In the Loop. On the Loop. Out of the Loop.
This month three things caught my attention: the “cyborg workplace” argument, the contrasting Twitter essay warning that nothing done on a screen is safe from AI substitution, and the launch of Open Claw, which I will discuss separately. Here, I focus on the first two.
The first is “The cyborg workplace” published in The Economist (January 31, 2026) and the second is Matt Shumer’s widely shared post, “Something Big Is Happening” (X, 2026), which argues that in the medium term “nothing that can be done on a computer is safe.”
Both pieces are analytically serious, both deserve attention, yet they point in different directions. I offer a framing that reconciles them.
I. Task Recomposition Rather Than Occupational Extinction
The Economist article advances a historically grounded thesis: artificial intelligence is reducing the cost of specific cognitive activities without eliminating entire professions. Drawing on U.S. employment data since 2022, it highlights three empirical patterns:
Aggregate white-collar employment has increased rather than collapsed: roughly 3 million white-collar jobs have been added since late 2022, while blue-collar employment has remained flat.
Real wages in professional and business services have risen by around 5%, and the white-collar wage premium now stands at approximately one-third above blue-collar earnings - nearly triple the premium recorded in the early 1980s.
Growth has concentrated in roles combining technical expertise with oversight and coordination: project managers and information-security experts up around 30%, business operations specialists up nearly 60%, mathematical science occupations up around 40% with real wages rising by roughly a fifth. Routine back-office functions have moved in the opposite direction: insurance-claims clerks down 13%, secretaries and administrative assistants down 20%.
The underlying mechanism is consistent with established labour economics. Occupations are bundles of tasks, and technologies typically substitute for routine, codifiable tasks while complementing non-routine analytical and interpersonal tasks.
AI appears to follow this pattern: it drafts, codes, summarises and processes at lower marginal cost. Tasks involving discretion, contextual judgment, coordination and accountability, however, remain structurally embedded in human roles.
The key analytical claim, therefore, is not that AI leaves white-collar work untouched, it is that it recomposes work at the task level. Substitution and complementarity coexist within the same occupation, often within the same role.
II. The Automation Frontier and the Screen-Based Warning
Matt Shumer’s “Something Big Is Happening” advances a more expansive proposition: if the core of one’s job consists of reading, writing, analysing or communicating through a keyboard, AI will absorb significant parts of it. Eventually, robotics will extend this automation frontier into physical labour.
This argument is directionally correct. AI systems have expanded the domain of tasks that can be automated beyond routine clerical processing into higher-order cognitive activities. The boundary between “routine” and “non-routine” work is shifting, and that shift is not trivial.
However, two clarifications are necessary.
First, AI performance remains uneven. It demonstrates what researchers describe as jagged intelligence: high competence in some domains and brittleness in edge cases. A useful indicator of current scale: only around 4% of occupations use AI across three-quarters or more of their tasks. In many professional contexts, the fraction of tasks involving discretion, liability and ethical responsibility cannot be treated as residual. A system that performs 95% of a task reliably may still require human oversight if the remaining 5% carries asymmetric risk.
Second, economic impact depends not only on technical feasibility but on organisational design. Even when a task can be automated, firms may choose hybrid arrangements to manage risk, compliance and reputational exposure. The presence of a screen does not determine the absence of human responsibility.
The automation frontier is real, but its translation into wholesale occupational displacement is mediated by institutional and organisational constraints that technical capability alone does not dissolve.
III. Workflow Architecture and the Position of the Loop
The more precise question is not whether AI will transform work. It is where human judgment sits inside the workflow. Three configurations define the structural options.
In a human-in-the-loop configuration, a human defines the problem, validates outputs, exercises judgment and carries accountability for decisions. AI augments productivity but does not substitute human responsibility. The workflow is augmented, not automated.
In a human-on-the-loop configuration, the human designs, supervises and orchestrates systems, managing AI agents, workflows and interfaces. Value lies in coordination, integration and governance. AI executes; humans audit and intervene when necessary.
In a human-out-of-the-loop configuration, AI operates autonomously within tightly defined boundaries. Human involvement occurs before the process begins, in system design and rule-setting, and after it concludes, in review and compliance, but not at each transaction. AI executes within limits that humans define. This is where the exposure to substitution is most direct.
Empirical patterns suggest that roles characterised by expertise, oversight and coordination are expanding, while roles dominated by routine execution are contracting.
For individuals, this implies three strategic imperatives:
Increase discretionary content in your role Seek responsibilities involving judgment, ambiguity and decision rights rather than pure execution.
Develop orchestration capabilities Learn to work with AI systems, manage outputs, and integrate them into broader organisational processes.
Invest in transferability Analytical reasoning, systems thinking and interpersonal competence remain complements to intelligent systems.
Empirical patterns suggest that roles characterised by expertise, oversight and coordination are expanding, while roles dominated by routine execution are contracting. The debate between expansion and extinction is therefore incomplete. The more consequential distinction is architectural: where, in a given workflow, is human judgment positioned, and what does that imply for accountability, risk and governance?
That question is worth examining carefully, not only at the level of individual roles, but in how organisations are actively choosing to design the workflows through which consequential work now flows.