind small requests

11. Keep important work from disappearing behind small requests

A task list can repeatedly select short, externally prompted items while leaving meaningful personal work untouched. The short tasks are not necessarily trivial. Replies and practical favours may matter. The problem arises when every available minute is allocated reactively and no explicit choice remains for a goal you continue to say matters. SI can make that pattern visible, but it should not decide that your relationships, learning or creative work have the same value for everyone. The corrective choice belongs to you.

Begin by naming one protected contribution at an appropriate scale. “Complete one twenty-minute exercise from the course I already chose” is clearer than “prioritise self-development”. It creates a task that can compete honestly with other work. You can still defer it when a real constraint changes, as Mei did, but repeated deferrals become evidence to inspect. Perhaps the task is too large, the goal is no longer wanted, or your commitments genuinely consume the available capacity. Each explanation leads to a different repair.

Avoid turning protected personal work into an inflexible command. A chosen creative session may yield to an unexpected responsibility. The important question is whether the change is deliberate and whether a replacement decision is made later. “Deferred because a verified handover needed the capacity” is different from “lost because I kept opening new messages”. A brief reason protects against both harsh self-judgment and easy rationalisation. It lets you see whether the priority system is serving the life you actually want rather than the life implied by your inbox.

Use a backlog review to retire intentions you no longer endorse. An old course comparison may keep reappearing because no one has said it is finished or unwanted. Ask yourself whether you would add it today if it were absent. If the answer is no and nobody is relying on it, removing it can be a good decision. Do not let the assistant automatically delete records, cancel services or withdraw commitments because a backlog item is removed. The task list is a representation of intention; external state requires its own appropriate action.

When the assistant recommends quick wins, ask what is being displaced. Five ten-minute tasks use fifty minutes, even if each feels harmless. If those tasks prevent you from completing a forty-minute agreed deliverable, the apparent efficiency is misleading. On the other hand, a short unblocking reply can be a genuine high-value action. The repair is not a universal ban on small tasks. It is a habit of comparing their combined cost and consequence with the larger action they compete against.

At review, look for a pattern across comparable decision windows. Did important self-chosen work receive any actual capacity? Were deferrals justified by verified changes? Did optional browsing repeatedly enter as if it were mandatory? Keep the record proportionate and private. You do not need a score for every hour of your life. A handful of short observations can reveal whether the selected priorities match your stated purpose. If the mismatch persists, change the task size, reduce commitments or reconsider the goal instead of demanding a more persuasive ranking.

Back to contents · Next: 12. Execute from a short card and verify real completion

12. Execute from a short card and verify real completion

Prioritisation creates an intention. Execution changes something in the world. Keep that distinction visible so that the assistant does not report a task complete because it has described how to do it. A useful execution card includes the next action, its completion condition, the needed materials, the stop condition and the next review trigger. For an ordinary file task, completion might mean “the exported file has been opened and the date checked”. For a message, it might mean “the human has reviewed and sent the reply to the verified recipient”.

Separate ready, active, waiting and complete states. A permission request can be complete as a request while the dependent use remains waiting. A backup operation can have finished copying while verification remains incomplete. A draft can be ready for review without being sent. These distinctions prevent a priority board from overstating progress. They also help the next decision: if the current task is waiting for someone else, you can choose another ready action without pretending the blocked result has already been achieved.

Limit active work in a way that fits the situation. For a personal decision window, one active attention-demanding task is often a simple starting policy, though it is not a universal scientific optimum. Finish a useful unit or leave a clear stopping note before switching. You can then resume from a known state rather than reconstruct the entire decision. Where a tool runs in the background, verify whether you truly have another independent activity available; do not assume all waiting time can be converted into productive capacity.

Tool access does not equal permission. If your assistant can send email or change a calendar, specify the authorised action and destination when you genuinely want it to act. For a prioritisation exercise, keep it in proposal mode. Do not ask it to “sort out everything” and then infer that every generated cancellation, delegation or purchase was approved. The safest useful workflow often has the assistant prepare the plan and the human perform ordinary actions, especially while the task rules and completion checks are still being learned.

Generative AI can produce confidently incorrect content. NIST’s Generative AI Profile identifies confabulation as a risk and also discusses data privacy and human–AI configuration. That is a reason to check source facts and actual results, not evidence that this particular prioritisation method has been scientifically validated. In this guide, verification is concrete: read the real deadline, open the actual file, confirm the arithmetic and check whether the required person agreed. A polished explanation from the same assistant is not an independent confirmation.

