How do you break a job into tasks Super Intelligence can help with? Start by treating the job title as a bundle rather than a single activity. List the repeated units of work inside the role, separate information processing from judgment, relationships, physical action and formal authority, then evaluate each task for context, frequency, verifiability, consequence and reversibility.
This article follows How to Map a Workflow Before Adding Super Intelligence. The workflow map shows how work moves across states and people. This page owns the next question: task decomposition. It asks what each person actually does inside those states and which parts are suitable for Super Intelligence.
In the eduKateSG workplace series, Super Intelligence is the practical machine-intelligence layer commonly described as artificial intelligence, generative AI, assistants, copilots, agents and connected automation. The goal is not to automate job titles. The goal is to understand the work precisely enough to redesign the human–SI division of labour.
A Job Is Not One Task
A job title hides many different forms of work. A manager may read reports, prepare meetings, negotiate, approve budgets, coach staff, resolve conflict, write updates and make decisions. A teacher may explain, diagnose misconceptions, prepare materials, mark work, communicate with parents, manage a class and protect student welfare.
These tasks differ dramatically in their suitability for SI. Drafting a first version of a report and deciding whether to terminate an employee can both sit inside a manager’s role, but they should not receive the same autonomy.
Task Decomposition Prevents Job-Level Overclaiming
Statements such as “AI will replace accountants” or “AI can do marketing” are too coarse for operational design. A role is a bundle of tasks, responsibilities, relationships and authority. Some tasks may become heavily automated while the role remains valuable because the remaining tasks become more important.
Task decomposition therefore gives organisations a more realistic way to discuss productivity and job redesign: which units of work change, which disappear, which become faster and which require stronger human capability.
The Seven Task Families
1. Information acquisition
Search, retrieve, read, listen, collect and inspect. These tasks gather raw information. SI can be strong where the sources are accessible and the objective is clear.
2. Information transformation
Summarise, extract, classify, translate, reformat, compare and reorganise. These are often excellent early SI tasks because inputs and outputs can be inspected.
3. Generation
Draft, brainstorm, code, design alternatives, create examples and prepare plans. SI can create candidate artefacts quickly; human or deterministic verification may remain necessary.
4. Analysis and reasoning
Interpret, diagnose, infer, evaluate evidence, identify patterns and propose explanations. These tasks can benefit from SI but require careful attention to evidence quality and uncertainty.
5. Coordination
Schedule, route, notify, assign, hand off, track and monitor. SI can reduce administrative load when connected to tools, though permissions and current state matter.
6. Judgment and authority
Approve, prioritise, negotiate, decide exceptions, allocate resources, accept risk and make consequential choices. SI can prepare evidence, but authority may remain human or institutional.
7. Relationship and embodied work
Build trust, persuade, coach, care, lead, de-escalate, observe physical environments and perform hands-on action. SI may support preparation but often cannot own the full task.
The Task Card
For every meaningful task, write a small card. The card turns a vague duty into something that can be evaluated.
- Task name: one clear verb phrase.
- Trigger: what starts the task?
- Input: what information or object arrives?
- Output: what should leave the task?
- Receiver: who uses the result?
- Frequency: how often does it occur?
- Context: what must be known?
- Standard: what does acceptable work look like?
- Verification: how can the result be checked?
- Consequence: what happens if it is wrong?
- Reversibility: how easily can error be corrected?
- Authority: what may the task decide or change?
The task card becomes the basic unit for SI opportunity analysis. It is small enough to compare across roles and detailed enough to expose hidden risk.
Start With Verbs
Job descriptions are full of nouns: strategy, operations, customer service, leadership, compliance. Task decomposition becomes clearer when those nouns are converted into verbs.
- Research competitors
- Extract contract obligations
- Classify customer requests
- Draft weekly report
- Compare supplier proposals
- Approve budget change
- Negotiate delivery date
- Coach employee
- Inspect equipment
- Escalate incident
- Reconcile account
- Schedule meeting
Verbs reveal the cognitive operation. They also make overlap visible across departments. “Compare” can occur in procurement, legal, finance and research even though the surrounding context differs.
Separate Preparation From Decision
Many apparently judgment-heavy tasks contain a large preparatory component. A manager deciding on a budget may spend hours gathering figures, comparing scenarios and assembling evidence. SI can help with that preparation while the manager retains the decision.
Decomposition therefore creates intermediate automation opportunities that job-level thinking misses.
Separate Decision From Action
Choosing what should happen and executing the choice are different tasks. A system can recommend a refund without issuing it. It can suggest a code deployment without deploying it. It can prepare a hiring recommendation without sending a rejection.
This separation allows organisations to grant useful cognitive leverage without giving unearned authority.
Separate Interpretation From Retrieval
A task may appear to require reasoning when much of the burden is finding context. A lawyer interpreting a clause may spend time locating precedent. A support agent may spend time finding the current policy. A teacher may spend time gathering prior student work.
By separating retrieval from interpretation, the organisation can automate the search burden while preserving expert judgment.
Separate Drafting From Fact Establishment
Drafting language and establishing facts are different tasks. SI can write quickly, but the factual layer should come from authoritative sources or validated data.
This distinction matters for customer communication, reports, proposals, public statements and professional work.
Separate Analysis From Approval
A system can calculate, compare or identify anomalies without being authorised to approve a transaction, policy exception or business commitment.
Task decomposition should therefore mark the point where analysis ends and institutional authority begins.
Separate Human Relationship Work From Administrative Work
Many roles mix relationship and administration. A salesperson may spend time preparing account notes and time building trust. A teacher may prepare worksheets and counsel a learner. A manager may assemble performance data and conduct a difficult conversation.
SI can reduce administrative load so humans can spend more attention on the relationship-heavy part of the role.
The Decomposition Sequence
- Choose one job or role.
- List recurring outputs the role produces.
- For each output, list the major tasks required.
- Convert each task into a verb phrase.
- Split tasks that contain both preparation and authority.
- Split retrieval from interpretation.
- Split decision from external action.
- Mark relationship-heavy and physical tasks.
- Create a task card for each meaningful unit.
- Evaluate SI fit task by task.
The Frequency Test
High-frequency tasks deserve attention because small improvements compound. A five-minute saving that occurs one hundred times per week can matter more than automating a two-hour task performed once a year.
Frequency alone is not enough. High-frequency high-consequence decisions may still require strong human control.
The Information-Intensity Test
Tasks dominated by reading, search, comparison, extraction or transformation are often good SI candidates. The more the task depends on processing large amounts of unstructured information, the more leverage machine intelligence may create.
The Context Test
Can the system receive the local information required for correct work? If the task depends on undocumented history or tacit knowledge, the first project may be context capture rather than automation.
The Standard-Clarity Test
Can the organisation explain what good output looks like? Rubrics, examples, policies, schemas and tests increase task suitability.
If experts disagree about the standard, automation should wait until the ambiguity is understood.
The Verification Test
Can the output be checked without repeating the entire task? Source citations, deterministic checks, tests and qualified review improve task fit.
A task that is easy to generate but expensive to verify may produce little net benefit.
The Consequence Test
What happens if the task is wrong? Consequence affects autonomy. A low-risk internal summary and a high-stakes employment decision should not receive the same workflow design.
The Reversibility Test
How easily can an error be corrected? Reversible tasks make good pilot candidates. Irreversible actions require stronger verification and authority controls.
The Latency Test
Does speed matter? Faster classification may improve customer service. Faster strategic judgment may not be valuable if the decision deserves deliberation.
