MBA Module · Organization Design · v2 — evidence & executive edition

Designing the Organization

From the information-processing theory of the firm to the axes of organizational design — and their extension into AI-augmented and AI-native work.

An organization is an information-processing and coordination system. Its structure is a set of design choices that trade the cost of processing information against the uncertainty and interdependence of its work.

The argument in 90 seconds

  1. There is no one best organization. Structure must fit the work — its uncertainty, its ambiguity of meaning, and how tightly people's tasks depend on each other. A century of research converges here (Part 1).
  2. The design space has eight levers — four independent spines (decision rights, coordination, channels, knowledge), three modifiers (boundary, incentives, psychological safety), one composite (how much process you impose). You are already somewhere in this space, whether you chose it or not (Part 2).
  3. AI does not change the levers; it reprices them. Coordination, memory, and channel richness get cheaper — so settings that used to be unaffordable become viable, and your current position stops being optimal (Part 4).
  4. The matching rule: AI executes codified work, co-pilots adaptive work, and augments emergent coordination. The costly mistake is the leap from "AI could help with this" to "AI can run this."
  5. Your job shifts from designing processes to governing a loop — sense → interpret → decide → act — answering four constitutional questions: who approves workflows, who owns outcomes, what autonomy is permitted, what must not change (Part 4).

Four parts · use the sidebar to navigate

Part 1 — Foundations

How scholars learned to see the organization

Organization theory is not a pile of unrelated models. It is one long argument that slowly converged on a single idea: coordinating knowledge work is fundamentally a problem of processing information under uncertainty. Here is the line of thought.

The master lineage

Four moves turned a theory of individual decision-making into a theory of organizational design. Read left to right.

1947

Bounded rationality

Simon: people have limited attention and compute, so they satisfice rather than maximize.

1958–63

Behavioral firm

March & Simon; Cyert & March: slack, sequential attention, loosely coupled sub-units.

1973–74

Information-processing design

Galbraith: reduce the need for information, or raise the capacity to process it.

1990s+

Networks · knowledge · self-management

The canon extends into ties, tacit knowledge, modularity and post-bureaucratic forms.

"A wealth of information creates a poverty of attention." Herbert A. Simon, Designing Organizations for an Information-Rich World, 1971

Landmarks — click a node

Fourteen works that define the field. Click any point on the timeline to read why it matters.

Select a landmark

The intellectual landscape

Each node is a foundational contribution. Together they trace the shift from "the firm as a production function" to "the firm as a knowledge-creating, coordinating system." Click a node to explore.

The twelve lineages of thought

The synthesis groups the literature into twelve clusters. Expand each to see the core idea and its key thinkers.

Why this matters for design: Contingency theory (Burns & Stalker; Lawrence & Lorsch) supplies the licence for everything that follows — there is no one best way. Different environments select different points in design space. That is exactly why Part 2 is a set of axes, not a single recommendation.
Part 2 — The Organization (macro scale)

The dimensions you actually choose

At the whole-organization scale, if structure is a set of choices, what are the choices? A common starting point uses three axes — decision rights, communication channels, and knowledge. The literature says that is valid but incomplete. Below: a four-spine core, three orthogonal modifiers, and the contingencies that decide which setting is right.

Build an organization

Move the four core sliders — or load a known archetype — and watch the profile update. These four "spines" are the cleanest, most independent design levers.

Your organization resembles

Adjust the sliders to characterize a design.


The consolidated set — 4 spines, 3 modifiers, 1 composite

The literature throws up a dozen candidate "dimensions," but they aren't all independent. Consolidating for statistical distinctness leaves eight genuine design dimensions — plus two contingency inputs the design must fit rather than choose.

Design dimensionPolesType

Spine independent core axis   Modifier orthogonal add-on   Composite merges several correlated dimensions

What got merged: three heavily-correlated dimensions — formalization / standardization, span of control / attention, and the mechanistic ↔ organic spectrum — co-vary so tightly that they collapse into a single "structuring intensity" factor. Keeping them apart would have manufactured false orthogonality.