When work stops, write the smallest useful continuity note. “Brief exported and checked; placeholder used because permission has not arrived; backup still needed” is enough to support a later choice. Avoid copying sensitive correspondence into the note when a role and status will do. The next review should use this current record rather than the original untouched task list. A priority system becomes reliable when it can recognise what actually happened, including partial completion, waiting states and decisions that deliberately removed work.

Back to contents · Next: 13. Review decisions without measuring your worth

13. Review decisions without measuring your worth

The review asks whether the selection served its purpose under the information available at the time. It does not ask whether you were a sufficiently productive person. Start with the concrete record: selected actions, actual completions, time used, verified changes and deferred items. Compare like with like. A day disrupted by a new fixed responsibility is not directly comparable with an uninterrupted session. A plan that protected a promise and removed unnecessary work may be successful even if it produced fewer checkmarks than an earlier plan.

Separate a bad outcome from a bad decision. A sensible request for permission may receive no response. If the plan had a legitimate fallback, that outcome does not make the request irrational. Conversely, an unjustified assumption may happen to work once. If a person unexpectedly agrees after you already acted, the favourable ending does not make the earlier lack of permission acceptable. Review the evidence and boundaries available when the choice was made, as well as the eventual result. This prevents luck from becoming a misleading planning rule.

For ordinary work, a few useful observations are enough. Did any confirmed commitment fail because optional work displaced it? Did you spend significant time on something that no longer served a goal? Did an early check prevent avoidable rework? Did a task remain blocked because the owner or required decision was unclear? These are diagnostic questions. Each points to a repair. A single completion percentage compresses them into a number that may conceal more than it reveals, especially when tasks have very different sizes and consequences.

Record estimate errors with their cause when you can. A task may take longer because its scope expanded, because the original estimate omitted checking, because a file was missing or because the work was unfamiliar. Those causes should not all produce the same correction. If scope changed, clarify the completion condition next time. If checking was omitted, include it. If uncertainty was genuine, retain a range or use an early test. Simply multiplying every estimate by two may conceal useful differences among the tasks.

Review deferrals as decisions with expiry conditions. A deferred task can return when its dependency arrives, when its promised review date comes, or when a consequence changes. It should not automatically jump to the top because it has been carried forward. Ask whether its purpose remains valid. If it has become unnecessary, retire it. If someone is still waiting, communicate through an authorised route rather than relying on your private list to repair their expectations. A backlog cannot renegotiate a promise by itself.

Improve one rule at a time. You might decide to verify deadlines before ranking, include file-opening checks in estimates, or stop comparing new courses while an existing course remains your chosen goal. Test the rule over relevant ordinary decisions and see whether it helps. Avoid adopting a grand personal operating system on the evidence of one example. The practical benefit of SI is that it can help you notice a specific failure and prepare a specific correction. It should leave your decisions easier to explain and easier to revise.

Back to contents · Next: 14. Diagnose the common prioritisation failures

14. Diagnose the common prioritisation failures

When everything is labelled urgent, investigate the labels before creating finer shades of red. Ask which deadlines are verified, what happens if each item waits and who has authority to renegotiate. Some tasks will turn out to have preferences rather than fixed thresholds. Others may be genuine conflicts requiring a conversation. The repair is evidence and agreement, not a more forceful ranking. An assistant should be willing to say that the supplied commitments cannot all be met with the available capacity.

When the assistant always selects the shortest task, inspect the objective you gave it. “Help me clear my list” can encourage a focus on completion count. Restate the purpose in terms of outcomes and commitments, then ask what the short tasks displace. Give substantial work a bounded next action so that it is comparable. A forty-minute useful draft can compete with four ten-minute errands; “finish my whole project” cannot be assessed from the same information. The failure may be in task definition rather than in the ranking alone.

When scores seem convincing but the result feels wrong, look for missing gates, duplicated criteria or invented evidence. A task might receive separate points for “impact”, “value” and “importance” even though all three refer to the same benefit. A permission-dependent action may receive a high score despite being ineligible. A deadline may have been guessed from common practice. Repair the inputs and comparison rules before adjusting weights. Otherwise the scoring system becomes a sophisticated way of expressing the assumptions that caused the problem.

When the order changes every time you ask, check whether the facts or preferences changed. If neither changed, ask the assistant to explain the difference and identify which comparison is genuinely uncertain. You may have a tie rather than a clear winner. Choose an acceptable option and set a review trigger instead of repeatedly seeking certainty. Frequent reranking can itself consume the capacity intended for action. Stability should come from stable reasons, while flexibility should come from new evidence, not from the assistant’s variation in wording.

