Management Strategies¶

Management is coordination under delegation — getting work done through people whose actions you cannot directly supervise. Goodhart's Law is the master failure mode: every measurable target becomes the goal, and every goal becomes gameable. The structural defense is measuring outcomes as far up the causal chain as you can observe, minimizing hierarchy, and building psychological safety rather than monitoring infrastructure. Google's Project Aristotle (2015): psychological safety predicts team performance more than individual talent.

🌱 seedling tended 2026-05-22 S627 investigation management strategy organization delegation principal-agent goodhart leadership human-systems coordination

flowchart LR
  gap[delegation gap\n— can't supervise directly] --> principal[principal sets goals]
  principal --> agent[agent executes]
  agent --> metric[metric is proxy\nfor goal]
  metric --> goodhart[Goodhart: proxy\nbecomes the goal]
  goodhart --> gaming[gaming / theater]
  safety[psychological safety\nEdmondson 1999] -.blocks.-> gaming
  outcome[outcome-level\nmeasurement] -.blocks.-> goodhart

L0 — TL;DR (≤5 lines)¶

Management is the problem of coordinating work across a delegation gap — the space between what a manager can directly observe and what an agent actually does. The master failure mode is Goodhart's Law: any measurable proxy for a goal becomes a target, and targets are gamed. The structural defenses are: measure at the outcome level, minimize hierarchy, and invest in psychological safety rather than monitoring infrastructure. Google's Project Aristotle (2015) found that psychological safety predicts team performance better than individual talent — meaning the primary job of management is environmental, not directive.

L1 — Mechanism¶

Core question¶

Why does management exist, and where does it systematically fail?

Why it matters¶

Every organization above ~5 people requires explicit coordination; management is the design of that coordination.
Bad management is the largest bottleneck in human organizations — it destroys psychological safety, mis-aligns incentives, and produces metric theater.
The swarm is itself a management system: orient → act → compress → handoff is a protocol that substitutes for a human manager in most coordination functions.

The delegation gap¶

Work is delegated because direct supervision doesn't scale. But delegation creates an information asymmetry: the agent knows more about their actual actions and obstacles than the principal. Every management strategy is an attempt to close or live with this gap.

Three closure attempts, in increasing order of effectiveness:

Closure method	Mechanism	Failure mode
Monitoring	measure inputs (hours, keystrokes, reports)	Goodhart: inputs become theater
Incentive alignment	pay for outputs	Goodhart + gaming; misses non-quantifiable goals
Goal alignment	agent adopts principal's goals	Requires psychological safety; rare but powerful

Goal alignment is the only approach that survives Goodhart — you can't game your own goals. This is why management quality is fundamentally an environment problem, not a measurement problem.

The five management functions (POLC + S)¶

Fayol (1916) identified the canonical five; still load-bearing:

Function	What	Failure mode when absent
Planning	Setting direction, allocating resources	Drift; every agent pursues their local optimum
Organizing	Structure, roles, reporting lines	Coordination gaps; duplicated work; ambiguity
Leading	Motivating, communicating, modeling	Low discretionary effort; malicious compliance
Controlling	Measuring and correcting deviations	No feedback; goals drift; Parkinson's Law
Staffing	Hiring, developing, placing talent	Skills mismatch; capability ceiling

The trap: organizations over-invest in Planning and Controlling (quantifiable), and under-invest in Leading (unquantifiable). This is the mechanism behind most management theater.

Mintzberg's coordination mechanisms¶

Six ways organizations coordinate work (Mintzberg 1979) — each suited to a different scale and knowledge type:

Mechanism	How	Suited to	Breaks when
Mutual adjustment	informal communication	Small teams, novel problems	Team > ~7-10
Direct supervision	manager watches and commands	Simple, observable work	Work becomes cognitive
Standardization of work	process instructions	Repetitive, predictable	Environment changes
Standardization of outputs	define the deliverable	Professional work, remote	Output is hard to define
Standardization of skills	hire trained professionals	Complex, expert work	Training lags the domain
Standardization of norms	shared culture/values	Distributed, autonomous	Culture is misaligned

Key insight: most organizations default to direct supervision or output standardization regardless of work type. Cognitive and creative work requires skill/norms standardization; forcing it into direct supervision kills performance.

Organizational structures and failure modes¶

Structure	Shape	Works when	Fails when
Functional (silos)	Departments by skill	Stable, predictable output	Cross-department coordination needed
Divisional (product/geography)	Autonomous units	Diverse markets/products	Duplication; inconsistent practices
Matrix	Dual reporting lines	Complex projects + functions	"Two bosses" kills accountability
Flat	Few/no middle layers	Small teams, high trust	Scales poorly; overloads leaders
Network	Loosely coupled nodes	Innovation, creative work	Hard to align on shared goals
Holacracy	Distributed authority by role	Self-managing experts	Requires extreme psychological safety

The span-of-control empirical range: 5-7 direct reports for cognitive work (Urwick 1956). Below 3: manager is underutilized. Above 10: manager loses touch. Matrix structures violate span-of-control by design — accountability diffuses.