The Bottleneck Test
Is this task constraining the workflow? Improving a non-bottleneck may have little end-to-end impact. Use the workflow map to identify whether the task is actually limiting throughput, quality or attention.
The Transfer Test
Would improving this task create a reusable capability across other workflows? Retrieval, document extraction, structured handoffs and evaluation are often transferable mechanisms.
Learning value can justify prioritising a task whose immediate return is moderate.
Worked Role: Manager
- Read project updates — strong SI support.
- Summarise risks — strong SI support with source verification.
- Prepare meeting agenda — strong SI support.
- Compare resource scenarios — useful SI collaboration.
- Decide strategic priorities — human-led with SI preparation.
- Coach employee — human-led relationship work.
- Approve sensitive personnel action — human authority.
- Write follow-up note — SI drafting support.
- Monitor commitments — SI/automation opportunity.
The manager role does not disappear. Its task composition shifts toward interpretation, relationships and accountability.
Worked Role: Salesperson
- Research account — strong SI support.
- Prepare call brief — strong SI support.
- Draft follow-up — strong SI support.
- Summarise CRM history — strong SI support.
- Identify open questions — useful SI collaboration.
- Negotiate terms — human-led.
- Build trust — human-led.
- Approve pricing exception — governed authority.
- Record next steps — automation opportunity.
Worked Role: Teacher
- Generate candidate examples — SI support.
- Adapt explanation — SI collaboration.
- Create practice questions — SI support with verification.
- Mark routine structure — possible SI assistance.
- Diagnose misconception — human + SI collaboration.
- Decide instructional response — human pedagogical judgment.
- Motivate learner — human relationship work.
- Manage classroom — embodied human work.
- Communicate progress — SI drafting with teacher ownership.
Worked Role: Accountant
- Extract document fields — strong SI support.
- Reconcile exact figures — deterministic systems first.
- Investigate anomaly — human + SI collaboration.
- Draft variance commentary — strong SI support.
- Interpret accounting consequence — professional judgment.
- Approve material adjustment — governed authority.
- Prepare management report — SI support around authoritative numbers.
Worked Role: Lawyer
- Retrieve clauses and authorities — strong SI support.
- Compare contract versions — strong SI support.
- Prepare issue list — SI collaboration.
- Draft candidate language — SI support.
- Interpret legal consequence — qualified human judgment.
- Negotiate risk allocation — human-led.
- Give final legal advice — professional authority.
Worked Role: Software Engineer
- Explain unfamiliar code — SI support.
- Generate bounded function — SI collaboration.
- Write tests — SI support plus deterministic execution.
- Debug likely causes — SI collaboration.
- Review architecture trade-off — human + SI.
- Approve merge — engineering control.
- Deploy production — governed workflow.
- Handle incident — human-led with SI preparation and diagnostics.
Worked Role: Customer Support Specialist
- Classify request — strong SI candidate.
- Retrieve policy — strong SI candidate.
- Summarise case history — strong SI candidate.
- Draft response — strong SI candidate.
- Decide routine eligibility — possible bounded automation.
- Handle dispute — human-led.
- Make exception — human authority.
- Record case note — automation opportunity.
Worked Role: Operations Manager
- Monitor conditions — automation opportunity.
- Classify exceptions — SI support.
- Prepare incident summary — SI support.
- Compare runbook options — SI collaboration.
- Coordinate teams — human-led with SI support.
- Approve high-impact action — governed authority.
- Conduct postmortem — human + SI synthesis.
Worked Role: Recruiter
- Draft role description — SI support.
- Schedule interviews — automation opportunity.
- Organise candidate evidence — SI support.
- Prepare interview questions — SI support.
- Summarise notes — SI support with caution.
- Judge candidate fit — human-led and fairness-sensitive.
- Make hiring recommendation — human authority with structured evidence.
- Communicate outcome — SI drafting under human ownership.
Worked Role: Research Analyst
- Generate search terms — SI support.
- Find relevant sources — SI/retrieval support.
- Extract study characteristics — SI support.
- Compare evidence — SI collaboration.
- Assess source quality — human judgment.
- Interpret causal meaning — human expertise.
- Draft synthesis — SI support.
- Make recommendation — human-led with SI evidence preparation.
Worked Role: Executive
- Receive brief — downstream use.
- Ask questions of evidence — human + SI.
- Compare scenarios — SI preparation.
- Surface contradictions — SI support.
- Negotiate alignment — human-led.
- Make strategic decision — human authority.
- Communicate decision — SI drafting support.
- Own consequence — human/institutional accountability.
The Task Inventory
A useful task inventory can be built over one week. Ask employees to log recurring tasks in broad categories rather than minute-by-minute surveillance. The objective is to identify patterns, not monitor individuals.
- Task name
- Frequency
- Average active time
- Inputs
- Output
- Receiver
- Systems used
- Context sources
- Consequence
- Verification
- Frustration or friction
- SI candidate role
The Task Inventory Should Capture Hidden Work
Hidden work includes searching, formatting, checking, chasing, copying, reconciling and rebuilding context. These tasks rarely appear in job descriptions but often create the largest SI opportunities.
The Task Inventory Should Capture Rare High-Value Work
Not every valuable task is frequent. Strategy preparation, major proposals or incident response may be rare but important. Include them separately because the correct SI role may be evidence preparation rather than automation.
The Task Inventory Should Capture Unpleasant Work
Tasks employees dislike can be useful signals, but dislike alone does not justify automation. Some disliked tasks are consequential and require expertise. Evaluate the mechanism, not sentiment.
The Task Inventory Should Capture Learning Work
Employees spend time learning policies, systems and adjacent skills. SI can support explanation and retrieval, but organisations should preserve enough independent capability for verification and recovery.
Task Granularity: Not Too Large
“Manage customers” is too large. It contains research, communication, negotiation, record keeping and decision-making. Large tasks hide different SI boundaries.
Task Granularity: Not Too Small
“Open spreadsheet”, “click cell” and “type number” are too small for strategic analysis unless the purpose is detailed software automation. The useful unit normally corresponds to a meaningful cognitive or state-changing operation.
The Granularity Test
A task is usually the right size when it has one dominant verb, one recognisable input, one useful output and one clear verification method. If it contains several verbs with different authority, split it.
The Task Boundary Test
Ask where the task begins and what makes it complete. If completion depends on another person or system, record the handoff rather than pretending the task owns the full outcome.
The Authority Boundary Test
If the task both prepares evidence and approves the result, consider splitting it. This makes later SI integration safer and more flexible.
The Relationship Boundary Test
If part of the task is relational and part administrative, split them. An account manager can automate preparation without automating trust.
The Physical Boundary Test
If the task requires observing or manipulating the physical world, separate digital preparation from embodied action. SI may schedule, explain or analyse while a person performs the physical step.
The Uncertainty Boundary Test
If routine cases are clear and exceptions require judgment, split them. This creates the common pattern: automate the stable core, escalate the edge.
The Task Decomposition Anti-Pattern: Job Replacement Thinking
Starting with “Can SI replace this role?” encourages all-or-nothing reasoning. Start with tasks and redesign the bundle.
Anti-Pattern: Automation by Annoyance
A task is irritating, so the team automates it without considering consequence or hidden learning value. Evaluate task fit first.
Anti-Pattern: Collapsing Preparation and Authority
The same step prepares evidence and approves action. SI then inherits authority accidentally. Split the states.
Anti-Pattern: Ignoring Verification
A task looks easy to generate but difficult to check. The organisation discovers later that review cost removes the benefit.