When tasks disappear between revisions, require a reconciliation by identifier. Every original item should be complete, active, waiting, deferred, removed or explicitly replaced. If a large task becomes three smaller ones, record the relationship and remove double counting. If a task is no longer needed, give the reason. This is especially important after an interruption or a compressed summary, when an assistant may preserve the top items and lose the rest. A shorter answer is useful only if it preserves the decisions that matter.

When planning feels like avoidance, look at its stopping condition. Can you name the fact still needed to choose? Would another comparison alter the action? If not, select a safe, reversible next step and do it. When action keeps stalling, inspect readiness rather than demanding motivation. Perhaps the file is missing, the next action is vague, the task exceeds the current capacity or the goal is no longer wanted. The right repair is the one that addresses the actual obstacle, not a generic instruction to become more disciplined.

Back to contents · Next: 15. Independent exercises with complete checking guidance

15. Independent exercises with complete checking guidance

Exercise one: Alex has ninety usable minutes. Ten are reserved for choosing and recording the plan and ten for overrun, leaving seventy action minutes. Task A is a fifteen-minute agreed reply due before noon that unblocks an organiser. Task B is a thirty-minute check and export of a promised handout due today. Task C is twenty-five minutes of personal practice with no external deadline. Task D is forty minutes changing the handout’s decorative layout, which the receiver has said is unnecessary. Which selection fits, and what should the assistant refuse to assume?

Answer one: A, B and C total seventy action minutes. With ten planning and ten reserve, they use the ninety-minute capacity exactly. D is deferred or removed from this decision window because it would displace either a commitment or the chosen practice without improving the receiver’s stated requirement. A must occur before noon and B before its verified deadline; aggregate minutes do not prove clock-time feasibility. The assistant should not assume the receiver agreed to a later handout, that decoration can replace factual checking or that another person will complete A.

Exercise two: Sara has fifty action minutes after planning and reserve have already been deducted. A twenty-minute task requires permission that has not arrived. A ten-minute request could seek that permission. A thirty-minute alternative uses material Sara already has permission to use and meets the same agreed output. A separate ten-minute check is needed before either output is sent. Propose a selection that does not depend on a favourable reply, and identify an attractive but unsupported plan.

Answer two: Sara can spend ten minutes on the permission request, thirty minutes preparing the already-authorised alternative and ten minutes checking it, totalling fifty. She may also decide the request is unnecessary for this window and leave those ten minutes unused. The unsupported plan counts the request as if approval automatically follows, then starts the permission-dependent action without a verified answer. If permission does arrive, Sara should still compare the remaining effort and avoid double-starting both routes. The fact that one route becomes possible does not make switching to it compulsory.

Exercise three: Dev’s original action budget is 120 minutes. He completes a twenty-minute unblocking task and a fifteen-minute source check. He then loses thirty minutes of future capacity. The remaining candidates are a forty-minute promised output, a twenty-minute verification and save step required for that output, and a twenty-five-minute optional improvement. How much action capacity remains, which tasks fit, and what must happen if the essential output cannot be reduced safely?

Answer three: The revised total action budget is ninety minutes. Thirty-five have already been used, leaving fifty-five. The promised output and required verification need sixty, so they do not fit. Dropping the optional improvement is necessary but insufficient. Dev must find an agreed safe reduction of at least five minutes, obtain more real capacity, or renegotiate the affected commitment. The assistant must not announce a feasible plan merely because it removed the twenty-five-minute improvement. Nor may it remove verification while claiming the original completion standard remains satisfied.

Exercise four: Three tasks receive model-generated scores of 8.8, 8.7 and 8.6. The first has an unverified deadline. The second is blocked on another person’s consent. The third is ready and supports your chosen goal. What can you conclude from the scores, and which questions should come before execution? Do not invent additional facts to force a single universal answer.

Answer four: The scores alone establish no trustworthy order. Verify the first deadline and its consequence. Treat the second as permission-blocked, possibly replacing it with an authorised request. Check that the third’s completion condition and effort fit the available capacity. If the deadline verification reveals an immediate genuine obligation, the first task may rise. If it reveals no urgency, the third may be the best ready action. The exercise tests whether you preserve uncertainty, rather than whether you guess the hidden answer the model appears to expect.

Exercise five: A friend has not replied to your proposed delegation. Your assistant removes a thirty-minute collection task from your plan and fills the space with a hobby activity. The collection is an existing promise you made. Identify the first incorrect assumption, produce a repair, and explain whether a well-written message draft resolves the problem.

