Research program
Unbundling Work studies tasks, decisions, handoffs, authority, evidence, risk, and outcomes instead of treating the job title as the unit of analysis. The aim is to preserve scarce human judgment and accountability, encode work that does not require them, and recompose the operation around the best mix of people, AI, software, and machines.
Which tasks, decisions and handoffs need scarce human judgment and accountability, and which can be encoded, so an operation can be rebuilt around the best mix of people, AI, software and machines?
Claim: recommendation
Most plans to change work begin with the job title. That is too blunt. A job bundles tasks, decisions, authority, credentials, evidence, handoffs, exceptions, relationships, physical acts, and liability. The bundle may reflect law or sound practice. It may also reflect old software, payment rules, labor scarcity, organizational boundaries, or habit.
Unbundling Work studies the work itself before deciding what should remain human, move to a machine, be shared, be redesigned, or disappear.
Claim: recommendation
The useful unit has a purpose, trigger, inputs, source priority, decisions, actions, outputs, authority, credential need, risk, handoffs, systems, time, failure modes, learning value, rights, and an accepted outcome.
A description is not enough. The record must be checked against observed work, source events, exceptions, and the people who receive the result.
Claim: recommendation
A team that automates a bundled role carries old assumptions into the new operation. It may remove a visible task while adding review, escalation, integration, or patient burden somewhere else.
First ask why the work exists and who imposes, performs, pays for, benefits from, waits for, and can change it. The answer may be to retain, assist, automate, redesign, renegotiate, or remove the work.
Claim: recommendation
The finished unit is an AI Workforce™ made of people, AI workers, deterministic rules, applications, source systems, data stores, devices, outside services, and governance. Each carries work according to judgment, authority, trust, repeatability, observability, reversibility, physical need, and consequence.
The test is the accepted outcome, not the number of automated tasks.
Claim: recommendation
Native Alpha™ is the judgment, trust, knowledge, access, relationship, data, rights, workflow position, or execution that changes the result and is not easily reproduced. For each human activity, the research asks whether to preserve it, amplify it, encode it, or reject the claim that it is special.
It also asks what routine work forms future judgment. Removing today's burden must not remove tomorrow's apprenticeship.
Claim: observation
Healthcare makes weak assumptions visible. Credentials, professional scope, reimbursement, privacy, fragmented records, physical care, trust, liability, and patient and caregiver labor all shape the work.
Studies follow work across clinicians, staff, payers, vendors, patients, caregivers, and systems. A saving is not counted when the burden simply moves.
Claim: recommendation
Operational Truth™ separates intended, reported, recorded, inferred, observed, and verified state. Outcomes Acceptance Testing asks whether the real result occurred within stated safety, time, cost, and evidence limits.
Authority rises only after measured performance and falls when a model, source, policy, population, dependency, or incident changes the risk.
Claim: recommendation
Why does the work exist and whose incentive keeps it there? How will people move and continue to learn? Can one shared fault harm many outcomes at once? What evidence supports cause rather than correlation? Who owns the data, know how, worker contracts, tests, and operating record?
A design that cannot answer these questions is not ready to operate.
Program
Native Alpha™ is the unusual capability of a person or organisation that materially changes an accepted outcome, and that ordinary peers with ordinary tools and AI cannot reproduce. Recomposition is dividing work again so that capability is spent where it matters. The goal is a working operation in which people, machines, software, and outside services each carry the work they can perform well and lawfully.
Program
We hand capable engines important work and then hand them a thin description of the people, companies, customers, competitors, products, and systems the work is about. An Eidolon is a working digital reflection of one of those things: not a replica, not a clone, but enough grounded reality for an AI to reason about the original without guessing.
Program
Labor as Code™ is the practice of writing a unit of work down (its purpose, evidence, authority, limits, exceptions, and tests) precisely enough that it can be assigned to a person or a machine and then inspected and changed. Work becomes governable when its purpose, evidence, authority, limits, exceptions, and tests are written clearly enough to inspect and change.
Program
This program asks how an organization can know that its mixed workforce acts on reliable evidence and produces acceptable outcomes. Consequential work needs evidence of what happened, clear limits on authority, known failure modes, and a test tied to the result in the world. It matters because a completed field can be well formed and still be wrong.
Program
A job title bundles work, authority, trust, evidence, and liability. Open the bundle before deciding what should move and what must stay with a person.
