The Paradox of Designing Work That Disappears

What if the best system is the one that makes work vanish?

The dream of automation is usually told as a story about efficiency: faster output, fewer errors, lower cost. But there is a deeper and more unsettling question hiding underneath it: if machines become better at doing the work, what exactly is left for people to do?

That question matters because work is not only an economic function. It is also a site of meaning, identity, and social organization. The strange tension is that design often tries to reduce human effort, while a humane vision of work insists that labor should be creative, fulfilling, and self directed. These goals sound compatible until you ask what happens when systems are actually good at taking tasks away from humans.

The answer is not simply “more automation” or “less automation.” The real challenge is to design human AI systems that separate the parts of work that should disappear from the parts that should deepen. That requires thinking like a systems designer, not just a technologist. It also requires confronting a contradiction at the heart of modern production: we want to minimize labor time, yet we still measure value by labor time.

The central problem is not whether machines can do more. It is whether we can design work so that what remains is more human.

The hidden unit of design is not the tool, but the function

Most conversations about AI begin too late. They start with a tool, a model, or a product and then ask where it might be used. That approach invites design fixation, the tendency to get trapped by an existing solution and treat it as inevitable. A better approach starts one level deeper: with the function model of the human AI system.

A function model asks a simple but radical question: what must happen for the overall system to achieve its goal? In other words, before we decide whether a human, a machine, or both should do something, we first identify the functions required. This shift matters because it prevents us from confusing the current arrangement of work with the actual nature of the work.

Think of a hospital intake desk. The visible job may be “the receptionist checks patients in.” But the underlying functions could include verifying identity, prioritizing urgency, explaining procedures, collecting data, reducing anxiety, and routing people correctly. Once those functions are named, the design space opens up. Some functions may be better carried by a human voice, some by software, and some by a combination.

This is where the deeper connection to the future of work appears. If work is defined only as a block of time spent performing tasks, then automation looks like subtraction. But if work is understood as a constellation of functions, then automation becomes a question of recomposition. The point is not just to remove labor. The point is to reallocate functions so that the human role becomes more meaningful and the machine role more precise.

From labor time to function allocation: the real conflict

The classic contradiction is that capital tends to reduce labor time while still treating labor time as the source and measure of wealth. That contradiction is not just economic theory. It is visible in everyday AI deployment. Organizations want systems that are cheaper, faster, and more scalable, but they still organize jobs as if value comes from visible human effort.

This is why many automation efforts fail socially even when they succeed technically. They may optimize a narrow output while damaging the larger structure of work. A chatbot can shorten response times, but if it removes the part of customer service that builds trust, the organization may save seconds while losing loyalty. A radiology AI can flag anomalies, but if it collapses the clinician’s interpretive role into rubber stamping, the job becomes less intellectually satisfying and potentially less safe.

The deepest issue is not replacement, but division of cognition. Who notices, who decides, who explains, who escalates, who reassures, who learns? These are distinct functions. A good design does not ask whether AI should take over everything. It asks how to divide the cognitive and emotional labor so that each actor does what it does best.

That is why a function model matters. It reveals work as a system of interdependent responsibilities rather than a single indivisible occupation. Once those responsibilities are visible, the real design question emerges: which functions should be carried by humans, which by machines, and which by the interaction between them?

Designing the collaboration, not just the automation

A useful way to think about human AI systems is through function carriers. Once functions are identified, the designer maps each one to possible carriers: a human, a machine, a workflow rule, a interface element, a sensor, a policy, or some combination. A morphological chart helps by laying out candidate carriers for each function and allowing different combinations to be explored systematically.

This matters because there is rarely one correct design. There are many possible architectures, each with different tradeoffs. One version may maximize accuracy but reduce transparency. Another may increase user trust but slow down throughput. Another may be cheaper but more brittle under unusual conditions. The temptation is to let the highest score decide, but that is too shallow. The better move is to build a design narrative that explains why a particular combination of carriers is chosen over others.

That narrative is important because work systems are not only technical artifacts. They are social contracts. If a medical triage system assigns urgency to an algorithm, the rationale must address not only prediction accuracy but also accountability, explainability, and the lived experience of patients and clinicians. If a hiring system filters applicants, the design must justify how it balances efficiency with fairness and human judgment.

Here is the surprising insight: automation is never just about automation. Every automated function changes the role of the human around it. Sometimes the human becomes supervisor, sometimes operator, sometimes interpreter, sometimes educator, sometimes the last line of defense. Designing the system therefore means designing the human role as deliberately as the machine role.

