How Leadership Works | How People Coordinate Direction, Trust and Action

Leadership Is Partly a Sensemaking Job

Before people can coordinate, someone has to help the group interpret what is happening. Which signals matter? What changed? Is the problem temporary or structural? Which uncertainty can be ignored and which must be investigated?

This is sensemaking: turning a confusing field of observations into a workable shared model without pretending uncertainty has disappeared.

Observe → interpret → expose uncertainty → build a shared model → decide what deserves attention next.

Leadership fails when interpretation becomes storytelling designed to protect status rather than a model that remains correctable by new evidence.

Authority, Power and Influence Are Different

Authority is legitimate decision right attached to a role. Power is the practical ability to shape outcomes. Influence is the ability to change another person’s judgement or action without necessarily possessing formal authority.

A leader may hold authority but little influence because trust has collapsed. A technical expert may hold little rank but substantial influence because others depend on their knowledge. A gatekeeper may possess power through control of information or access without being formally senior.

High-resolution leadership maps all three rather than assuming the organisation chart describes the real control system.

Decision Architecture Matters More Than the Leader Making Every Decision

A scalable leader designs which decisions belong where. Routine, reversible and local decisions can often move outward. Rare, irreversible or cross-system decisions may require higher coordination.

• Local decision: enough context sits close to the work.
• Consultative decision: one owner decides after gathering relevant perspectives.
• Expert decision: specialised knowledge should dominate.
• Collective decision: legitimacy or distributed consequences make participation central.
• Escalated decision: risk, authority or interdependence exceeds the local role.

Leadership quality includes putting the decision at the lowest level that has enough capability, information and authority to make it well.

Leaders Face a Principal–Agent Problem

When one person delegates work to another, their goals and information may not align perfectly. The person doing the work may know more about local conditions, while the person assigning the work may control reward, budget or evaluation.

This creates a recurring leadership problem: how do you create enough autonomy for expertise to operate without making the system impossible to inspect?

Clear outcomes, aligned incentives, transparent reporting and proportionate audit are stronger than either extreme of total control or total trust.

Metrics Become Dangerous When They Replace the Mission

Leaders use metrics because complex systems need compression. But once a metric becomes a target, people may optimise the number while weakening the actual outcome.

A school can raise completion rates by lowering difficulty. A customer-service team can shorten call times by ending difficult calls too quickly. A project can appear “green” because reporting hides unresolved dependencies.

Measure the mission, but keep an independent route back to the receiver so the metric cannot become the whole reality.

Psychological Safety Is an Information Property, Not Simply Comfort

A psychologically safer team is one where people can surface uncertainty, admit mistakes, ask questions and challenge assumptions without disproportionate interpersonal penalty.

This does not remove standards or accountability. It improves the chance that failure information arrives while the system can still respond.

Leadership therefore needs a dual standard: high candour + high responsibility. Safety without standards can become drift; standards without safety can become concealment.

Normal Operations and Crisis Leadership Need Different Control Ratios

In stable conditions, distributed authority, experimentation and local judgement can improve learning and speed. In an acute crisis, temporary centralisation may be necessary because time is short and coordination failure is costly.

The danger is carrying emergency control into normal life after the emergency has passed. Strong leadership changes the control ratio with the environment and then deliberately returns authority outward when the reason for centralisation disappears.

Strategy and Execution Need an Explicit Bridge

A strategy describes choices about where to focus and what trade-offs to accept. Execution converts those choices into owned actions, resources, sequences and review points.

Priority → owner → resources → first action → dependency → measurable receipt → review date → continue, correct or stop.

Without this bridge, organisations can produce sophisticated strategy documents while daily behaviour remains unchanged.

Leadership Must Track Who Receives the Consequence

Leaders often decide far from the people who experience the result. This creates a structural risk: benefits may be visible upward while burdens remain local and weakly reported.

A strong leadership system therefore asks who decides, who implements, who benefits, who bears the cost, and how the receiver can return evidence to the decision layer.