Specialized SaaS won largely because configuring general-purpose platforms was expensive. If AI collapses that cost, the reason for the historical result stops applying. This note sets out that hypothesis, states what is being measured to test it, and names the parts that stay hard.
Represent trust as evidence about a narrow capability, approved sources, authority, incidents, and recertification. An AI credential is a task specific operating record that can be narrowed, suspended, or renewed as evidence changes. It is not a professional license or legal credential.
This note sets a rule for granting and withdrawing a worker's authority: evidence earns more authority, and a material change in model, source, policy, population, workflow or incident cuts it until retesting is complete. It matters because strong performance alone does not justify wider authority. Increase authority only after evidence and reduce it automatically when material conditions change.
Show who requires work, who performs it, who pays, who benefits, who waits, and who can change it. The matrix exposes cost shifting and the party able to remove the cause, but it does not settle the policy choice, because different parties may have sound reasons for the same work.
Identify which repeated work creates future judgment and replace that learning path before automation removes it. This map matters because automating routine work can strip future workers of the cases and feedback they need to become competent. It preserves the learning, not the burden.
Choose enough comparison to tell whether a recomposed workflow caused the claimed change. The plan matches the study design to the claim and narrows the claim when the design cannot distinguish the intervention from staffing, case mix, policy, or season, because operational acceptance and causal effect answer different questions.
Mark every competitor claim as observed, disclosed, reported, or inferred, so analysis built on it cannot quietly promote a guess. Labelled claims let an engine say when a recommendation rests on inference rather than evidence, and contradictions between credible sources are kept as findings.
This note gives a way to measure how much credentialed time, authority and liability a workflow uses per accepted outcome, before and after a redesign. Lower exposure is not automatically better, so the redesign stops when accepted outcomes, responsibility, trust or capability formation worsen. Count where credentialed time, authority, supervision, and liability enter one accepted outcome.
Training, credential, licence, supervision, and liability cover exist as a stack because each layer is insufficient alone. An AI worker arrives with none of them.
Enough reflection of an account that anyone can prepare before acting, bounded by confidentiality terms, privacy obligations, and contractual restrictions. The customer Eidolon is useful only if restricted material is decided and held separately before writing, because a reflection that keeps material the company had no right to retain is a liability.
Most assembled systems fail before any model runs, because the data they depend on is incomplete, inconsistent, or somewhere nobody checked. Before a workflow is a candidate for automation, answer four questions in writing: availability, quality, gaps, and alignment. Repair the data substrate first.
This note finds that case accuracy does not predict population harm from a shared fault, so scaling should stop when one untested dependency can affect more outcomes than the organization can detect, contain and recover. The map gives the concentration and blast radius thresholds to check. Show which workers share models, sources, policies, tools, identities, vendors, and people so one fault cannot spread unseen.
Keep every consequential operational claim connected to its source, custodian, transformations, and useful life. The result is an evidence chain a reviewer can follow in both directions, from the real event to every derived field and decision. Provenance shows where a claim came from, not that the source is correct.
Businesses do not need CRM software; they need work performed. Decompose the outcome into roles, tasks, workflow, and capabilities, and the application stops being the organizing unit.
Automation follows understanding. Work becomes explicit, then observable, then repeatable, and only then is it a candidate for AI execution with human exception handling. Skipping a stage does not speed the sequence; it moves the failure later, where it costs more.
Define how consequential work continues or stops when AI, data, tools, identity, or outside services are unavailable or untrusted. The plan sets modes from normal operation to safe stop, and a fallback counts only once it has been exercised. Deployment stops when the operation cannot fail safely.
Make work inspectable, assignable, testable, versioned, and governable before attempting broad automation. The pattern turns observed work into Work Units and Worker Contracts that can be tested in shadow mode and versioned, and it stops when specification adds more burden than clarity.
Separate who creates, controls, authorizes, owns, transfers, and finances risk at each stage. The map gives a stage by stage view of risk and responsibility with unresolved questions, and a recomposition is rejected when no party clearly owns a consequential failure. Technical architecture does not establish legal liability.
The smallest institutional reflection that changes AI output: what the company does, how it is organised, who decides, what it sells, and what it will not claim.