Anti-Pattern: Ignoring Context Assembly
The team automates drafting while employees still spend most of their time finding the information. Decompose retrieval and drafting separately.
Anti-Pattern: Ignoring Downstream Work
The task becomes faster but the receiver spends more time correcting or clarifying. Include receiver effort in the task card.
Anti-Pattern: Treating Every Repetitive Task as Low Risk
Some repetitive decisions affect rights, money or safety. Frequency does not make consequence disappear.
Anti-Pattern: Treating Every Judgment Task as Non-Automatable
Judgment-heavy work often contains large preparatory subtasks that SI can accelerate. Decompose before concluding.
Anti-Pattern: Treating Physical Work as SI-Irrelevant
Physical roles contain planning, documentation, diagnosis, scheduling and knowledge tasks that can benefit from SI even when the core physical action remains human.
The Human Capability Map
For tasks likely to remain human, record the capability they require: domain knowledge, relationship skill, negotiation, ethical judgment, professional authority, physical dexterity or systems thinking.
This map helps training shift toward the capabilities that become more valuable as routine cognitive production becomes cheaper.
The Capability-Preservation Test
Ask whether automating a task would remove practice from a skill humans still need for verification or recovery. If yes, design deliberate practice, explanation or fallback training.
The Task-to-Role Map
After decomposition, assign a likely SI role: Assistant, Copilot, Coworker, Agent or no SI. Then compare with the autonomy level appropriate for the workflow.
This keeps role and task separate. A low-risk extraction task may be an Agent subtask inside a workflow whose final judgment remains human.
The Task-to-Control Map
For every SI candidate, write the control required: source check, schema validation, test, approval, spending limit, read-only permission or escalation.
Controls should be proportional to consequence, not uniform across all tasks.
The Task-to-Metric Map
Choose the metric closest to the task’s contribution: search time, draft acceptance, classification accuracy, error rate, queue time, response time, reviewer effort or downstream rework.
The task metric should connect to the workflow outcome so local optimisation does not become organisational waste.
What This Article Owns
This page owns job-to-task decomposition. It does not own workflow mapping, use-case prioritisation or bottleneck diagnosis. The workflow map identifies where work moves; this article identifies the units of work inside roles; the next articles decide which of those units create the highest leverage.
The deletion test is direct: if this page disappeared, the workplace series would jump from workflow states to use cases without a precise method for breaking human roles into evaluable task units.
The Task Decomposition Matrix
After the first task inventory, place tasks into a matrix using two axes: machine suitability and human authority requirement. This separates tasks that are cognitively easy for SI from tasks the organisation should still keep under human control.
- High SI suitability + low authority: strong candidates for assistance or automation, such as summarisation, extraction and routine classification.
- High SI suitability + high authority: strong candidates for preparation, not blind execution, such as contract analysis, financial anomaly review or employee evidence packets.
- Low SI suitability + low authority: may remain human or be solved with ordinary software if the value is small.
- Low SI suitability + high authority: usually human-led; SI may support context or documentation only.
The matrix prevents a familiar mistake: because SI can perform the cognitive part, the organisation assumes it should also inherit the decision right.
The Task Decomposition Ladder
Level 1 — Duty
A duty is broad: manage customer accounts, produce financial reports, teach mathematics, maintain software. Duties are too large for SI-fit analysis.
Level 2 — Work package
A work package produces a recognisable deliverable: prepare account brief, close the month, plan the lesson, implement a feature.
Level 3 — Task
A task has one dominant verb and clear input/output: retrieve customer history, compare actuals with budget, generate practice questions, write unit tests.
Level 4 — Step
A step is the smallest operational action that matters for automation: query CRM, extract amount, run test, send notification. Detailed step-level mapping is useful only when integration or automation requires it.
Most workplace SI decisions belong at task level. Work packages are useful for human delegation; steps are useful for technical implementation.
How to Decompose a Knowledge-Work Role
- List the outputs the role produces in a normal week.
- For each output, list the work packages that create it.
- Turn each work package into verb-based tasks.
- Separate information acquisition from interpretation.
- Separate generation from verification.
- Separate preparation from judgment.
- Separate judgment from execution.
- Mark relational, physical and authority-bearing components.
- Create task cards for the recurring or high-value tasks.
- Score task fit and bottleneck relevance.
How to Decompose a Frontline Role
Frontline roles mix real-time interaction, policy, record keeping and physical or relational work. Start from the customer or service event rather than the job description.
A retail employee may greet, identify need, retrieve product information, answer questions, check stock, recommend options, process payment, resolve exceptions and document an issue. SI fit differs at each point.
How to Decompose a Professional Role
Professional roles contain research, interpretation, drafting, judgment and formal responsibility. Separate the evidence-processing work from the professional conclusion.
A lawyer can delegate comparison and retrieval more aggressively than legal advice. A clinician can use SI for record organisation without delegating diagnosis. An accountant can automate extraction while retaining professional sign-off.
How to Decompose a Leadership Role
Leadership work includes information compression, scenario analysis, relationship management, negotiation, authority and accountability. SI can reduce the cost of preparation, but leadership value often remains in choosing priorities and aligning people under uncertainty.
Do not measure leadership automation by how many decisions the model makes. Measure whether leaders receive better evidence and spend more attention on the decisions that require them.
How to Decompose a Physical Role
Physical jobs still contain digital and cognitive tasks: scheduling, diagnosis, inspection planning, documentation, parts lookup, safety checks and coordination.
Separate the physical action from the knowledge work around it. SI may improve preparation and troubleshooting even when a person must perform the physical task.
How to Decompose a Creative Role
Creative work includes research, ideation, selection, production, taste, revision and audience judgment. SI can expand options and accelerate first drafts, while humans remain central to intent, taste and originality.
Split divergent tasks—generate many possibilities—from convergent tasks—choose which possibility deserves to exist. SI is often strongest in divergence; human judgment often dominates convergence.
How to Decompose a Decision-Making Role
A decision can be broken into problem framing, evidence gathering, option generation, comparison, uncertainty assessment, recommendation, authority and execution.
SI may help with most of the first six while the accountable person retains authority. This decomposition creates a safer and more useful alternative to asking the model “What should we do?”
How to Decompose a Communication Role
Communication includes fact gathering, audience analysis, message design, drafting, factual verification, tone review, approval, sending and follow-up.
The draft may be highly automatable while factual authority and promises remain governed.
How to Decompose a Research Role
Research includes question framing, search design, source discovery, source evaluation, extraction, synthesis, interpretation, uncertainty and recommendation.
SI can dramatically increase throughput in discovery and extraction. Human expertise remains essential for source quality and meaning.
How to Decompose a Compliance Role
Compliance work includes requirement identification, evidence collection, comparison, issue classification, interpretation, remediation tracking and sign-off.
SI is useful for evidence organisation and comparison, but regulatory or professional authority remains with the appropriate role.
How to Decompose an Operations Role
Operations work contains monitoring, classification, triage, scheduling, coordination, escalation, action and post-event learning.
Routine monitoring and classification can be strong SI candidates. High-impact operational actions need explicit authority and rollback.
The Task Input Types
- Text: emails, reports, contracts, tickets and notes.
- Structured data: tables, fields, metrics and transactions.
- Images: screenshots, diagrams, forms and visual evidence.
- Audio/video: meetings, calls and demonstrations.
- System state: live records, permissions, inventory, project status.
- Human context: relationship history, tacit judgment and local knowledge.
- Physical state: real-world conditions observed by people or sensors.