Receiver feedback is not a courtesy. It is part of the control system.

Ethical Leadership Is Not Separate From Operational Leadership

Efficiency cannot answer every leadership question. Some actions may be effective at reaching a target while violating rights, dignity, safety or legitimate process.

Good leadership therefore maintains protected constraints that are not traded away merely because a shortcut would improve one metric. This is especially important when power asymmetry prevents affected people from refusing the trade.

Succession Should Transfer Judgement, Not Just Instructions

A successor who inherits only procedures may struggle when conditions change. Durable leadership transfers principles, decision boundaries, information routes, institutional memory and enough judgement to know when the old procedure no longer fits.

The deeper succession test is therefore not “Can the next person copy me?” but “Can the system continue to make good decisions when I am no longer the reference point?”

A High-Resolution Leadership Audit

1. Reality: What is actually happening, and which observations are being ignored?
2. Sensemaking: What shared model explains the current state?
3. Purpose: What outcome is worth coordinating around?
4. Priority: What matters now, and what is deliberately not being done?
5. Legitimacy: Why should people accept this direction?
6. Authority: Which decisions formally belong to whom?
7. Power: Who can actually shape outcomes regardless of title?
8. Decision architecture: Which choices should be local, expert, collective or escalated?
9. Incentives: What behaviour does the system reward in practice?
10. Metrics: Could people improve the number while harming the mission?
11. Information: Can bad news and dissent reach the leader early?
12. Safety/standards: Can people speak candidly while remaining accountable?
13. Execution: Does every priority have an owner, resource and receipt?
14. Receiver: Who bears the consequence, and can their signal travel back?
15. Adaptation: Which assumptions should change when evidence changes?
16. Succession: What capability and judgement remain after the leader leaves?

Connect Leadership to the Wider eduKateSG Mechanism Estate

• How Power Works — how real ability to shape outcomes can differ from formal rank.
• How Trust Works — why legitimacy and information quality depend on calibrated reliance.
• How Decision-Making Works — how evidence, uncertainty, values and thresholds become commitments.
• How Responsibility Works — how authority, ownership and accountability should remain aligned.
• How Work Works — where leadership becomes production through people, tools and coordinated tasks.

How to Use Super Intelligence to Prepare for Leadership Responsibilities

A first leadership responsibility often arrives as a small, consequential job: make a priority clear, coordinate people with different information, and return a decision to the person who actually owns it. Use this practice lab to prepare a briefing, a delegation and a difficult conversation before trying them in real work.

Here, Super Intelligence (SI) means the AI assistance used to organise and challenge your preparation. The exercise does not assume that a tool has superior judgement about people. You keep the decisions about priorities, commitments and relationships, within your actual authority.

Everything in the two case packets is fictional. All named people are adults; the records, hours and exchanges are invented teaching inputs. The weak AI-style answer and rehearsed replies are authored examples, not transcripts from a tested system. A worked answer shows how to check this case; it is not evidence that this exercise improves leadership performance.

How to practise: read records A1–A6, write your own three-sentence briefing, then compare it with the worked repair. Rehearse the changed-capacity conversation aloud. Finally, complete the second case before opening its checking key. You can do the whole exercise on paper; a live AI tool is optional.

1. Read the responsibility before planning the work

Case A: Priya is acting lead for one internal help-card project. It is Monday morning. Six cards will help colleagues practise a room-booking tool in a training sandbox. Priya is preparing the team’s work, not assessing anybody’s suitability for employment. Treat the following records as the complete starting packet; do not invent extra staff, tools, permissions or working hours.

