21. The Oscillation · Coupling

After the Failure

A company suffers a major outage.

Customers are affected. Executives demand answers. Regulators ask for assurances that it will not happen again.

The response is immediate:

more approvals before release,
additional testing requirements,
new review boards,
clearer escalation paths,
tighter controls around deployment.

For a while, the changes appear to work. Incidents decrease. People become more careful. The organization feels more disciplined.

Then pressure shifts. A competitor releases features faster. Revenue targets tighten. Product deadlines slip.

Leadership begins asking a different question: Why has delivery slowed down so much?

Gradually, the same controls introduced after the incident begin weakening: exceptions get approved, reviews become rushed, temporary shortcuts become normal, output becomes more important than depth.

The organization accelerates again. Then another failure happens.

Usually with a different technical cause. Usually with the same structural pattern underneath: responsibility sounded clearer than it actually was.¹

This cycle appears far beyond software. Governments expand oversight after public failures, then loosen enforcement under economic pressure. News organizations tighten editorial standards after scandals, then weaken them again while chasing engagement and revenue. Corporations centralize governance after crises, then decentralize when speed and growth become more valuable.

Different industries. Different language. The same structural swing.

Why Systems Swing

Large systems rarely move steadily toward better design. They oscillate.

The reason is simple: the pressures acting on scaled systems point in opposite directions. Organizations need speed, coordination, adaptability, consistency, oversight, local judgment, centralized visibility, innovation, and reliability. No structure optimizes all of these simultaneously. Improving one pressure usually increases another cost somewhere else.

So systems swing between partial corrections. After visible harm comes reform. After reform slows performance comes acceleration. After acceleration fragments responsibility comes stabilization. After stabilization becomes too rigid comes another push for flexibility.

Each phase emerges because the previous phase created real pain. That is why oscillation persists. The system is not irrational. It is continuously compensating for tradeoffs it cannot permanently resolve.²

Reform

Reform begins after consequences become impossible to ignore. A failure exposes drift between decision-making and reality.

The institution responds by trying to restore clarity: clearer ownership, stronger controls, additional oversight, more explicit accountability, tighter boundaries.

At its best, reform restores learning. It reconnects consequence to the people capable of redesigning the system.

But reform also increases friction. Approvals take longer. Coordination costs rise. Throughput slows. Eventually pressure builds against the new constraints. That pressure creates the next phase.

Acceleration

Acceleration begins when the organization feels constrained by its own controls. The institution must move faster: ship more, publish more, scale more, deliver more. Speed becomes evidence of health. Throughput becomes easier to measure than understanding.

At first, acceleration feels productive because motion increases visibly. But systems can move faster than they learn.

Software teams experience this as rising deployment velocity while recovery quality stagnates. Institutions experience it as expanding initiatives layered onto unresolved structural problems. Media organizations experience it as distribution growth while editorial cohesion weakens.

The pattern is similar in each case: the system increases activity faster than consequence can return clearly enough to improve design.

Coordination debt accumulates quietly underneath the motion. More meetings appear because boundaries are unclear. More reporting appears because trust weakens. More synchronization becomes necessary because ownership has fragmented.

The organization remains active while becoming harder to understand locally.³

Entropy

Acceleration eventually creates fragmentation. Temporary exceptions become permanent. Dashboards replace direct understanding. Teams optimize metrics they can control instead of outcomes they can no longer fully see.

Responsibility becomes distributed across so many boundaries that no one can confidently explain how the full system behaves anymore.

This is organizational entropy. Not collapse. Not chaos. Drift.

The institution continues functioning while becoming slower to learn from reality. The dangerous part is that entropy often remains invisible for long periods because the system still appears operational. Reports still look healthy. Metrics still move. Meetings still happen.

But fewer people understand how decisions connect to consequences across the full structure.⁴

Stabilization

Eventually the accumulated drift produces enough pain that the system slows itself down again. Stabilization begins.

Rules harden. Exceptions narrow. Variance becomes suspicious. The institution focuses on risk management.

Sometimes this preserves learning. Sometimes it merely preserves procedure. That distinction matters.

Healthy stabilization makes failures visible early enough that redesign remains possible. Unhealthy stabilization optimizes for provable compliance while reality changes underneath the structure.

This is where bureaucracy often thickens. The institution stores the memory of past failures in procedure: review gates, approval chains, audit trails, mandatory reporting, standardized workflows.

Some of this is necessary. But over time, systems can become so focused on proving that procedures were followed that they stop adapting to the environment those procedures were meant to manage.

The organization becomes safer in the short term while growing less flexible in the long term.⁵

Neither Side Wins Permanently

It is tempting to describe these swings as battles between freedom and control, innovation and bureaucracy, builders and regulators, decentralization and centralization.

But none of those oppositions fully captures the structure. Each phase solves a real problem. Each phase creates a new one.

Reform restores accountability but increases friction. Acceleration restores adaptability but weakens coherence. Stabilization restores predictability but can suppress learning.

No pole wins permanently because scaled systems must continuously balance pressures that cannot be perfectly reconciled. That is why organizations repeatedly rediscover similar cycles under new language: agile transformation, platform centralization, digital modernization, governance reform, decentralization, standardization.

The labels change. The oscillation remains.⁶

What This Chapter Is Arguing

This chapter is not arguing that reform is useless or that acceleration is reckless. It is arguing something narrower: large systems often mistake the current corrective phase for the permanent solution.

But the pressures that created the previous phase never disappear completely.

The goal is not to eliminate oscillation. The goal is to preserve learning while moving through it.

Because systems rarely fail all at once. They fail when consequences can no longer return clearly enough for redesign to keep pace with drift.

Bridge to Chapter 22

This chapter described the oscillation. The next chapter turns toward cohesion: clear ownership, stable boundaries, role integrity, visible responsibility.

Not because cohesion eliminates tradeoffs. But because systems with stronger cohesion require less coordination machinery to remain understandable under pressure.

The challenge is not preventing acceleration forever. The challenge is surviving acceleration without dissolving responsibility every time speed becomes valuable again.

Systems do not permanently solve the tension between adaptability and control. They swing between corrections.

Systems do not permanently solve the tension between adaptability and control. The question is whether the swing still preserves the ability to learn.

Gene Kim et al., The Phoenix Project (IT Revolution Press, 2013), on incident-driven controls and delivery pressure; and Nicole Forsgren, Jez Humble, and Gene Kim, Accelerate, on velocity versus stability signals. ↩
Donella Meadows, Thinking in Systems: A Primer (White River Junction, VT: Chelsea Green Publishing, 2008), on competing goals and compensating dynamics in large systems. ↩
Donald T. Campbell, "Assessing the Impact of Planned Social Change" (1976), on activity metrics displacing learning; and Chapter 12 in this book on acceleration without ownership. ↩
See the interlude Coherence Under Scale and the Part V bridge in this book on drift and coordination substitutes; and Chapter 4 on coordination pressure. ↩
Chapter 20 in this book on bureaucracy as institutional memory; and James C. Scott, Seeing Like a State, on legibility and rigidity (structural risk, not anti-planning polemic). ↩
Christopher Pollitt and Geert Bouckaert, Public Management Reform: A Comparative Analysis, 3rd ed. (Oxford: Oxford University Press, 2017). ↩