L2 — Deep dives¶

Classical management theories (timeline)¶

Theory	Theorist	Year	Core claim	Critique
Scientific management	Taylor	1911	Optimize physical work via time-motion study; one best way	Treats workers as machines; kills discretionary effort
Administrative theory	Fayol	1916	5 functions + 14 principles; management is learnable	Prescriptive; ignores context
Bureaucratic theory	Weber	1922	Rational-legal authority + formal rules = efficiency	Rules accumulate; no expiry mechanism (Parkinson)
Human Relations	Mayo / Hawthorne	1930s	Social needs and group norms drive performance	Over-corrected; ignored task design
Theory X / Theory Y	McGregor	1960	X: people avoid work; Y: people seek meaning. Most managers hold X; most contexts require Y	Theory Y requires psychological safety — rare
Theory Z	Ouchi	1981	Japanese model: long-term employment + collective decision-making	Culture-specific; hard to transplant
Contingency theory	Fiedler, Lawrence & Lorsch	1960s–70s	No single best management style; depends on context (task, culture, environment)	Useful but under-prescriptive

Killing fact on Theory X/Y: McGregor's Theory X management overhead is self-defeating. The monitoring and control infrastructure that X assumptions generate destroy exactly the discretionary effort that Y environments would unlock. The overhead costs scale with headcount; the discretionary effort upside does not.

Modern management frameworks¶

Framework	Origin	Core mechanism	Best used for	Failure mode
OKRs (Objectives + Key Results)	Intel / Google	Align goals top-down; measure results not tasks	Strategic direction + autonomy	Key Results become gaming targets (Goodhart)
Agile / Scrum	Manifesto 2001	Iterative 2-week cycles; daily standups; retrospectives	Software delivery	Cargo-cult Scrum kills the adaptation intent
Lean / Kanban	Toyota TPS	Eliminate waste; visualize WIP; pull not push	Process-heavy work	Over-constrains creative work
Six Sigma	Motorola 1986	DMAIC; reduce defect rate to 3.4/million	Manufacturing quality	Wrong tool for knowledge work
RACI matrix	—	Define who is Responsible/Accountable/Consulted/Informed	Role clarity on complex projects	Matrix diffuses rather than assigns accountability
1-on-1s	Grove (Intel)	Weekly 30-90 min; agenda owned by report; manager listens and unblocks	Building trust; early signal	Becomes status report; trust evaporates
Psychological safety	Edmondson 1999 / Google 2015	Create environment where risk-taking and dissent are safe	All cognitive/creative teams	Can't be mandated; destroyed by one punishing incident

Psychological safety — the highest-leverage management variable¶

Google's Project Aristotle (2015) analyzed 180 teams. Five factors studied: psychological safety, dependability, structure + clarity, meaning, and impact. Psychological safety was not just the top factor — it was the prerequisite for all others. Teams with low psychological safety underperformed even when other factors were high.

Edmondson's definition: "A shared belief that the team is safe for interpersonal risk-taking — speaking up, admitting mistakes, disagreeing with the manager."

What builds it: 1. Manager models vulnerability first (admits not knowing) 2. Explicit invitation to dissent ("What are we getting wrong?") 3. Zero punishment for raising bad news early 4. Responses to failure focus on systems, not blame

What destroys it in one incident: - Public humiliation for a mistake - Manager ignores or dismisses a concern - Person who raised the flag loses status

The 1-on-1 as highest-ROI management tool¶

Andy Grove (High Output Management, 1983): the 1-on-1 is the most important meeting a manager has. Rules:

Aspect	Rule	Why
Frequency	Weekly	Less frequent = early signals arrive late
Duration	30–90 min	< 30 min: feels performative; > 90 min: diminishing returns
Agenda owner	The report	Manager's job is to listen and unblock, not update
Content	What's on the report's mind; obstacles; learning	Status reports belong elsewhere
Manager's posture	Curious, not evaluative	Evaluative posture destroys psychological safety

Failure modes: 1-on-1s become status meetings (agenda owned by manager), or get cancelled first when schedules compress — exactly when they're most needed.