Answer five: The first incorrect assumption is that the delegation is agreed. Restore the collection to your responsibilities unless the friend accepts or another authorised arrangement is confirmed. Remove or defer enough optional activity to keep the plan feasible. A draft can help you ask clearly, but it does not constitute acceptance by the friend and does not change the original promise. Even a sent request remains a request. The repair must preserve a feasible fallback while waiting, rather than transfer the risk of silence onto the person expecting the collection.

Exercise six: Return tomorrow without looking at the model answers. Write five ordinary candidate actions from a fresh fictional situation, give each an effort and completion condition, and create a budget smaller than the total. Include one genuine commitment, one blocked action and one self-chosen long-term contribution. Select a feasible set, state what waits and explain one fact that would change the decision. Then inspect your answer using the checks below before comparing it with this article.

Checking guidance for exercise six: There is no single correct set because you create the inputs. A complete answer has non-overlapping capacity categories, preserved permission boundaries, an explicit reason for each selected task, a state for every unselected task and a feasible response to the named change. Check arithmetic independently. If your blocked action is selected for execution, repair it into a request or a fallback. If your personal contribution disappears, explain why rather than assuming external requests always win. Transfer means you can defend a new decision, not repeat Mei’s list.

Back to contents · Next: 16. Questions that matter when choosing your next task

16. Questions that matter when choosing your next task

Should I always do the most important task first?

Only if it is eligible, ready and feasible in the available window, and if doing it first does not cause a more time-sensitive commitment to fail. A major creative task may deserve substantial capacity but require materials that are unavailable until later. A brief confirmed reply may need to happen first because someone else is blocked. Separate overall importance from immediate sequence. You can protect the important task’s capacity while performing a small enabling action earlier, provided the smaller work does not expand and consume the space you intended to protect.

Is an urgent–important matrix enough?

It can help begin a conversation, but it does not by itself represent permission, dependencies, effort, capacity or the source of a deadline. Two tasks in the same quadrant may need very different treatment. One may be ready and the other blocked. One may take ten minutes and the other two hours. Use the matrix as a rough view if it helps, then add the facts needed for an actual decision. Do not treat placement in a box as proof that the task can be delegated, delayed or executed safely.

How many priorities should I choose?

Choose a set that fits your actual capacity and serves the decision window. There is no universal number in this guide. Three large actions may exceed a whole afternoon; six small connected steps may form one useful output. Count effort and check completion conditions rather than selecting a fashionable number. Keep the active card short enough to use, but preserve the state of other tasks somewhere reliable. If the choice is between a complicated ranking and one clear next action with an honest deferred list, the latter may be sufficient.

Can SI tell me what should matter most in my life?

It can help you articulate trade-offs, notice inconsistencies and compare the consequences of preferences you provide. It cannot determine your values on your behalf. If the assistant repeatedly treats income, speed, visibility or completion count as the default goal, correct the premise. You may care more about keeping a promise, learning a craft, being available to someone or reducing unnecessary obligations. Those choices should be stated by you and reflected in the task packet. A prioritisation system should make your reasoning more visible, rather than quietly substitute its own.

What if every task really is necessary?

Then the problem may be insufficient capacity, incompatible deadlines or excessive accepted commitments. Ranking alone does not solve it. Identify which constraints can change and who must agree. You may need to reduce scope, move a deadline, arrange an accepted handoff or explain that the complete set cannot be delivered. SI can prepare an accurate comparison of options and a draft request. It should not conceal the shortfall, promise impossible completion or mark essential checks optional simply to make the arithmetic work on paper.

Should I keep asking until the assistant gives a better ranking?

Ask again when you have corrected a fact, clarified a preference or identified a specific defect. Repeatedly requesting “a better answer” without changing the decision packet can produce variation without improvement. Require reasons, check the arithmetic and identify which comparisons are genuinely uncertain. If two choices are both acceptable, a human choice may be enough. Set a review trigger and proceed with a reversible action. The aim is not to obtain an answer that feels increasingly impressive; it is to make a decision that remains understandable when you act on it.

How do I start today without building a complicated system?

Choose one ordinary decision window and list no more detail than you need. Record the usable capacity, a few concrete candidate actions, verified commitments and any blockers. Ask SI for one feasible selection and one meaningful alternative, with reasons for what waits. Check the facts and arithmetic yourself. Execute from a short card, then record what actually happened. Keep only the parts of the method that improve the next decision. A modest, repeatable practice is more useful than an elaborate board that you cannot maintain or no longer trust.

Back to contents · Next: Sources and further routes