Tasks become easier to evaluate when the input type is explicit. A model may handle text well but lack the live system or physical state needed for correct action.
The Task Output Types
- Information: summary, explanation or retrieved evidence.
- Structured fields: extracted or classified data.
- Artefact: draft, report, code or plan.
- Recommendation: proposed action or ranked options.
- Decision: authorised choice.
- External action: changed record, sent message or executed tool call.
- Relationship effect: trust, persuasion, coaching or negotiation.
- Physical effect: real-world manipulation or service.
Output type predicts control requirements. Information and drafts are easier to reverse than decisions and external actions.
The Task Risk Types
- Factual risk: output may contain incorrect claims.
- Decision risk: wrong recommendation may influence a consequential choice.
- Authority risk: system may exercise power it should not have.
- Privacy risk: sensitive information may be exposed or misused.
- Security risk: tool access may alter protected systems.
- Fairness risk: decisions about people may be systematically distorted.
- Relationship risk: communication may damage trust.
- Operational risk: errors may disrupt service or create rework.
The Task Verification Types
- Source check: compare claim with authoritative evidence.
- Calculation: recompute exact numbers.
- Schema validation: verify structure and required fields.
- Test execution: run software or technical checks.
- Peer review: another qualified person evaluates output.
- Professional review: authorised expert interprets consequence.
- World return: external system confirms action.
- Outcome measurement: downstream result confirms usefulness.
The Task Dependency Types
Tasks can depend on other tasks, information or approvals. Mark dependencies because an apparently automatable task may still wait on a human upstream.
For example, drafting a report is easy to automate, but the report cannot be accurate until five teams submit current data. The bottleneck lies elsewhere.
Task Bundles That Should Stay Together
Some tasks should remain bundled because separating them increases coordination cost or destroys context. A difficult customer conversation may combine listening, interpretation and response in real time. Splitting every sentence into SI sub-tasks may reduce rather than improve performance.
Decomposition is an analytical tool, not a requirement to fragment work indefinitely.
Task Bundles That Should Be Split
Split when one subtask has a different authority, verification or risk level. Preparing a payment and approving it should be separate. Drafting a public statement and authorising publication should be separate.
The Task Adjacency Test
If SI performs one task, examine the tasks immediately before and after it. The biggest value may come from connecting adjacent context or handoff rather than improving the task itself.
This test links task decomposition back to the workflow map.
The Task Saturation Test
Ask what happens if SI makes the task ten times faster. Does the workflow improve, or does work pile up at the next state? A task whose acceleration creates a queue downstream is not the sole bottleneck.
The Task Quality Test
Define quality before testing. Drafting quality may include factual accuracy, completeness, tone and usability. Classification quality may include accuracy, calibration and exception handling.
A vague “looks good” standard makes it difficult to know whether SI is actually helping.
The Task Autonomy Test
Decide whether SI should Assist, Collaborate, Automate or Operate around the task. The same task can exist at several autonomy levels.
For example, email drafting can remain assistive, become collaborative with customer context, or become automated for a narrow low-risk class.
The Task Permission Test
What information and tools does SI need? A task that only requires a public document should not receive customer records. A task that only needs to draft should not receive send permission.
The Task Exception Test
What cases should leave the normal SI path? Missing information, conflicting sources, sensitive categories or high consequence may trigger escalation.
A task without an exception rule is rarely ready for high autonomy.
The Task Learning Test
How will recurring corrections improve the task design? Update the source, instruction, example, taxonomy, validation or boundary rather than relying on reviewers to repeat the same fix forever.
Worked Decomposition: Weekly Report
- Collect updates.
- Check completeness.
- Retrieve prior commitments.
- Normalise terminology.
- Compare progress with milestones.
- Identify blockers.
- Draft narrative.
- Check figures.
- Prioritise risks.
- Approve report.
- Publish and route actions.
Tasks 1–7 contain strong SI opportunities. Task 8 may use deterministic validation. Task 9 is collaborative judgment. Task 10 is human authority. The job-level label “reporting” hides this structure.
Worked Decomposition: Customer Complaint
- Identify customer.
- Retrieve history.
- Classify complaint.
- Find policy.
- Summarise evidence.
- Generate resolution options.
- Decide exception.
- Draft response.
- Approve commitment.
- Send.
- Record case.
- Monitor follow-up.
SI can support most information tasks, but exception and commitment authority may remain human. Routine classes can later receive more automation.
Worked Decomposition: Contract Review
- Identify contract type.
- Retrieve standard template.
- Extract changed clauses.
- Compare obligations.
- Flag unusual terms.
- Retrieve relevant guidance.
- Draft issue list.
- Interpret legal consequence.
- Negotiate changes.
- Approve final position.
- Store final agreement.
The decomposition shows a large machine-assistable preparation layer around a smaller but critical professional judgment layer.
Worked Decomposition: Hiring
- Define role criteria.
- Draft job description.
- Publish role.
- Receive applications.
- Extract job-related information.
- Schedule interviews.
- Prepare interview questions.
- Conduct interviews.
- Organise evidence.
- Evaluate candidates.
- Make decision.
- Communicate outcome.
- Onboard selected candidate.
Administrative tasks are strong SI candidates. High-impact employment judgments require clear fairness, privacy and human accountability.
Worked Decomposition: Software Feature
- Clarify requirement.
- Inspect codebase.
- Plan change.
- Write code.
- Write tests.
- Run tests.
- Review security and architecture.
- Open pull request.
- Review change.
- Merge.
- Deploy.
- Monitor.
SI can contribute across the technical preparation chain, while deployment and approval remain governed by engineering controls.
Worked Decomposition: Lesson Planning
- Define objective.
- Review student evidence.
- Identify misconception.
- Choose explanation.
- Generate examples.
- Design practice.
- Verify answers.
- Plan sequence.
- Teach.
- Observe response.
- Adjust instruction.
- Retest.
SI can expand candidate materials and explanations, but the educator remains responsible for learner diagnosis and instructional judgment.
Worked Decomposition: Executive Decision
- Frame decision.
- Gather evidence.
- Validate sources.
- Generate scenarios.
- Compare trade-offs.
- Identify assumptions.
- Surface risks.
- Seek stakeholder views.
- Decide.
- Communicate.
- Allocate resources.
- Monitor outcome.
SI can compress and expand evidence around the decision, but executive authority and organisational consequence remain human.
Worked Decomposition: Procurement
- Define requirement.
- Collect proposals.
- Extract commercial terms.
- Normalise scope.
- Compare service levels.
- Check exclusions.
- Assess risk.
- Request clarification.
- Prepare decision packet.
- Select supplier.
- Approve commitment.
- Contract and monitor.
The machine-suitable layer is substantial, but selection and commitment remain governed.
Worked Decomposition: Incident Response
- Detect alert.
- Gather telemetry.
- Classify severity.
- Build timeline.
- Generate hypotheses.
- Check runbooks.
- Choose containment.
- Execute containment.
- Communicate status.
- Recover service.
- Verify recovery.
- Conduct postmortem.
SI can help with evidence and hypotheses. High-impact containment and public communication may require stronger human control.
Repair Path: Job Is Too Broad
If the role cannot be decomposed because every task sounds vague, return to outputs. What does the role actually produce each day or week? Build the task list around those deliverables.
Repair Path: Standards Are Unclear
If experts disagree on what good work looks like, gather representative examples and define the acceptance criteria before automating.
Repair Path: Tacit Knowledge Dominates
Interview experienced workers and capture the cues they use. Some tacit knowledge may become explicit context; some will remain human judgment.
