More Output, Less Judgement: Parkinson's Law and the AI Productivity Trap

Dr Danie Adendorff DSc (c.h), MSc

Artificial intelligence has not abolished bureaucracy. It has given bureaucracy a faster production engine.

That is the uncomfortable point missing from much of the AI productivity debate. The discussion still concentrates on speed, automation, efficiency, labour savings, and output. Those measures matter, but they are incomplete. A business does not exist to produce more internal material. It exists to serve customers, execute strategy, manage risk, protect value, and fulfil its mission. When AI increases internal output without strengthening judgement, the organisation becomes more productive in form and weaker in substance.

This article follows directly from the argument developed in "Jevons Paradox and the AI Age." Jevons explains the output surge. When the cost of producing cognitive work falls, the volume of cognitive work rises. AI reduces the cost of drafting, summarising, analysing, reporting, briefing, monitoring, and coordinating. Organisations then produce more of all of it.

Parkinson explains what happens next. Organisations absorb the extra output and turn it into bureaucracy. Reports create reviews. Reviews create meetings. Meetings create summaries. Summaries create action registers. Action registers create dashboards. Dashboards create escalation loops. Internal production becomes self-feeding.

Staff reduction completes the failure chain. When experienced people are removed because AI appears to increase productivity, the organisation loses the judgement required to distinguish useful work from administrative noise. The result is the AI productivity trap: more output, less judgement, and a business that becomes more efficient at moving away from its own mission.

Parkinson's Law Before AI

C. Northcote Parkinson's famous law is usually compressed into one sentence: work expands to fill the time available for its completion. That summary is useful, but it understates the force of the original argument. Parkinson was not merely describing procrastination. He was describing administrative expansion.

His British civil-service examples, including the Colonial Office, made a sharper point: administration expands even when the external task contracts. The real-world mission shrinks while the internal apparatus grows. Staff numbers, files, committees, supervisory layers, correspondence, and internal coordination develop their own logic. The bureaucracy no longer mirrors the mission. It feeds itself.

That principle applies far beyond government. A business loses market focus but increases reporting. A company misses operational targets but creates more governance forums. A department fails to serve customers but improves its internal documentation. A leadership team postpones hard decisions while producing strategy packs, risk registers, meeting notes, and performance dashboards.

At that point, the organisation has confused activity with effectiveness. It looks managed. It looks controlled. It looks busy. It is drifting.

Parkinson's Law is therefore not a joke about slow work. It is a warning about organisational self-expansion. Work expands because structures, incentives, and habits allow it to expand. Bureaucracy protects itself by generating further work. Administrative systems turn motion into apparent legitimacy.

AI does not remove that tendency. It strengthens it when executives fail to impose mission discipline.

AI Changes the Law's Operating Environment

Before AI, internal output had friction. A report required a person to write it. A briefing pack required a team to assemble it. A transcript required someone to produce it. A policy note, compliance summary, risk review, or customer analysis consumed visible labour.

That friction acted as a crude control. It did not prevent bureaucracy, but it limited its speed.

AI removes much of that friction. A meeting now produces a transcript, summary, task list, decision note, follow-up email, and status update. A minor issue becomes a briefing paper. A preliminary thought becomes a polished proposal. A routine dataset becomes a dashboard. A customer interaction becomes an analysed theme. A compliance question becomes a structured memorandum.

The material looks professional. That is the danger. AI gives weak work the surface appearance of strong work. It formats uncertainty. It polishes fragments. It turns undeveloped thinking into presentable prose. It lowers the visible cost of organisational production and raises the risk that leaders mistake generated material for managerial progress.

This is no longer a theoretical warning. The BetterUp Labs and Stanford Social Media Lab workslop research surveyed 1,150 U.S. full-time desk workers in September 2025. It reported that 40 percent had received AI-generated workslop in the previous month, that workers estimated 15.4 percent of received work fell into that category, and that each incident took nearly two hours to resolve. Treat this as reported workplace evidence, not as hard operational telemetry. Its value lies in the mechanism it exposes: weak work travels further because it looks finished, and the labour of discovering the weakness moves downstream.

The earlier Jevons argument explains the first-order effect: cheaper cognitive production increases cognitive consumption. Parkinson explains the second-order effect: the organisation builds administrative processes around the increased output. The two principles combine. AI lowers the cost of producing internal work, and bureaucracy expands to consume it.

The result is not always better productivity. In this failure pattern, it is administrative inflation.

The Staff-Reduction Trap

The most dangerous version of this pattern appears when AI output is combined with staff reduction.

Executives see the machine producing summaries, reports, responses, draft policies, decision notes, and customer communications. They conclude that fewer people are needed. Some reductions are rational. Repetitive work, duplicated reporting, and routine drafting do not deserve protection merely because humans currently perform them.

