Steering 3 — Scale Modularly
Steer systems to scale through modular containment, not centralised abstraction.
Scale is not free. It trades judgment for throughput. As systems grow, they must rely on abstraction—rules, models, automation, statistical inference—to remain economically viable. This is unavoidable.
The failure is not automation itself.
The failure is how systems behave once abstraction breaks.
Large systems fail when they cannot contain, correct, and reverse errors locally.
Scale Fails on Salt Curves (Goldilocks Zones)
Most large systems implicitly assume simple relationships between input intensity and outcomes:
- more effort → better results
- more rules → more safety
- more automation → lower cost
- more analysis → better decisions
- stronger incentives → better behaviour
This assumption is wrong.
Many critical system inputs follow salt curves (Goldilocks zones):
- too little input → failure
- an optimal band → effectiveness
- too much input → failure again
The input may be control, incentives, automation, analysis, oversight, or coordination effort.
Early movement along the curve produces real gains. Performance improves. Errors drop. Costs stabilise. Because these improvements are genuine, increasing input feels rational.
The failure occurs at the upper inflection point.
Beyond this point:
- additional input yields diminishing returns
- then negative returns
- then systemic harm
Crucially, the harm is delayed. The system continues to appear functional. When failure becomes visible, it is usually misdiagnosed as insufficient effort rather than excessive intensity.
Scale failures are therefore not design mistakes in kind.
They are errors of degree.
Why This Matters for Scale (Agency)
As systems scale, they implicitly assume a trade:
scale requires more control
more control requires less agency
This assumption is widespread—and wrong.
In reality, agency is a primary scaling constraint. Systems that suppress agency hit a ceiling early. Systems that solve the agency problem can scale 20× to 100× further before brittleness, overhead, and collapse set in.
This is not ideological. It is structural.
Agency is how systems:
- correct errors where they occur
- adapt to edge cases
- surface information early
- prevent small failures from compounding
When agency is suppressed, scale becomes fragile. Throughput increases, but resilience collapses. Correction costs explode. Growth slows, then reverses.
Agency itself follows a salt curve:
- too little agency → passivity, gaming, learned helplessness
- an optimal band → initiative, correction, local intelligence
- too much unbounded agency → incoherence
Most large systems overshoot on control and undershoot on agency at the same time.
That combination is lethal.
Solving the agency problem does not merely improve culture or morale.
It unlocks an order-of-magnitude increase in sustainable scale.
The Scale–Agency Paradox
Scale increases complexity.
Complexity increases edge cases.
Edge cases require judgment.
Which means:
the larger the system, the more agency it actually needs.
Yet large systems respond to complexity by removing agency—replacing judgment with rules, discretion with procedure, and responsibility with process.
This creates a structural paradox:
scale demands more local judgment
scale design eliminates local judgment
Errors are no longer corrected where they occur. They propagate upward, harden into precedent, and require force to undo. Agency is diffused everywhere and concentrated nowhere.
This is how systems slide from Quadrant I Gain or Quadrant IV Advantage into Quadrant III Collapse at scale.
Salt Curves and the Four Os
Salt-curve failure at scale expresses itself through four recurring patterns.
Overcorrection
Inputs are intensified past the optimal band. Early success justifies escalation. False positives rise, but are tolerated.
Oversight
Judgment is replaced by layered procedure. Humans exist, but without authority, accountability, or outcome responsibility.
Overlay
Instead of reducing input, organisations add compensatory subsystems—appeals, exception handlers, privacy managers, make-good gestures.
Offside
Resolution requires force external to the system. Regulators, courts, or ombudsmen must intervene.
Each O is an attempt to avoid admitting the system has overshot its Goldilocks zone.
The Paragentist Resolution
The Paragentist solution is not to reject scale.
It is to change what scales.
Do not scale constraints.
Do not scale abstraction.
Do not scale centralised control.
Scale agency modularly.
This means:
- pushing judgment down, not up
- keeping authority close to consequences
- allowing local reversal without global precedent
- bounding behaviour with interfaces, not micromanagement
Constraints still exist—but they bound modules, not people.
Control defines limits; agency does the work inside them.
Extreme Proof: Valve
Valve is a decisive, empirical proof that getting agency right unlocks extreme scale.
Valve is a video-game developer, publisher, and platform company best known for creating Steam, the dominant global digital distribution platform for PC games. Steam serves hundreds of millions of users worldwide and handles payments, DRM, matchmaking, updates, and marketplace transactions at enormous scale.
What makes Valve remarkable is not just what it builds, but how it is organised.
Valve operates with:
- no formal managers
- no fixed reporting lines
- self-selecting teams
- internal markets for talent allocation
- peer-based accountability tied to outcomes
Employees choose which problems to work on, form teams organically, and are evaluated by peers on delivered value rather than process compliance.
Despite this lack of traditional control, Valve generates extraordinary economic output.
In recent years, Valve has produced roughly $40–50 million in revenue per employee, depending on the estimate—20× to over 100× higher than large technology firms such as Apple, Google, Microsoft, Meta, or Amazon, whose revenue per employee typically sits below $2–3 million.
Valve did not achieve this by:
- tightening controls
- adding management layers
- centralising decision-making
It achieved it by treating agency as the primary scaling mechanism.
Coordination still exists. Constraints still exist. But judgment lives locally, authority sits close to consequences, and correction happens where errors arise.
This model is not universally transferable—but it falsifies the claim that scale requires agency suppression.
Reframing the Salt Curve
At scale, the relevant salt curve is not control alone.
It is the balance between control and agency.
Most systems fail because they:
- keep increasing constraints past effectiveness
- keep diffusing agency past usefulness
- mistake compliance for reliability
Once past the inflection point, every additional layer:
- slows correction
- increases resentment
- amplifies error cost
- destroys optionality
At that point, the system is not managing risk.
It is manufacturing it.
The Steering
Scale Modularly means designing systems where agency scales with complexity, not against it.
Before scaling, ask:
- where does judgment live?
- who can reverse errors locally?
- what does correction cost compared to resistance?
- how quickly can agency re-enter the system?
Scale works when failure is cheap, local, and reversible.
When it isn't, automation doesn't reduce cost—
it merely defers it, with interest.
That interest is paid in time, trust, and Fuckwittery.