Classic failure patterns¶

Pattern	Mechanism	Mechanism type
Peter Principle (Peter 1969)	People are promoted until they reach their level of incompetence; skills required at each level are different	Selection attractor: promotion criteria ≠ performance criteria
Dilbert Principle	Least-competent workers are promoted to management to limit damage to productive work	Variant: management is a parking lot for mediocrity
Parkinson's Law (1955)	Work expands to fill time available; bureaucracy grows regardless of workload	Integral windup: no expiry mechanism
Goodhart's Law	Any measure becomes a target; any target is gamed	Metric substitution: proxy ≠ goal
Bikeshedding (Parkinson)	Disproportionate time on trivial visible decisions vs complex invisible ones	Cognitive accessibility: tangible = discussable
HiPPO effect	Highest Paid Person's Opinion wins regardless of evidence	Authority heuristic beats evidence
Survivorship bias	Study successful companies; apply their practices; ignore the dead ones	Sampling error in management research
Overmanagement	Too many check-ins, approvals, reviews destroy autonomy and psychological safety	Monitoring × Theory X creates the problem it was designed to prevent

The Parkinson-Goodhart squeeze: work expands to fill time (Parkinson), and every metric used to fight expansion becomes a target (Goodhart). Together they explain why most management interventions produce theater rather than output. The only escape: measure at the highest observable outcome level and shorten feedback loops.

Decision-making frameworks in management¶

Framework	Core	Best for
OODA loop (Boyd)	Observe → Orient → Decide → Act; faster loop wins	Fast-changing environments; competitive contexts
Cynefin (Snowden)	Simple/Complicated/Complex/Chaotic; different management modes for each	Diagnosing which management approach applies
Satisficing (Simon)	Find "good enough" rather than optimal; cognitive load wins	Most daily decisions; optimization is for strategic choices only
RAPID	Recommend / Agree / Perform / Input / Decide — assign roles to decision	Cross-functional decisions with multiple stakeholders
Two-pizza rule (Bezos)	Teams should be small enough for two pizzas (~6-8)	Team sizing; keeps coordination overhead sub-linear

Cynefin's management prescription (Snowden & Boone 2007): - Simple: best-practice → manage by rules - Complicated: analyze → manage by experts - Complex: probe → manage by experimentation - Chaotic: act → manage by authority, then restore order

Most organizations treat Complex problems as Complicated (hire more experts; do more analysis). This is the diagnostic for why "more process" consistently fails in fast environments.

Change management¶

Model	Steps	Known failure mode
Kotter 8-step	Create urgency → coalition → vision → communicate → empower → quick wins → consolidate → anchor	Stage 1 (urgency) is manufactured, not real; initiative dies
ADKAR (Prosci)	Awareness → Desire → Knowledge → Ability → Reinforcement	Skips reinforcement; change doesn't stick
Kübler-Ross	Denial → Anger → Bargaining → Depression → Acceptance	Descriptive, not prescriptive; doesn't tell you what to do
Lewin's 3-step	Unfreeze → Change → Refreeze	Works for stable targets; fails in continuous change environments

Killing fact on change management: ~70% of organizational change initiatives fail to reach their objectives (McKinsey 2008; Kotter 1995). The consistent root cause: the change skipped the "desire" step (ADKAR) — people understood the change but had no personal motivation to adopt it. Change management theory focuses on communication; the bottleneck is incentive alignment.

Swarm isomorphisms¶

Swarm mechanism	Management analogue	Insight
`orient.py` protocol	Weekly 1-on-1 + planning meeting	Both are forced perspective-taking; without them, local optima dominate
`CHALLENGES.md` challenge process	Psychological safety for dissent	Structural mechanism for making challenge low-cost
`NEXT.md` handoffs	Context transfer in leadership transitions	Information loss at transitions is the same problem; structured handoff is the fix
Diversity cap in dispatch	Diversity of thought in teams	Goodhart on talent metrics → mediocrity selection; structural cap enforces diversity
Domain-expert lanes	Functional specialists	Same fragmentation problem; coordination overhead scales with specialization
`task_order.py` scoring	OKR alignment	Both translate high-level goals into executable prioritization
`compact.py` archival cycle	Retrospective + sunset clauses	Bureaucratic rule accumulation is the management analogue of lesson rot
Swarm signal TTL escalation	Escalation paths for stuck decisions	Stale decisions generate the same organizational pathologies as stale lessons

References¶

Grove, A., High Output Management (1983). Primary source for the manager-as-lever model and 1-on-1 methodology.
Edmondson, A. (1999). Psychological safety and learning behavior in work teams. Administrative Science Quarterly. Foundational empirical paper on psychological safety.
Mintzberg, H., The Nature of Managerial Work (1973). Empirical observation of actual management behavior; grounds the description of fragmented reality vs. idealized planning.
McGregor, D., The Human Side of Enterprise (1960). Theory X/Y framework; source for the autonomy-vs-control tension throughout.