Contingency inputs (not design choices)

These describe the work, not the design. You don't set them — you fit the eight dimensions to them.

InputRangeRole

Fit, not fixed optima

Because the tradeoffs invert with the work, there is no universally best setting. Pick the nature of the task and see how the design should shift (Galbraith / Daft & Lengel contingency). Uncertainty is missing data; equivocality is disagreement about what the data means — they demand different remedies.

Task uncertainty & equivocality

MODERATE
Routine · well-understoodNovel · ambiguous
Live debates the framework keeps open: the centralization tradeoff inverts with task complexity (Bavelas–Leavitt); "mirroring" of team and product structure is a tendency, not a law (Colfer & Baldwin); media richness is learned, not fixed to a medium (channel expansion); codification vs. personalization is a strategic either/or, not a blend (Hansen et al.); and brokerage (Burt) vs. closure (Coleman) each buy something different.

One axis, two winning firms — the knowledge spine in the wild

The clearest evidence that these are choices with consequences, not abstractions, comes from Hansen, Nohria & Tierney's study of consulting firms (HBR 1999).

Case · codification
Ernst & Young / Andersen Consulting

Invested heavily in document repositories and reuse: solutions codified once, sold many times, staffed with leveraged junior teams. Won on scale and price for repeatable problems.

The strategy only pays where volume and repetition are high — the contingency table above, in action.

Case · personalization
McKinsey / Bain

Invested in networks of people, not documents: knowledge moves person-to-person, staffing is expertise-matched, fees carry a premium. Won on bespoke judgment for one-of-a-kind problems.

Hansen's rule: pick a dominant strategy (~80/20 mix). Firms that straddled the middle underperformed on both.

Part 3 — The Dyad (micro scale)

Zooming in: how two collaborators work best

The macro axes describe the whole organization. But most real work happens in pairs — two peers, a manager and a report, and now a human and an AI. The dyad is its own design unit, and it is where AI actually lands day to day.

The core idea: effective collaboration is the ongoing joint construction of common ground (Clark & Brennan), supported by trust, psychological safety, and shared mental models, and structured by matching channel richness/synchronicity to task equivocality. Communication is a coordination problem, not a transmission problem.

Three kinds of dyad

They share a spine but differ in their primary risks and mechanisms. Pick a type to see its profile.

The designable properties of a dyad

Synthesis across theories yields a compact set of properties — each with a validated measure. (A fuller ~13-construct set is in the dyad synthesis.)

What the theory recommends

Design moves that repeatedly show up across the literature. Expand each.

The human–AI dyad

Shares the human dynamics (grounding, trust, role clarity) but adds its own apparatus: joint cognitive systems, levels of automation, and the calibration of reliance.

Level of automation

MODERATE
1 · human does it10 · AI decides & acts

Calibrating reliance

The twin failure modes are over-trust (using AI when it's wrong) and under-trust (ignoring it when it's right). The goal is appropriate reliance — deployed systems sit well below the ideal (RAIR ≈ .54, RSR ≈ .64).

Follow correct AI

Human correctly updates to good AI advice — raises RAIR.

Over-trust

Human follows incorrect AI advice.

Under-trust

Human ignores correct AI advice.

Correctly reject AI

Human overrides wrong AI advice — raises RSR.

Counterintuitive finding: explanations don't automatically help — they can increase uncritical acceptance rather than appropriate reliance (Bansal et al. 2021). More transparency ≠ better collaboration.

Diagnose & design a dyad

A staged playbook. Click a stage.

Select a stage

Five stages

From diagnosis to redesign-or-exit. Click a stage above to see what it involves.