• A1 · Sponsor’s charter, Monday 09:00. Mina asks for six source-checked, independently tested cards and an issue log by Thursday 13:00, ready for her Thursday 14:00 review. Priya may sequence the work, divide drafting between qualified writers, and defer extras that Mina has not authorised. Mina retains changes to the six-card scope, the deadline, other assigned duties and all distribution approval. Priya cannot authorise overtime, spend money, waive testing or promise publication. If a dependency or failed check threatens the package, Priya must tell Mina promptly and present a bounded choice.
• A2 · Available task time and capabilities. Jo has Tuesday 09:00–11:00 and Wednesday 09:00–11:00: four hours. Jo can verify the current sandbox steps and independently test cards; Jo could also write, but must not independently test any card Jo authored. Arun has Tuesday 13:00–16:00 and Wednesday 13:00–14:00: four hours, for drafting. Priya has Tuesday 15:00–16:00 and Wednesday 11:00–12:00: two hours, for drafting or bundle assembly. These are the only project-task slots after other work and routine coordination. No extra evening work is available.
• A3 · Work and acceptance rules. First verify all six steps against the current approved sandbox version and record the version: two hours, Jo. Then write six cards: three hours total, Arun and/or Priya. Those three hours may be divided between them and include merging the text. Then independently run all six cards in the sandbox and record pass/fail against each step: two hours, Jo. Finally assemble the versioned cards and issue log: one hour, Priya. Each stage needs the previous stage’s completed output. A card passes only if its source version is recorded, its steps match that version, and Jo’s independent run succeeds. Missing evidence is not a pass. Priya checks the review bundle; Mina decides whether it may be distributed.
• A4 · Conflicting request, Monday 09:05. A colleague who will introduce the session asks for an additional illustrated slide deck by Thursday 13:00. It needs four further drafting hours by Arun or Priya and cannot replace any of the six cards. Mina has not approved this extra. The requester says both outputs are important; the request does not change Priya’s charter or the available hours.
• A5 · Incomplete handover. Last Friday’s tracker asked Arun for a draft or a blocker update by 16:00. The handover field is blank. A separate Friday 15:40 message from Arun says: “Two old screenshots disagree. I need Jo’s current step list before I finish the wording.” The supplied records contain no reply and no draft. They do not establish whether another update was sent elsewhere, whether the step list arrived later, or why the tracker was left blank.
• A6 · Communication and return path. At an agreed checkpoint the worker posts the current file, or the specific blocker and the decision needed, in the project tracker and alerts Priya. Priya acknowledges it and updates the plan. A new blocker is reported when noticed, rather than hidden until the next checkpoint. Corrections should state the observation, work impact, question and next agreement without guessing motives. Only the real people involved can confirm what happened and accept the next commitment.

2. Separate competing explanations from established facts

Two interpretations of A5 deserve checking. The work may have been appropriately paused at an unresolved source dependency: Arun’s 15:40 message supports that possibility. The reporting agreement may also have broken down: the required tracker entry is absent. These explanations can both be true. A blocked draft does not automatically explain the missing update, and a missing update does not prove laziness, resistance or lack of leadership potential.

The first questions are narrow: “Did the verified list arrive, and when?” and “Where, if anywhere, was the blocker update recorded?” Check the message and tracker history with the people concerned. If there is no additional record, keep the cause unknown and repair the reporting agreement. Do not turn SI’s fluent explanation into a personnel judgement.

Initial direction in three sentences: “Our priority is six checked cards and a clear issue log for Mina’s Thursday review. The work must follow source verification, drafting, independent testing and assembly, within the listed hours. The slide deck is unapproved extra work; I will explain the capacity conflict and return any requested scope change to Mina.”

3. Give SI a preparation task with a checking rule

If using an approved AI tool, provide only the fictional A1–A6 packet and this prompt. In real work, first check your organisation’s tool and data rules; do not copy private personnel records, health information, disciplinary material or identifiable colleague conversations into a rehearsal. Removing a name alone may not make a situation anonymous.

Preparation prompt: “Use only A1–A6. Prepare a priority recommendation, a person-by-person time and dependency check, a decision-rights map, a delegation to Arun, and an opening question about the missing handover. Cite the record supporting every constraint. Distinguish facts, interpretations and unanswered questions. State what must stop or return to Mina. Do not invent hours, send messages, assign personality traits, make personnel recommendations or claim authority for me. End by listing any statement in your own draft that the packet cannot support. I will verify the answer and decide what to use.”