Stabilisation Path: SI Helps One Expert Only
Compare what the expert provides that others do not: better sources, better decomposition, stronger domain knowledge or better review. Convert reusable parts into shared task cards and examples.
Extension Path: Task Is Stable
Once a task is stable in collaborative mode, connect the required source or tool, automate deterministic checks and define exceptions. Expand one permission at a time.
The Task Decomposition Audit
- Does every major task have one dominant verb?
- Are retrieval and interpretation separated?
- Are generation and verification separated?
- Are preparation and approval separated?
- Are decision and execution separated?
- Are relationship-heavy tasks visible?
- Are physical tasks visible?
- Is authority visible?
- Are inputs and outputs named?
- Can quality be defined?
- Can the result be verified?
- Is consequence known?
- Is reversibility known?
- Is frequency known?
- Is bottleneck relevance known?
- Is the likely SI role clear?
Frequently Asked Questions
How small should a task be?
Small enough to have one dominant cognitive operation and clear input/output, but large enough to produce a meaningful result. Split tasks when authority or verification changes.
Should every job be decomposed?
Not in full detail. Prioritise roles with repeated information work, visible bottlenecks or high strategic importance.
Can SI help with judgment-heavy roles?
Yes, often through evidence gathering, comparison, scenario generation and drafting. The final judgment can remain human.
What is the easiest task type to start with?
Frequent, information-heavy, low-to-moderate-risk tasks whose output can be checked: summarisation, extraction, comparison, first drafts and retrieval are common examples.
What task types should stay human?
Relationship-intensive, embodied, high-authority or hard-to-verify tasks often remain human-led, though SI can support preparation.
How do I know if I decomposed too far?
If the units are clicks and keystrokes rather than meaningful cognitive or state-changing operations, move one level up.
How do I know if I decomposed too little?
If one task contains several verbs with different authority or verification needs, split it.
What comes next?
Continue to How to Find High-Leverage Super Intelligence Use Cases in Your Workplace. Task decomposition creates the candidate inventory; the next article identifies which tasks deserve priority.
The Core Task-Decomposition Rule
Do not automate the job title. Decompose the role until the real cognitive operation, evidence, authority and outcome are visible.
When work is divided at the right level, Super Intelligence becomes easier to place, humans retain meaningful responsibility and the organisation can redesign roles without relying on simplistic replacement narratives.
Department Task Library: Marketing
- Research customer language
- Summarise market evidence
- Compare competitor positioning
- Generate campaign angles
- Draft copy variants
- Repurpose long content into shorter formats
- Check factual claims
- Apply brand guidance
- Schedule content
- Analyse performance
- Interpret customer response
- Decide positioning
Research, summarisation, variant generation and repurposing are strong SI candidates. Claim verification, taste and final positioning remain governed by human strategy.
Department Task Library: Sales
- Research account
- Summarise prior interactions
- Prepare meeting brief
- Generate discovery questions
- Draft follow-up
- Update CRM
- Compare proposals
- Prepare commercial options
- Negotiate terms
- Approve discount
- Manage relationship
- Forecast opportunity
Preparation and record-keeping create substantial SI leverage. Negotiation, pricing authority and relationship judgment remain more human-centred.
Department Task Library: Customer Service
- Identify customer
- Retrieve history
- Classify issue
- Find relevant policy
- Summarise context
- Draft response
- Check entitlement
- Route exception
- Approve special resolution
- Send communication
- Record case
- Monitor follow-up
This role is especially suitable for routine/exception decomposition. Stable cases can move toward automation while disputes and high-value exceptions remain human-led.
Department Task Library: Finance
- Extract invoice data
- Reconcile figures
- Investigate variance
- Compare periods
- Draft commentary
- Prepare forecast scenarios
- Check supporting evidence
- Approve adjustment
- Authorise payment
- Close reporting period
- Explain results to management
The line between exact calculation and interpretive language matters. Deterministic systems should own arithmetic and authoritative state; SI can support explanation and evidence organisation.
Department Task Library: Human Resources
- Draft job description
- Schedule interviews
- Prepare onboarding pack
- Answer policy questions
- Organise interview notes
- Summarise employee feedback
- Draft training material
- Prepare performance evidence
- Evaluate employee
- Approve promotion
- Handle grievance
- Conduct sensitive conversation
Administrative and knowledge tasks are highly assistable. Consequential employment decisions, fairness-sensitive judgments and relationship work require stronger human ownership.
Department Task Library: Legal
- Retrieve precedent or clause
- Compare contract versions
- Extract obligations
- Summarise authority
- Draft issue list
- Draft candidate language
- Identify factual gaps
- Interpret legal effect
- Advise client
- Negotiate position
- Approve final wording
A large preparation layer can be accelerated without giving SI final legal authority.
Department Task Library: Procurement
- Define requirement
- Collect proposals
- Extract pricing
- Normalise terms
- Compare scope
- Check exclusions
- Identify supplier risk
- Prepare clarification questions
- Build decision packet
- Select supplier
- Approve commitment
Comparison and evidence preparation are machine-suitable. Supplier selection and commercial commitment remain governed decisions.
Department Task Library: Operations
- Monitor state
- Detect exception
- Classify incident
- Retrieve runbook
- Prepare diagnostic summary
- Coordinate response
- Schedule resources
- Approve high-impact action
- Document event
- Conduct postmortem
- Track corrective actions
Monitoring and diagnostics can be highly automated, while high-impact action and cross-team coordination often remain human-led.
Department Task Library: IT
- Classify ticket
- Retrieve known solution
- Draft response
- Gather system state
- Run diagnostic
- Prepare change
- Grant access
- Reset credential
- Deploy change
- Monitor outcome
- Escalate incident
Low-risk repeatable service tasks can move toward automation; privileged actions and production changes require stronger controls.
Department Task Library: Software Engineering
- Read requirement
- Inspect code
- Plan implementation
- Generate code
- Write tests
- Run tests
- Debug
- Review security
- Review architecture
- Open pull request
- Approve merge
- Deploy
- Monitor
The task inventory reveals a spectrum rather than one “coding” task. SI can contribute throughout, while production authority remains controlled.
Department Task Library: Research
- Frame question
- Generate search strategy
- Find sources
- Extract evidence
- Compare findings
- Evaluate source quality
- Identify contradictions
- Draft synthesis
- Interpret meaning
- Recommend action
- Publish
High-throughput evidence handling is a strong SI domain. Research judgment stays anchored to source quality, method and uncertainty.
Department Task Library: Education
- Define learning objective
- Review learner evidence
- Diagnose misconception
- Generate explanation
- Generate examples
- Create practice
- Verify answers
- Teach
- Observe response
- Provide feedback
- Reteach
- Assess independent performance
SI can greatly expand candidate instructional material while teachers retain diagnosis, sequencing, safeguarding and relationship judgment.
Department Task Library: Leadership
- Review briefing
- Identify strategic question
- Request evidence
- Compare scenarios
- Challenge assumptions
- Negotiate alignment
- Prioritise
- Allocate resources
- Communicate decision
- Own consequence
- Monitor result
Leadership decomposition shows why SI often creates leverage around the decision rather than replacing the decision.
Task Prioritisation Band A — High-Leverage Preparation
These tasks are frequent, information-heavy, checkable and low-to-moderate risk. Examples include retrieval, summarisation, extraction, comparison, first drafts and structured handoffs.
They are often the best early SI candidates because they reduce cognitive load without moving authority too far.