The scale between the dyad and the organization — the team and the unit. Most management happens at a scale this module compresses: the 5–15 person team and the business unit. The dyadic constructs above were mostly validated on groups of 3+ (Hackman's team-effectiveness tradition; Edmondson's safety work is team-level), so they extend upward better than the "dyad" label suggests — and the mirroring/modularity material in Parts 1–2 (Conway, Team Topologies) is precisely team-scale design. The unit scale is where the Part 4 loop gets assigned and where the companion diagnostic operates. Treat this as a known compression, not an omission by accident.
Caveats: most theories come from WEIRD, mid-20th-century samples; many "team" constructs were validated on groups of 3+, not pure dyads; psychological safety, trust, and support overlap empirically; human–AI findings are recent and partly speculative; and the widely-cited "65% of startup failures are people problems" figure is directional, not precise.
Part 4 — The AI-Native Organization

The organization as a control system

The information-processing view (Part 1) assumed human decision-makers with fixed cognitive limits. AI changes the cost, speed, and division of the organization's core loop — how it senses, makes sense, decides, and acts — and how it adapts that loop over time. The design question becomes: who runs each part of the loop, how fast does it turn, and what bounds its autonomy?

Why a loop, not a stack. An earlier draft modeled this as a five-"layer" architecture. We retired it: the "layers" mixed incompatible categories — a store (memory), a set of functions (capability), and a control loop (adaptation) — into a false stack. A feedback loop is the more honest structure: one organizing principle, with a genealogy in cybernetics (Beer's Viable System Model), Boyd's OODA loop, and Argyris's double-loop learning. conceptual framework
Speed of learning — how fast the loop turns — may be the dominant competitive variable in an AI-augmented environment, more than the quality of any single decision. Adapted from the cognitive-electricity thesis
Adjacent evidence, honestly labeled. This thesis is conceptual — but it has a decade-long existence proof in one domain. The DORA / Accelerate research program (Forsgren, Humble & Kim 2018; annual State-of-DevOps surveys) found that loop-speed metrics — deployment frequency, lead time for change, time-to-restore — separate high-performing organizations from low ones, and predict organizational performance, in software delivery. That is evidence that how fast the loop turns can beat the quality of any single decision in at least one knowledge domain. Whether it generalizes to whole organizations is exactly the open empirical question this part flags.

The operating loop: sense → interpret → decide → act

The core cycle every organization runs (Boyd's OODA), oriented by intent, drawing on memory, wrapped by governance, and adapted by a meta-loop. Click any stage or element.

Select an element

A feedback control system

Four operating stages turn continuously — oriented by intent (the reference signal), fed by memory (state), adapted by the meta-loop, and bounded by governance. The human/AI division of labor differs at every stage; click to explore.

Who runs each stage

The human ↔ AI division of labor is not one setting — it differs stage by stage. Assign each stage and see whether it fits, and the reliance risk.

Calibrate reliance, don't maximize it. The failure modes are over-trust (following AI when it's wrong) and under-trust (ignoring it when it's right); accountability stays human even where AI acts. Deployed systems sit well below ideal reliance, and explanations can even worsen over-trust (Bansal et al. 2021).

Matching AI to the work

How much of a stage AI can run depends on the work — the operational form of the Part 2 uncertainty/equivocality contingency. Slide it.

Nature of the work

ADAPTIVE
Codified · predictableEmergent · novel · political
The trap: "because AI can help with this, AI can run this." The precise formulation: AI executes codified work, co-pilots adaptive work, and augments emergent coordination. Each needs a different integration strategy.

Instrumenting the loop — what to measure

Part 3 gave the dyad validated instruments and thresholds. The loop deserves the same discipline: if the loop is the model, these are the gauges. None of these is a validated scale yet — they are operational proxies, DORA-style, and should be treated as a starting instrument panel, not a standard.