Deliberately weak AI-style answer: “Show decisive leadership by promising both cards and slides. Arun has eight hours, so give him the writing and let him publish when ready. Jo can verify while Arun starts from Friday’s screenshots, then run the tests if time permits. Arun missed the handover because he resists accountability; tell him that this needs to improve before he can be trusted.”

A confident tone does not make that answer usable. Before rewriting it, identify the first consequence that would be unsafe to carry forward: promising both outputs without enough authorised capacity. Then check the dependency, release and conversation errors separately. A polished rewrite that leaves any of those errors intact is still a failed preparation.

• Capacity error → A2–A4. The core package needs 2 + 3 + 2 + 1 = 8 task-hours. The deck adds 4, making 12 against 10 available. Arun has four hours, not eight. More importantly, Jo’s four hours are already needed for verification and independent testing; the writers have six hours between them, of which four are needed for drafting and assembly. Only two remain, not the four needed for the deck.
• Dependency and quality error → A3 and A5. Old conflicting screenshots cannot stand in for the verified step list. Drafting follows completed verification; testing follows completed drafting. “If time permits” violates an acceptance requirement. If a check fails, keep that failure visible and arrange a repair; do not relabel it as complete.
• Authority error → A1 and A3. Arun may draft. Priya coordinates and checks the bundle. Mina owns distribution approval. Delegating an output does not transfer a decision the lead does not possess.
• Interpretation error → A5 and A6. The blank tracker and source query are observable. Resistance and untrustworthiness are invented explanations. Ask about the missing information, describe the operational impact, and agree a reportable next step.

4. Replace the weak answer with an executable brief

Checked first plan: Jo verifies the six steps Tuesday 09:00–11:00. Arun drafts Tuesday 13:00–16:00. Jo independently tests Wednesday 09:00–11:00. Priya assembles Wednesday 11:00–12:00 for the Thursday 13:00 handover. Priya’s Tuesday 15:00–16:00 and Arun’s Wednesday 13:00–14:00 remain unused in this plan. Those two hours are not interchangeable with Jo’s specialist testing time, and they do not guarantee that every possible defect can be repaired before review.

Decision-rights map: Priya chooses the sequence and the qualified writer allocation, and may defer the unapproved deck. Jo decides whether each test meets the stated criterion and reports the evidence; Jo cannot waive it. Arun decides wording within the verified steps, asking about ambiguity rather than changing the underlying procedure. Mina decides changes to scope, other duties or deadlines and whether to distribute. SI drafts and challenges; it owns none of these decisions.

Completed briefing to the team: “We are preparing six accurate cards for Mina’s review, with one independent run recorded for every card. Jo will produce the verified, versioned step list Tuesday by 11:00; Arun will draft from that list by 16:00; Jo will test Wednesday morning and I will assemble the bundle by noon. Please report a missing source, failed step or lost work slot as soon as it appears, with the file and decision needed. We have ten project hours for an eight-hour core package. We will not commit the additional four-hour deck unless Mina changes the authorised work and we check a feasible replacement plan.”

Completed delegation to Arun: “Please produce six plain-language draft cards from Jo’s verified step list, using Tuesday 13:00–16:00. You may improve wording and layout; do not infer a new booking step or use an unverified screenshot. Put the source version on each card and mark any unresolved wording question. At Tuesday 14:30, post the current file or exact blocker and decision needed in the tracker and alert me; I will acknowledge it. Return the six-card file by 16:00 for Jo’s independent run on Wednesday. If the verified list is missing, stop source-dependent drafting and alert me immediately. If the work no longer fits, bring me the remaining work and available time rather than using overtime. Neither of us is approving distribution.”