Task Prioritisation Band B — Collaborative Judgment
These tasks involve interpretation, trade-offs or domain knowledge but can benefit from machine-prepared evidence. Examples include scenario comparison, anomaly analysis, research synthesis and diagnosis support.
Use SI to widen and organise the evidence while retaining human judgment.
Task Prioritisation Band C — Routine Automation
These tasks are stable, repeatable and measurable with clear exceptions. Classification, routing, reminders, structured extraction and routine tool actions may move toward automation.
Automation should follow a proven collaborative version rather than precede it.
Task Prioritisation Band D — Consequential Authority
These tasks make or execute decisions affecting rights, money, safety, employment, security or major commitments. SI can support preparation, but authority should remain bounded and governed.
Task Prioritisation Band E — Relationship and Embodied Work
These tasks depend heavily on presence, trust, negotiation, empathy, physical skill or situational awareness. SI may help with preparation and documentation, but the human contribution is often the core value.
The Task Opportunity Formula
A useful task opportunity can be thought of qualitatively as Value × Frequency × Cognitive Fit × Verifiability ÷ Risk and Integration Cost. This is not a mathematical score that should be treated as exact. It is a reminder that leverage depends on several interacting dimensions.
A high-frequency task with weak verification can be a poor automation candidate. A low-frequency task with strong strategic value can still justify collaborative SI if it saves scarce expert time.
The Task Friction Types
- Search friction: finding information repeatedly.
- Reading friction: processing long material.
- Transformation friction: converting formats or structures.
- Comparison friction: holding many alternatives in working memory.
- Drafting friction: turning intent into first-pass output.
- Coordination friction: moving work across people and systems.
- Monitoring friction: repeatedly checking for change.
- Verification friction: inspecting output for errors.
- Exception friction: handling cases outside the normal path.
Task decomposition should identify the specific friction. That makes the SI role more precise.
The Task Leverage Types
- Speed leverage: complete the same task faster.
- Scale leverage: process more cases with similar quality.
- Quality leverage: improve consistency, completeness or evidence.
- Attention leverage: redirect human focus toward exceptions and judgment.
- Knowledge leverage: make organisational information easier to reuse.
- Coordination leverage: improve handoffs and follow-through.
- Learning leverage: accelerate explanation, practice and feedback.
- Coverage leverage: inspect more alternatives, documents or cases than humans could manually.
Different tasks create different forms of leverage. A project should state which one it expects.
Task Decomposition and Role Redesign
Once SI takes more of the preparation layer, the human role may change. A customer-service specialist may spend less time searching policy and more time resolving unusual cases. An analyst may spend less time extracting and more time interpreting. A teacher may spend less time generating routine practice and more time diagnosing misconceptions.
Role redesign should be explicit so the organisation knows where the released capacity goes.
Task Decomposition and Performance Management
If SI changes the task bundle, performance expectations should change too. Do not keep measuring employees primarily on first-pass production if the system now performs much of it.
New measures may emphasise decision quality, exception resolution, customer outcomes, learning, system improvement or knowledge contribution.
Task Decomposition and Training
Training should follow the remaining and emerging tasks. People may need stronger verification, problem framing, systems thinking and relationship skills.
The goal is not to make everyone an AI engineer. It is to make employees capable operators of their redesigned work.
Task Decomposition and Capability Atrophy
If humans stop performing a task entirely, decide whether the underlying skill must remain. Skills needed for verification, exception handling or outage recovery may require deliberate practice.
Not every automated skill needs preservation. The organisation should identify which ones are operationally critical.
Task Decomposition and Job Architecture
HR can use task inventories to update job descriptions around outcomes and human capabilities rather than lists of routine production activities that may change quickly.
This can make roles more resilient as tools evolve because the job is anchored to responsibility and outcome.
Task Decomposition and Team Design
When many roles contain similar retrieval, drafting or monitoring tasks, a shared SI capability can support the team. When tasks require different data and authority, local workflows may remain better.
Task-level analysis therefore informs which capabilities should be centralised.
Task Decomposition and Workflow Mapping
The workflow map and task inventory should connect. The workflow tells where the task occurs; the task card tells what cognitive operation happens there.
Together they create a complete design surface: state transition plus task characteristics.
Task Decomposition and Bottleneck Analysis
A task can be highly automatable without being worth automating first. The bottleneck article later in this series identifies which tasks constrain throughput or quality.
Use decomposition to create the candidate list; use bottleneck analysis to choose where improvement matters most.
Task Decomposition and High-Leverage Use Cases
The next article turns the task inventory into a use-case portfolio. It will compare value, feasibility, risk and transfer learning so the organisation can choose a small number of projects instead of trying to automate everything.
Task Transfer Test
Before copying an SI task from one role to another, compare input, output, context, consequence, verification and authority. The verb may be identical while the risk differs.
“Summarise” in an internal meeting and “summarise” a medical or legal record are not the same operating problem.
Task Standardisation Test
If employees perform the task differently for no useful reason, standardise the process before automating. If variation reflects legitimate judgment, preserve it.
Task Exception Test
Review the last ten unusual cases. Do they share a pattern? If yes, the task may be decomposed further into routine and exceptional branches.
Task Queue Test
If the task is accelerated tenfold, can the next state absorb the output? This protects against local productivity that creates downstream overload.
Task Review Test
Estimate the time required to review one SI output. Multiply by expected volume. If the review burden exceeds available capacity, redesign the task or validation.
Task World-Return Test
For task outputs that trigger action, identify how the organisation knows the external state changed. A generated action plan is not the same as completed action.
Task Data-Minimisation Test
List the fields the task actually needs. Do not give SI broader access merely because the data exists.
Task Source-Authority Test
Identify which source is authoritative for every material fact. If authority is unclear, improve the knowledge layer before high-autonomy use.
Task Currentness Test
State how fresh the information must be. A task using rapidly changing state may need real-time retrieval immediately before action.
Task Recovery Test
If SI performs the task incorrectly, can the organisation return to a safe state? Reversibility influences pilot design and permission scope.
Task Portfolio Worksheet
- Task
- Role
- Frequency
- Human time
- Main friction
- Input type
- Output type
- Context source
- Verification method
- Consequence
- Reversibility
- Authority level
- Bottleneck relevance
- Likely SI role
- Expected leverage
- Pilot priority
A spreadsheet with these columns is enough to compare dozens of tasks without losing the distinctions that matter.
The 60-Minute Job Decomposition Workshop
Minutes 0–10 — Outputs
List the recurring outputs the role is responsible for.
Minutes 10–25 — Tasks
Break each output into verb-based tasks. Split preparation from decision and decision from action.
Minutes 25–40 — Task cards
Add frequency, context, verification, consequence and reversibility.
Minutes 40–50 — SI roles
Mark candidate tasks for Assist, Collaborate, Automate or human-led work.
Minutes 50–60 — Priority
Choose the three tasks with the strongest combination of leverage, bottleneck relevance and manageable risk.
The One-Week Task Diary
For roles whose work is difficult to remember accurately, run a one-week task diary. Employees log broad tasks, not keystrokes. Capture what triggered the task, active time, wait time, systems used and whether rework occurred.
The diary often reveals hidden coordination and search work that formal job descriptions omit.
The Manager Calibration Session
After task diaries, compare employee and manager views. Managers may overestimate strategic work and underestimate administrative friction. Employees may overlook rare but critical responsibilities.
The calibrated inventory becomes a stronger basis for SI redesign.