A concrete example: the airport security checkpoint

Take airport security. If the only design goal were throughput, much of the checkpoint could be automated. But the real system has multiple functions: identity verification, threat detection, crowd management, reassurance, exception handling, and compliance signaling. A fully automated version might be fast in routine cases but collapse when the situation becomes ambiguous, tense, or emotionally charged.

A better design assigns functions strategically. Machines can scan baggage and detect patterns. Humans can handle anomalies, interpret context, and deescalate conflict. The interface can support both by surfacing risk in a way that is actionable rather than cryptic. The result is not simply a faster checkpoint. It is a more resilient one.

This is the broader lesson for AI in work. The best systems are not the ones that eliminate the human most aggressively. They are the ones that make the human contribution more precisely necessary.

The overlooked role of parameters: what can be controlled, and what cannot

Once a function model is established, the next step is to parameterise it. This means identifying which variables are controllable and which are not. A controllable parameter is something the designer can tune, such as interface layout, alert thresholds, or response timing. An uncontrollable parameter is something that affects the outcome but cannot be directly set, such as user stress, changing regulations, or the quality of incoming data.

This distinction seems technical, but it is philosophically important. Many failures in work design come from treating uncontrollable realities as if they were controllable by decree. Leaders assume that if they change a dashboard or add an AI assistant, the human behavior will adjust cleanly. It rarely does. People improvise, resist, misunderstand, and adapt in ways that the system designer did not predict.

Parameterising the function model forces humility. It says: here is what we can shape, here is what we can observe, and here is what remains outside our control. That honesty is crucial for designing work that survives contact with reality.

Imagine an AI writing assistant used in a legal team. A controllable parameter might be how aggressively the system suggests phrasing. An uncontrollable parameter might be whether the lawyer is under deadline pressure or whether the source documents are ambiguous. If the design ignores the second factor, the tool may look elegant in demos and fail in practice. If the design accounts for it, the system can adapt, for example by changing confidence displays, requiring human review at uncertainty thresholds, or routing complex drafts to senior staff.

This is where human work becomes more than execution. The human is not just a backup for machine error. The human is often the interpreter of context, the carrier of values, and the adaptive agent in a world of uncontrollable variation.

A new model for humane automation: remove toil, deepen judgment

The deepest synthesis here is that the future of work should not be framed as jobs versus machines. That framing is too blunt. A better framing is toil versus judgment.

Toil is repetitive, brittle, low meaning activity that consumes time without developing capability. Judgment is the interpretive, relational, and value laden part of work that requires context. Good AI design should remove toil where possible, but preserve and deepen judgment where it matters. In that sense, automation should be treated as a tool for redistributing dignity.

This is a more demanding standard than efficiency. It asks not only whether a system works, but whether it improves the human experience of work. Does it reduce mindless repetition without stripping away mastery? Does it make expertise more visible instead of less? Does it support people in making better decisions, or does it turn them into passive approvers? These questions reveal the moral stakes of design.

Consider two versions of the same task, approving insurance claims. In the first, an AI system approves or denies claims automatically, and humans only audit edge cases. In the second, the AI pre-sorts claims, highlights missing information, explains why a claim looks risky, and routes hard cases to human reviewers with context. Both systems automate. Only one preserves human judgment as a real function rather than a ceremonial one.

The difference is not subtle. The first design aims to replace labor. The second aims to redesign labor so that the remaining human work is more informed, more responsible, and more worthy of a skilled person’s attention.

Key Takeaways

1. Start with functions, not tools. Before choosing AI features, map the functions the system must perform. This prevents design fixation and opens a wider solution space.

2. Design the human role as carefully as the machine role. Every automated function changes what people do. Make sure the remaining human tasks involve meaningful judgment, not just error cleanup.

3. Separate toil from judgment. Use automation to remove repetitive, low value work, but protect the interpretive, relational, and ethical parts of the job.

4. Use controllable and uncontrollable parameters honestly. Identify what can actually be tuned and what cannot. Design for uncertainty instead of pretending it does not exist.

5. Write the design narrative, not just the score. When evaluating competing concepts, do not let numerical scores decide everything. Explain the tradeoffs and why the chosen system deserves to exist.

---

The future of work is not less human, but more deliberately human

The seductive promise of AI is that it will make work disappear. The better promise is harder to sell: it will force us to decide what in work should never have been treated as expendable in the first place.

That changes the meaning of progress. Progress is not a world where humans do nothing. It is a world where machines absorb what is routine, while humans retain what is judgmental, creative, and socially significant. Designing such a world requires more than technical competence. It requires a philosophy of work.

The real test of a human AI system is not whether it can replace a person in a task. It is whether it can make the overall system more worthy of human participation. That is the paradox at the center of modern automation: the best design may be the one that removes work, but only so that the work that remains can finally be called meaningful.