Read-back to request: “Tell me what you will return, what you can decide yourself and when you will ask for help.” A satisfactory example is: “Six versioned draft cards by Tuesday 16:00; I can change wording, but not the verified steps. I will post progress at 14:30 and report a source or capacity blocker immediately. Jo tests, and Mina decides distribution.” If a real colleague gives a different read-back, clarify the disagreement before treating the task as accepted.

Message to the deck requester: “The six-card package uses eight of our ten available project hours; your deck needs four more and is not yet authorised. I cannot commit both. I will keep the cards as the current priority and ask Mina to decide if you want a change to scope or resources.” This is a boundary with a return path, rather than a hidden promise that the team will somehow absorb the work.

5. Rehearse feedback, then repair the plan when the facts change

Opening to Arun: “The Friday 16:00 handover field is blank, and I can see your 15:40 question about conflicting screenshots. Without an updated file or blocker record, I cannot tell what is ready for the next stage. Can you walk me through what was ready, which source you were waiting for and where you reported the block? Then let us agree how today’s checkpoint will work.” This names a work problem while leaving the explanation open.

Authored rehearsal reply: Arun says, “I sent the question but did not put it in the tracker. I thought the question was enough. Also, Mina has assigned me queue cover in two of the hours on your roster.” Priya checks the fictional assignment notice rather than treating a new claim as verified merely because it was spoken.

A7 · Verified change introduced for this rehearsal. Mina’s Monday 09:10 notice assigns Arun queue cover Tuesday 15:00–16:00 and Wednesday 13:00–14:00. Those two project hours are no longer available. Priya may not cancel the cover. Arun can still draft Tuesday 13:00–15:00. All other slots, capabilities, task durations and rules stay as stated.

First attempted reply, to critique: “Thank you for explaining. Keep the original three-hour drafting promise and catch up after queue cover. I will tell Mina we are on track.” The opening is civil, but the first consequential error is preserving a commitment after two authorised work hours have disappeared. The reply also implies unlisted work time and hides a new capacity risk from the sponsor.

Repaired reply: “Thank you for explaining where the update was missed. Next time, please put the blocker and decision needed in the tracker as well as alerting me, so the handover is visible. The new queue-cover notice leaves you two drafting hours; I will not ask you to recover them outside the roster. You can draft Tuesday 13:00–15:00, and I can take the remaining drafting hour from 15:00–16:00. Jo’s Wednesday test slot and my assembly hour remain in sequence. Let us confirm the shared file and handoff at 15:00. I will tell Mina that the package now uses all available project time, with no repair reserve.”

Revised arithmetic and commitment: Jo uses four of four hours; Arun uses two of two; Priya uses two of two, split between one drafting hour and one assembly hour. Total: 8 required / 8 available / 0 reserve. The three drafting hours include the merge under A3, and Jo authors none of the cards, so the independent check remains intact. This is a feasible preparation plan under the stated durations, not a promise that all tests will pass.

Updated sponsor briefing: “Your queue-cover change removes two project hours. I have used my permitted drafting slot to keep the six-card review plan within eight hours, without changing cover or testing. There is no remaining repair reserve. If Tuesday’s verified list is late, drafting is incomplete at 16:00, or Wednesday’s run fails, I will hold the affected work as incomplete and return the actual gap to you for a time/resource decision. Distribution still awaits your approval.”

Feedback tied to the rules: The repair separates the handover observation from character, changes a commitment when A7 invalidates it, preserves the other assigned duty, uses a qualified writer within a real slot and keeps an independent test. Its remaining limitation is the zero reserve. The next practice should introduce a failed test or missing input, so Priya must return an unresolved decision instead of rescuing the plan with invented time.

6. Keep a readiness record that means what it says

Completed practice record for this worked example: Scope: one internal help-card project. Initial fault: the weak answer promised unauthorised work and release, ignored a dependency and invented a motive. Repair: an evidence-linked brief, bounded delegation, neutral clarification and a rechecked plan after A7. Evidence retained: the source packet, first answer, marked errors, revised hour allocation and read-back. Unresolved: the real source verification, test results, sponsor approval and how any real team member would respond.