The Receiver Calibration Session
Ask downstream colleagues what they actually need from the role’s outputs. This identifies task steps that exist only because the handoff is unclear.
The Decomposition Review Rule
Review task decomposition when the workflow changes materially, SI gains new capabilities, new tools are connected or the role’s authority changes.
Task inventories should evolve with work rather than become static HR documents.
The Final Task-Decomposition Checklist
- Role outputs identified.
- Recurring work packages listed.
- Tasks expressed as verb phrases.
- Retrieval separated from interpretation.
- Generation separated from verification.
- Preparation separated from approval.
- Decision separated from action.
- Relationship-heavy work marked.
- Physical work marked.
- Inputs and outputs named.
- Frequency estimated.
- Context sources named.
- Quality standard defined.
- Verification method defined.
- Consequence and reversibility understood.
- Authority level visible.
- Bottleneck relevance assessed.
- Likely SI role assigned.
- Human capabilities to preserve identified.
- Top three pilot candidates selected.
The Decomposition Rule in One Sentence
Break the role until each unit of work has one dominant operation, one visible authority boundary and one test of success.
At that level, SI fit becomes far easier to judge. The organisation can redesign jobs without pretending every task is equivalent and can increase machine leverage without making human responsibility disappear.
Five Task-Redesign Patterns
Pattern 1 — Compress
SI reduces the cost of reading or processing large amounts of information. Examples include summarising case histories, scanning research, comparing documents and preparing decision briefs.
Compression is useful when the human still needs to understand the result but cannot manually inspect every source at full depth. The redesign should preserve links back to evidence.
Pattern 2 — Expand
SI increases the number of alternatives a human can consider: campaign concepts, explanations, scenarios, code approaches or practice questions.
The human role shifts from generating one option to selecting and improving among many. Expansion is valuable only when the selection standard is strong.
Pattern 3 — Structure
SI converts messy information into consistent fields, categories, tables or handoff packets. This is powerful because structured output can feed deterministic systems and downstream workflows.
Verification should confirm that important information was not lost during structuring.
Pattern 4 — Monitor
SI or automation watches for changes, missing responses, threshold breaches or recurring patterns. Humans then respond to exceptions instead of polling continuously.
Monitoring redesign changes attention allocation rather than only production speed.
Pattern 5 — Prepare
SI gathers evidence and creates candidate output immediately before a consequential human decision or action.
Preparation is one of the safest ways to use SI in high-authority roles because it expands capacity without transferring the final decision right.
Task Contracts
For tasks chosen for SI, write a compact contract. The contract is more useful than a generic prompt because it defines the operating boundary.
- Purpose: why the task exists.
- Input: what the system receives.
- Approved context: which sources may be used.
- Operation: summarise, extract, compare, draft, classify or another verb.
- Output: the required structure.
- Verification: how important claims are checked.
- Prohibitions: what the system must not decide or do.
- Escalation: when the task stops.
- Receiver: who uses the output next.
- Closure: what proves the task has been accepted.
Task contracts make SI use transferable. Another employee can understand the same boundary without reverse-engineering a private conversation.
Capability Preservation by Task Type
Verification-critical skill
If humans must still verify the task, preserve enough underlying expertise to recognise errors. This applies to financial modelling, code, professional analysis and many other domains.
Recovery-critical skill
If people must take over during outages or exceptions, preserve the manual capability through practice, documentation or drills.
Judgment-critical skill
If the task teaches the domain judgment needed for later leadership or professional decisions, do not automate away all learning opportunities for junior staff.
Obsolete skill
Some tasks genuinely no longer need human practice when reliable software replaces them. Do not preserve work merely because it is familiar.
Task Decomposition and Career Development
As SI absorbs more routine cognitive production, junior employees may receive fewer opportunities to practise foundational work. Organisations should redesign development pathways deliberately.
One approach is to let SI accelerate production while requiring juniors to explain, verify and correct the output. Another is to maintain selected manual exercises or rotations. The goal is to preserve the reasoning required for higher-level tasks without forcing employees to repeat low-value work indefinitely.
Task Decomposition and Senior Expertise
Senior experts often create value by identifying what matters, spotting edge cases, interpreting ambiguity and taking responsibility. SI can increase the amount of evidence they can inspect, but it can also create more candidate material than they have time to review.
The redesign should therefore reduce expert preparation burden and protect expert attention rather than flood experts with machine output.
Task Decomposition and Team Composition
When routine tasks shrink, teams may need fewer people doing the same production work and more people capable of exception handling, customer interaction, systems improvement and domain judgment.
This does not imply a universal headcount outcome. It means the role mix may change even when the team size does not.
Task Decomposition and Organisational Memory
Task cards should capture what experienced employees repeatedly know: source locations, definitions, acceptance criteria and common exceptions. This turns tacit practice into organisational memory.
Do not confuse task documentation with complete expertise. Some judgment remains difficult to formalise and should stay human.
Task Decomposition and Process Simplification
Before assigning SI, ask whether the task should exist. Repeated copying, duplicate approval and reformatting may be removed or integrated directly.
The best SI task is sometimes no task at all because process redesign eliminates the need.
Task Decomposition and Data Quality
A task that depends on unreliable data should not be treated as an SI problem first. Improve the source, validation or intake.
SI can help detect inconsistency, but it should not silently compensate for missing authoritative state.
Task Decomposition and Change Control
When a task changes significantly—new policy, new tool, new data source or new authority—review the task card. The old SI boundary may no longer fit.
Task decomposition is a living design artefact, not a one-time workshop output.
Comparing the Same Task Across Departments
The same verb can have different risk. “Classify” a marketing lead, “classify” a support ticket and “classify” an employee case may use similar machine capability but very different data, consequence and fairness requirements.
Task transfer should therefore compare context and authority, not only operation.
The Cross-Department Verb Library
- Retrieve
- Read
- Extract
- Summarise
- Classify
- Compare
- Calculate
- Explain
- Draft
- Transform
- Translate
- Plan
- Monitor
- Detect
- Recommend
- Approve
- Negotiate
- Escalate
- Execute
- Record
This shared vocabulary helps organisations find reusable SI patterns while preserving domain-specific boundaries.
The Task Decomposition Control Room
For a major SI programme, maintain a portfolio of candidate tasks. Each task can have a status: discovered, mapped, piloting, stable, automated, retired or human-only.
The portfolio prevents teams from repeatedly rediscovering the same opportunities and makes it easier to share evidence across departments.
The Task Evidence Pack
For each task promoted beyond experimentation, keep representative inputs, accepted outputs, known failures, current instructions, source requirements and evaluation results.
The pack becomes the evidence that a task boundary is understood well enough for wider use.
The Task Promotion Gate
Move a task from Assist to Collaborate or Automate only when the current method is stable, quality is measurable, exceptions are understood and the next level removes meaningful friction.
Promotion should not occur simply because the product added a new feature.
The Task Demotion Gate
Reduce autonomy when errors rise, sources become unreliable, review capacity collapses, the task gains higher consequence or the organisation loses the ability to verify output.
Demotion can mean returning to drafts, removing tool access or routing all cases through human approval.
The Task Retirement Gate
Retire an SI task when the value is too small, maintenance cost is too high, a deterministic solution replaces it or the underlying business process disappears.
Retirement should remove obsolete credentials, triggers and documentation so the old system does not remain an invisible dependency.
A Complete Task Card Example — Project Risk Summary
- Task: summarise project risks.
- Trigger: weekly reporting cycle.
- Input: current project updates and risk register.
- Output: top risks with evidence, owner and status.