Find where a person's or organization's unusual capability materially changes an accepted outcome. The map tests whether the difference survives comparison with peers, tools, and AI, so the firm rejects the claim when ordinary peers reproduce the result. Native Alpha™ is a hypothesis, not praise.
This note gives a review for judging whether a redesign of work raised Native Alpha™ density and improved the outcome. It is a qualified view, not a universal numerical score, and the redesign is rejected when density rises on paper but outcomes, learning, trust or resilience worsen. Test whether scarce capacity is moving toward work where unusual capability changes or compounds the result.
Trace an operational claim to independent evidence and distinguish intended, reported, recorded, inferred, observed, and verified state. The output is a queryable evidence chain and the gap between reported and observed state. A complete record can still be false, and Operational Truth™ does not prove causality.
Test whether the intended real world result occurred within stated quality, safety, time, cost, and evidence limits. Acceptance is a continuous record tied to worker and workflow versions, and authority is reduced or stopped when thresholds fail. Passing an outcome test does not prove the worker caused the outcome.
Start by preparing evidence for a person, then grant narrow action only after review data supports it. Authority is promoted one level at a time, from observing to independent execution, using error, edit, escalation, and incident evidence. Some work should never progress beyond advice or packet preparation.
This note gives a before and after map of the unpaid work patients and caregivers do, so it counts in the economic and human result. A workflow improvement that merely exports more work to patients or caregivers than it removes is rejected. Count unpaid calls, forms, portals, travel, record gathering, repeated explanations, and home work around care.
When one reflection links to another, decide what a reader inherits, what it must request, and what it may never see. Permission has to travel with the link, or a worker cleared for one Eidolon can read every executive, customer, and system Eidolon it references.
Keep a role specification and a person's Eidolon as separate linked objects so work is not written for an office as though it were a personality.
Choose how to handle a claimed differentiated capability after testing what makes it valuable. The four choices are preserve, amplify, encode, or reject, and they can coexist across parts of one capability. Encoding stops when it weakens trust, learning, control, portability, or the advantage itself.
Judge a reflection by the work it improves. Add one fact, rerun the task, and keep the fact only if the accepted outcome gets better.
Facts age at different speeds. Date them, give the fast ones a review interval, and treat an unreviewed reflection as an unknown rather than a truth.
The realistic outcome is not fewer credentialed people. It is credentialed people supervising a distributed set of workers, human and otherwise. The role is recomposed, not deleted, and how many workers one person can oversee is the unknown that decides whether the economics work.
Track ownership and permitted use of know how, data, worker contracts, prompts, tests, evidence, and operating records. The resulting rights record can block release when a material asset is unresolved, and it supports qualified legal review without deciding ownership or enforceability.
A role name hides the scope, authority, evidence, exceptions, and outcomes that determine whether work can move. Comparing the formal description with actual tasks and decisions yields a role decomposition into distinct work units. A title can stay useful for employment, identity, and accountability even when it is a poor automation unit.
State what a worker may do, what evidence it must use, when a person reviews it, and how it fails safely. The contract is versioned so it survives a change of model or vendor. It governs the worker but grants no legal authority the organization lacks, and deployment stops when ownership or failure behavior cannot be stated.
Move from the current operation through shadow work and narrow authority while keeping rollback available. The plan stages the move with gates, costs, owners, and fallback, and reports time to net positive value, because steady state savings do not erase transition cost or temporary risk.
The conditions under which specialized software remains the right answer, stated precisely enough to be used in an actual review. Where requirements are certified, regulated, or deep in domain logic, Zero Neo has no claim and the hypothesis is treated as rejected. New software is the last resort, and sometimes that is where a case ends up.
Find whether work serves the outcome or was created by a rule, incentive, boundary, or information failure. Each burden is classified and given a disposition of retain, assist, automate, redesign, renegotiate, or remove. Institutional friction can be protective, so classification does not prove a burden is waste.
This note gives a versioned record that another person can inspect and test, so a piece of work is described by evidence and not by intended process. A description of intended process is not evidence of actual work, and the unit should be neither too broad to observe nor too narrow to own a result. Describe one piece of work by its purpose, evidence, authority, risk, handoffs, and accepted outcome.
Assign work across people, AI, rules, software, systems, devices, services, and governance as one operation. The map gives every Work Unit an owner and every handoff in writing, and it stops when a task saving raises end to end cost, delay, risk, or review burden.