GaugeWhat to countWhat it tells you
Signal-to-action latencyElapsed time from a material signal (market, regulator, incident, customer) to a changed action in the fieldHow fast the operating loop actually turns — the headline cadence number
Decision latency & queue depthAge and count of decisions waiting on a named human or committeeWhere decision rights are mis-assigned; which gates are the bottleneck
Experiment throughputProcess changes proposed / trialed / institutionalized per quarter (mutate → select rate)Whether the meta-loop turns at all — and whether Select keeps up with Mutate
Eval coverage% of AI-run workflow steps with automated evaluation and monitored qualityThe feasible ceiling on autonomy — you cannot open the loop past what you can see
Reliance calibration (RAIR / RSR)Sampled accept/override decisions on AI output, scored against ground truthOver-trust vs. under-trust per loop stage — the human side of the ceiling
Time-to-institutionalizeTime from a validated improvement to it becoming the default workflowGovernance drag: distinguishes deliberate locks from mere organizational friction

Reading the panel: latency gauges price the operating loop, throughput and institutionalization gauge the meta-loop, and coverage and calibration bound how far governance can safely open either. A locked process with fast institutionalization is a choice; a locked process with slow everything is decay.

The driver: AI reprices coordination

Why the loop is changing at all: AI lowers the cost of running each stage and of coordinating across people, agents, and firms (Coase, Galbraith). The master variable is whether you use it to automate or augment — Brynjolfsson's "Turing Trap." Toggle it.

AutomateAugment
The same argument, in CFO terms. Automating today's output has a hard ceiling: at best you reach cost-parity with every competitor running the same models — a margin you compete away. Augmentation is uncapped: new capabilities, products, and services that did not exist to be priced. The political-economy framing (Brynjolfsson, Acemoglu) and the P&L framing point the same way; use whichever your board hears.

When execution gets cheap, the bottleneck — and the value — migrates. Where it goes:

Coase, revisited: firms exist because internal coordination can be cheaper than the market. AI lowers both costs — but may cut external-coordination cost more. The AI-native firm may be a smaller core (intent, governance, sensitive capabilities) around a larger AI-coordinated ecosystem of partners and agents. This is the boundary-permeability modifier from Part 2, now in motion.
The strongest objection — and why it sharpens the claim. Williamson's transaction-cost economics says firms internalize not because coordination is expensive but because of asset specificity, opportunism, and holdup: you do not outsource what makes you hostage, even at zero coordination cost. Taken seriously, this predicts exactly what stays in the shrunken core — the high-specificity, high-holdup assets: intent, the proprietary data flywheel, regulatory accountability, and the capability to evaluate what you buy (absorptive capacity). And one inoculation: DiMaggio & Powell's institutional isomorphism warns that most reorganizations copy structures for legitimacy, not fit. Expect a wave of imitative "AI-native reorgs"; the framework's whole point is that the right configuration is contingent on your work — a copied one is fitted to someone else's.

The meta-loop: how the organization edits itself conceptual framework

A second-order loop watches the operating loop and changes it — observe, evaluate, mutate, select. Single-loop learning fixes actions within goals; double-loop revises the goals and assumptions themselves (Argyris). AI compresses this from months to minutes. Click a function.

Select a function

Observe → Evaluate → Mutate → Select

Unlike biological evolution — random mutation across generations — this runs within the organization's lifetime, with directed mutation and conscious selection. Click a node.

Governance: bounding the loop’s autonomy

Self-modification without governance is pathological — in biological terms, cancer. Every process sits on a spectrum from locked to open, deciding how much the loop may change without human authorization. Classify each; see if it fits.

The four constitutional questions: Who authorizes a workflow? Who owns the outcome? What autonomy is permitted? What must not change? Clear boundaries create freedom within them — they let the organization move faster, not slower.

Mapping to machinery you already own: in regulated firms, Locked / Experimental / Open is not a new control framework — it maps onto model-risk tiers (SR 11-7), the three-lines-of-defense structure, and EU AI Act risk categories. "Locked" processes are those your second line already attests; "Open" is what internal audit has classified as immaterial. Extend the existing machinery; do not duplicate it.

What failure looks like — four cases

Each stresses a different element of the control system. All four are from public reporting; none involves exotic technology — which is the point. The failure mode is organizational, not algorithmic.