- Receiver: project manager.
- Frequency: weekly.
- Context: milestones, previous risks and current dependencies.
- Standard: no unsupported claims; unresolved gaps visible.
- Verification: source links and manager review.
- Consequence: medium; missed risk can delay project.
- Reversibility: high before publication.
- Authority: recommend priorities, not approve project changes.
- SI role: Collaborate.
This level of detail is enough to design a pilot without turning the task card into a full technical specification.
A Complete Task Card Example — Refund Exception
- Task: decide whether a refund exception should be granted.
- Trigger: routine refund rule does not apply.
- Input: transaction history, customer context and policy.
- Output: authorised exception decision.
- Receiver: customer-service workflow.
- Frequency: occasional.
- Context: customer history, value, policy and prior exceptions.
- Standard: consistent with policy and commercial judgment.
- Verification: authorised manager review.
- Consequence: financial and relationship impact.
- Reversibility: moderate after payment.
- Authority: human manager.
- SI role: prepare evidence and recommendation only.
This example shows why the same task card can identify useful SI support while explicitly preserving human authority.
A Complete Task Card Example — Code Test Generation
- Task: generate unit tests for a bounded code change.
- Trigger: implementation branch exists.
- Input: specification and changed code.
- Output: candidate tests.
- Receiver: engineer and CI pipeline.
- Frequency: frequent.
- Context: codebase conventions and existing tests.
- Standard: valid tests covering required behaviour and edge cases.
- Verification: test execution and code review.
- Consequence: low-to-medium before merge.
- Reversibility: high.
- Authority: no production authority.
- SI role: Collaborate or bounded Agent.
Frequently Asked Questions About Task Decomposition
Why not start from the job title?
Job titles bundle many different operations and authority levels. Task-level analysis reveals which parts fit SI without making all-or-nothing claims about the role.
How many tasks should one job have?
Enough to capture meaningful recurring operations. Ten to thirty tasks is common for an initial inventory, though complex roles may have more. Group trivial micro-steps unless they matter for automation.
Should I time every task?
Approximate frequency and active time are usually enough for initial prioritisation. Use more precise measurement for high-value candidates.
Can SI help with physical jobs?
Yes. Physical roles often include planning, documentation, diagnosis, knowledge retrieval, scheduling and coordination that can benefit from SI even when the physical action remains human.
Can SI help with leadership?
Yes, primarily through evidence preparation, scenario comparison, briefings, monitoring and communication support. Purpose, alignment, trust and consequential decisions remain strongly human.
Should repetitive tasks always be automated first?
No. Repetition increases potential leverage, but risk, verifiability, bottleneck relevance and integration cost also matter. The next articles examine that trade-off.
What if the task is partly judgment and partly routine?
Split it. Routine preparation can often be automated while judgment remains human.
What if the task has no clear output?
Return to the work package or workflow outcome and ask what state the task is supposed to create. If nothing changes, the task may be poorly defined or unnecessary.
How do I know when a task is ready for automation?
The collaborative version is stable, normal and exceptional cases are understood, verification is efficient, permissions are scoped and the workflow can recover from failure.
What should I read next?
Continue to How to Find High-Leverage Super Intelligence Use Cases in Your Workplace. That article uses the task inventory to choose the opportunities most likely to create meaningful end-to-end value.
The Final Task-Decomposition Principle
Jobs are bundles; tasks are design units. When the organisation decomposes work far enough to see the operation, evidence, authority and receiver, it can place Super Intelligence precisely instead of making vague claims about replacing people.
The result is a more useful division of labour: machines handle scalable cognition where they fit, people retain responsibility where it matters, and the job itself evolves around the outcome rather than around old production habits.
The Task-Decomposition Return to the Workflow
After decomposition, return every task to the workflow map. A task may look attractive in isolation but create little value if it sits outside the bottleneck, depends on missing context or feeds a constrained downstream queue. This second pass prevents the task inventory from becoming a disconnected catalogue of automation ideas.
For each candidate, mark the state before it, the state after it and the receiver. Ask what changes if the task becomes faster, more accurate or more automated. If the answer is that nothing important changes downstream, lower its priority even if the task itself is technically easy.
The Task-Decomposition Return to Human Capability
Also return the selected tasks to the human role. Identify which capabilities become more important when routine production shrinks: verification, problem framing, customer interaction, negotiation, professional judgment, exception handling, systems thinking or teaching. The redesigned job should describe those responsibilities explicitly.
This matters because automation changes the shape of expertise. A role can become more valuable even as some tasks disappear, but only if the organisation develops the capabilities needed for the remaining work.
The Task-Decomposition Return to Organisational Learning
Finally, record what the organisation learned about the work itself. Task decomposition often exposes duplicated effort, missing knowledge, unclear authority and obsolete steps before SI is added. Those findings should improve the process even if the final decision is not to automate.
A good task-decomposition exercise therefore produces three outputs at once: a candidate SI portfolio, a clearer human role and a cleaner workflow. That combination is the reason task decomposition belongs before use-case prioritisation.
Task Decomposition for Roles That Are Changing Fast
Some roles change faster than job descriptions can be updated. Product managers, analysts, software engineers, educators, marketers and operations teams may add new SI-supported tasks while dropping old manual ones in the same quarter. In these roles, decomposition should be treated as a living operating exercise rather than an annual HR document.
Review the task bundle after a major tool rollout, model upgrade, process redesign or change in authority. Ask which tasks disappeared, which became easier, which now require more verification and which new exception or governance tasks appeared. This keeps the role description aligned with actual work.
Task Decomposition for Roles With High Accountability
Where the role carries legal, financial, safety, employment or professional accountability, decomposition should make the accountability-bearing task explicit. Do not hide it inside broad phrases such as “review” or “manage”. Name the actual decision: approve payment, sign advice, authorise release, decide exception or certify result.
This clarity allows SI to take on substantial preparation around the decision while keeping the authority boundary visible. It also makes the review step meaningful because the accountable person knows exactly what they are deciding.
Task Decomposition for Roles With High Relationship Value
Relationship-heavy roles often contain large amounts of invisible preparation. Account managers research clients, teachers prepare materials, managers gather performance evidence and consultants synthesise background before conversations. These preparatory tasks can be excellent SI candidates.
The redesign should use SI to protect relationship time rather than replace it automatically. If saved preparation time is converted into more customer contact, coaching or problem-solving, the human contribution becomes more concentrated rather than less important.
Task Decomposition for Roles With Heavy Exception Work
Some roles exist primarily because routine cases already flow through standard systems while people handle the exceptions. Examples include senior support staff, compliance specialists, operations managers and experienced engineers. In those roles, decomposition should focus on what makes an exception exceptional.
Identify the evidence, missing context, authority and decision pattern that cause the case to leave the routine path. SI can help assemble and structure the exception packet, but the role’s value may remain in resolving what cannot be standardised safely.
The Final Task-Portfolio Return
At the end of decomposition, do not produce a list of fifty possible automations. Produce a ranked task portfolio with three clear categories: tasks to test with SI now, tasks to keep human-led but SI-supported, and tasks to leave alone for the moment.
That portfolio creates a disciplined bridge into use-case selection. The next article will ask which tasks create the greatest leverage for the workplace as a whole, not simply which tasks are easiest to automate.
The Task-Decomposition Standard
A role is decomposed well when another person can see where information work ends, where judgment begins, where authority sits and what outcome each unit produces.
That standard is more useful than any claim about whether an occupation is “automatable”. It turns an abstract debate into a design map for real work.