The evidence is not anti-AI. Brynjolfsson, Li, and Raymond's study of 5,172 customer-support agents found a 15 percent average productivity gain after access to AI assistance, with the largest improvements among less experienced and lower-skilled workers. The same study is also a warning against crude replacement logic: the most experienced and highest-skilled workers showed small speed gains but small quality declines. Speed and quality did not move together uniformly.

That finding matters because experienced staff do more than produce documents. They carry memory, judgement, dissent, context, and informal warning systems. They know when a supplier explanation is weak. They know when a customer complaint signals a wider defect. They know which performance metric is technically correct but operationally misleading. They know the history behind failed initiatives. They know when a polished briefing hides an unresolved problem.

Holweg and Davenport's 2026 Harvard Business Review article makes the organisational version of the same warning. They describe AI slop as a driver of decay in the accuracy and quality of organisational knowledge. That is exactly the level at which the staff-reduction trap becomes dangerous. The scarce resource is no longer the ability to produce another paragraph, slide, or dashboard. The scarce resource is knowing what to trust, what to challenge, what to discard, and what to escalate.

When staff reduction removes the people who understand the mission, the organisation retains production while losing interpretation. It has more documents and fewer people capable of judging which documents matter. It has more dashboards and less operational sense. It has more summaries and less memory. It has faster communication and thinner accountability.

This is not transformation. It is cognitive disarmament disguised as efficiency.

More Output Is Not Mission Performance

The central management error is the confusion between output and mission performance.

Output is the material produced by the organisation: reports, documents, dashboards, emails, analyses, policies, plans, and summaries. Productivity measures output against input. Decision quality asks whether the output improves judgement under uncertainty. Mission performance asks whether the organisation achieves the purpose for which it exists.

AI improves the first measure easily, and the second in many bounded tasks. It improves the third and fourth only when human judgement, validation, and accountability remain strong, and only when the output is deliberately tied to mission outcomes rather than produced for its own sake.

A business that produces more internal material is not automatically better governed. A leadership team that receives more dashboards is not automatically better informed. A compliance function that generates more documentation is not automatically reducing risk. A customer-service department that produces faster responses is not automatically solving customer problems.

The distinction matters because AI makes the wrong thing easier to scale. If the organisation rewards internal output, AI increases internal output. If it rewards documentation, AI increases documentation. If it rewards apparent control, AI manufactures the artefacts of control. The mission then becomes secondary to the machinery around the mission.

Microsoft's 2023 Work Trend Index supports the wider context. It described digital debt as the inflow of data, emails, meetings, and notifications exceeding human processing capacity. Its survey reported that 64 percent of workers struggled to find the time and energy to do their job, and that those workers were 3.5 times more likely to struggle with innovation and strategic thinking. AI inserted into that environment without deletion discipline does not solve overload. It feeds overload in a more polished form.

How the Failure Develops

The failure does not begin with collapse. It begins with reasonable language. The pattern below is an analytical failure model, not a claim about one named company.

Leadership announces an AI productivity initiative aimed at efficiency, speed, and focus. Tools are deployed across communication, reporting, research, customer service, compliance, and internal knowledge management. Early results look promising: people produce faster, documents look sharper, routine drafting becomes easier, and managers see visible movement.

The organisation then starts reducing, freezing, or redeploying staff, on the reasonable-sounding logic that AI has increased capacity and the labour structure should adjust. What rarely happens at this stage is any disciplined mapping of which human capabilities are being removed. Roles are counted; judgement is not. Cost is tracked; memory is not. Output is measured; mission risk is not.

AI-generated material then multiplies on its own momentum, largely unchecked, because nothing in the previous stage created a mechanism to notice that this was happening. More summaries circulate, more dashboards appear, more reports get commissioned because they are now cheap to produce, and more meetings generate structured follow-up that itself demands review.

Coordination expands around this new volume. People reconcile conflicting AI summaries. Managers respond to generated action lists. Compliance teams validate drafts. Executives request further versions. Internal communication accelerates. Accountable decision-making does not accelerate with it.

The final stage is strategic misrecognition. Leaders mistake volume for control, reading the abundance of internal material as evidence of better management. In reality, the organisation has become more administratively active and less mission-centred - busier, not sharper.

That is how business mission failure develops in the AI era: not through a single dramatic error, but through the steady replacement of judgement by output.

The Executive Control Test

AI requires executive discipline, not passive adoption. The central question is not "How much can we produce?" It is "What must this output improve?"

Leaders should apply a strict control test before expanding AI-generated work:

· What mission does this output serve?

· Which decision does it improve?

· Who is accountable for interpreting it?

· What human judgement must remain in the loop?