Next decision: this written example is ready for a supervised rehearsal against the same rules. It does not certify a reader as ready to lead, justify an appointment, or establish that a team will trust them. A human supervisor and the people doing the work still need to confirm the actual remit, capabilities and arrangements. In real use, keep an authorised record of decisions and check what happened; do not turn an exercise score into a personnel rating.

7. Independent transfer: a late dependency and different authority

Stop before the checking key. Without asking SI to solve it first, prepare Noor’s next briefing and delegation from T1–T4. Decide whether the complete bundle is feasible, what useful work can proceed, what must stop, and who owns each unresolved decision. Include one sentence explaining why the previous case’s approval arrangement cannot simply be copied.

• T1 · New charter. Noor coordinates four room-briefing cards for an internal orientation. The complete review bundle is due Tuesday 16:00. Unlike Priya, Noor has authority to distribute the final cards without separate sponsor approval, but only after all four use confirmed room allocations and pass the independent check. Ben, the sponsor, controls the four-room scope, deadline and additional staff time. Only Leena, the venue owner, can confirm room allocations. Noor cannot treat Ben’s urgency as confirmation of those facts.
• T2 · Late source and fixed slots. Leena’s current allocation will not be available before Tuesday 15:30. Monday’s allocation is labelled superseded. Rafi is available Tuesday 13:00–15:00; Mei Tuesday 14:00–16:00; Noor Tuesday 15:00–16:00. There are no other approved slots or substitutes. After receiving the confirmed allocation, Rafi needs one hour to verify the room facts. Mei then needs two hours to write all four cards. Rafi then needs one hour to independently check them, and Noor needs one hour to assemble the bundle. These stages are sequential; nobody may sign their own independent check.
• T3 · Useful preparation before confirmation. Rafi can spend half an hour preparing an empty four-room verification checklist; Mei can spend half an hour drafting the generic welcome paragraph, with no room names or directions; Noor can spend half an hour assembling an interim status note. These tasks can be done before the allocation arrives. They do not reduce the five hours needed for the final sequence in T2. The note must say that the final cards are not ready.
• T4 · Conflicting pressure. An organiser asks Noor to reuse Monday’s allocation and mark everything ready by 16:00. Ben’s latest instruction retains all four rooms and says any deadline or resource change must come back to him. No one has approved a reduced scope, later work slots or reuse of superseded information. A late source does not by itself identify who is at fault.

Write a short response before opening the key. Then check your own response, not just whether you prefer the example’s phrasing. Alternative plans can be sound only if they respect every fixed fact or explicitly ask the right human to change one.

8. Check the skill you actually practised

• Evidence: does each factual constraint point to a packet record, with unknowns still visible?
• Direction: is there one clear current priority and an honest account of what is deferred or incomplete?
• Feasibility: do individual hours, capabilities, stage order and actual slots fit, rather than only the grand total?
• Authority: does each person receive only decisions they own, with the right return path for everything else?
• Delegation: are the outcome, input, discretion, evidence, checkpoint, help route and acceptance responsibility explicit?
• Conversation: does the reply distinguish observed work from assumed motive, invite missing context and make a checkable next agreement?
• Adaptation: does a new fact change the plan and commitment, with a visible hold when the conditions cannot be met?

Treat a missing authority boundary, invented capacity, bypassed check or unsupported character judgement as a reason to repair and repeat before using the draft. Do not average those errors away because the wording sounds encouraging. Ask SI to identify the earliest sentence that violates a cited rule, but verify the criticism yourself. Then repeat the explanation without AI help and have a responsible human challenge one changed condition.

Evidence note: this is an educational design, not a validated leadership assessment. The NIST Generative AI Profile (2024) identifies confident false output as a risk, calls for source verification, and distinguishes testing human proficiency from testing AI capability. Those cautions inform the checking approach here; NIST does not endorse this exercise or its example schedules.