· What obsolete work will be deleted?

· Which meetings, reports, dashboards, or approval layers will disappear?

· Are we reducing bureaucracy or automating it?

The deletion question is decisive. If AI is introduced and no old work disappears, Parkinson's Law remains active. The organisation has not created efficiency. It has created additional capacity for administrative expansion.

Every AI deployment should therefore carry a removal plan: which report stops, which meeting ends, which approval layer disappears, which duplicated summary is no longer produced, which metric is retired, and which process is simplified.

McKinsey's 2025 State of AI survey reinforces the same control principle from the enterprise-adoption side. Its high-performing organisations are nearly three times as likely as others to redesign workflows fundamentally, and they define when model outputs require human validation to ensure accuracy. That is the correct test. Value comes from redesigning work, not from layering AI output on top of existing bureaucracy.

The same discipline applies to staff reduction. If roles are removed, leaders must identify which judgement functions those roles performed. Who carried operational memory? Who challenged weak assumptions? Who understood customers? Who knew the regulatory history? Who detected nonsense before it reached senior management?

Cost reduction without judgement mapping is not modernisation. It is organisational amnesia.

Conclusion: The Productivity Trap

AI accelerates Parkinson's Law when executives add machine-generated output without removing obsolete work. It reduces the cost of internal production, and organisations expand to consume that production. This is the modern form of the old bureaucratic disease.

Jevons explains why AI increases output. Parkinson explains how organisations turn that output into bureaucracy. Staff reduction explains why judgement collapses. Mission failure is the consequence.

The remedy is not anti-AI sentiment. AI is useful when it removes waste, improves decision preparation, supports human expertise, and strengthens mission execution. It becomes dangerous when leaders treat output as evidence of value.

The business test is simple. Are customers better served? Are decisions sharper? Is risk understood earlier? Is execution stronger? Is unnecessary work disappearing? Is human judgement preserved where consequence demands it?

If the answer is no, the organisation is not becoming more intelligent. It is becoming more productive at producing its own administrative burden.

That is the AI productivity trap: more documents, more dashboards, more summaries, more apparent control - and less judgement where judgement matters most.

References

Adendorff, D. (05/26/2026) ‘Jevons Paradox and the AI Age’, Rumbls.com. .

BetterUp Labs and Stanford Social Media Lab (2025) ‘Workslop: The Hidden Cost of AI-Generated Busywork’. September 2025 survey of U.S. full-time desk workers. .

Brynjolfsson, E., Li, D. and Raymond, L.R. (2025) ‘Generative AI at Work’, The Quarterly Journal of Economics, 140(2), pp. 889–942. doi: 10.1093/qje/qjae044.

Dell'Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K.C., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K.R. (2026) 'Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality', Organization Science, 37(2), pp. 403–423. https://doi.org/10.1287/orsc.2025.21838

Holweg, M. and Davenport, T.H. (2026) ‘Don’t Let AI Slop Muck Up Your Company’s Processes’, Harvard Business Review, 16 June. Used for the organisational knowledge-decay argument.

Jevons, W.S. (1865) The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. London: Macmillan. Used for the original efficiency-consumption logic later associated with Jevons Paradox.

McKinsey & Company (2025) ‘The State of AI: Global Survey 2025’. Used for the workflow-redesign and human-validation governance evidence.

Microsoft (2023) ‘2023 Work Trend Index: Annual Report — Will AI Fix Work?’, 9 May. Used for the digital-debt context, communication overload, and the 64 per cent / 3.5x innovation-struggle finding.

Niederhoffer, K., Kellerman, G.R., Lee, A., Liebscher, A., Rapuano, K. and Hancock, J.T. (2025) ‘AI-Generated “Workslop” Is Destroying Productivity’, Harvard Business Review, 22 September, updated 25 September. Used for the workslop mechanism and survey findings.

OECD (n.d.) ‘AI and Work’. Used for the balanced context: AI productivity and job-quality benefits, alongside risks including automation, loss of agency, bias, privacy, lack of transparency, and the need for training and worker consultation.

Parkinson, C.N. (1955) ‘Parkinson’s Law’, The Economist, 19 November. Used for the administrative-expansion argument and the British civil-service examples.

Parkinson, C.N. (1958) Parkinson’s Law: The Pursuit of Progress. London: John Murray. Used for the expanded version of the administrative-expansion argument.

Author Workflow Disclosure

This article was developed through an AI-assisted but human-directed editorial workflow. AI was used for structuring, language refinement, and analytical development. The final argument, interpretation, and responsibility remain with the author.

contact@rumbls.com

Terms and Conditions

Privacy Policy

Cookie Policy

© 2026 WDA Publishing. All rights reserved.

Disclaimer

Contact .