Governance · ungoverned change
Knight Capital, 2012

A manual deploy left obsolete order-routing code live on one server; the automated loop executed it at full speed. ≈ $440M lost in 45 minutes; the firm did not survive independently.

Pre-LLM, and precisely why it matters: any loop that acts autonomously needs change-control and a kill-switch before it needs intelligence. "What must not change" is question four for a reason.

Accountability · the loop's outputs are yours
Air Canada's chatbot, 2024

The airline's support chatbot invented a bereavement-fare policy; a tribunal held the airline liable for its agent's answer, rejecting the argument that the chatbot was "a separate legal entity."

The clean legal statement of this module's rule: accountability stays human (and corporate) even where AI acts. Question two — who owns the outcome — is not optional.

Meta-loop · mutation without selection
Zillow Offers, 2021

A model-driven buying loop kept acting on forecasts as the housing market shifted; evaluation lagged action. Wind-down with > $500M in write-downs and a 25% workforce reduction.

The loop sensed and acted fast but the Evaluate–Select functions could not keep pace. Speed of learning means the whole loop — accelerating half of it is worse than accelerating none.

Reliance · calibrate, then recalibrate
Klarna's support automation, 2024–25

Announced its assistant did the work of ~700 agents; a year later leadership acknowledged quality had suffered and began re-hiring humans for higher-touch service.

Not a failure of automation but of the matching rule: codified tickets fit "AI executes"; ambiguous, trust-sensitive ones needed the co-pilot configuration. The correction — re-tuning the mix — is the meta-loop working.

How AI renegotiates the spines

The macro spines (Part 2) still apply — but AI shifts their poles, costs, and contingencies. Toggle the view.

View each spine:
Classical orgAI-augmented

…and the modifiers — the half that usually gets skipped

Part 2 established three orthogonal modifiers. Boundary permeability got its renegotiation above (Coase). The other two — incentives and psychological safety — are hit hardest by AI and discussed least. Same toggle applies.

The tacit ↔ explicit boundary moves

Nonaka's SECI spiral assumed converting tacit to explicit knowledge was slow, human, and costly. AI shifts several of these conversions — with real risks attached.

The leadership imperative: governed adaptability

Leadership shifts from designing fixed processes to governing a dynamic system — setting the reference (intent and identity), tuning who runs each stage of the loop, and bounding the meta-loop's autonomy. The thesis is not maximal adaptability (that dissolves into chaos) but governed adaptability: continuous adaptation within constraints that preserve coherence, accountability, and identity — homeostasis, not rigidity.

Open questions the 1970s categories can't answer on their own:

Design stance: treat AI-augmented cognition as a new region of design space — not something to force into 1970s boxes. The four spines still apply, but their poles, costs, and contingencies are being renegotiated. The gaps the canon under-theorizes — distributed/remote work (Olson & Olson), platform/ecosystem organizations (beyond Powell), radical self-management (Lee & Edmondson), and AI cognition — are exactly where new organizational research should concentrate.

Objections, answered honestly

The four pushbacks every executive raises — plus the one this framework raises against itself. Expand each.

From framework to diagnosis — run it on your unit

A framework you cannot apply is a lecture. The module ends where your decision starts: Part 5 embeds the diagnostic instrument that operationalizes everything above.

Part 5 · The AI Organization Configurator

Twenty-two questions about one unit's work, constraints, current operating baseline, and intent — never its structural preferences — returning target, feasible-now, the three best-fitting AI-organization configurations, rank, calibrated band, distance gap, tensions, exclusions, influences, and roadmap. Run it per unit and read the results as a portfolio.

Continue to Part 5 — run the diagnostic →

Also available standalone ↗ for sharing with a respondent who shouldn't need the course first.

Part 5 — Diagnostic

Run it on your unit

Everything so far becomes actionable here. The instrument below asks twenty-two questions about one unit's work, constraints, current operating baseline, and intent — never about its structural preferences — and returns target, feasible-now, the three best-fitting AI-organization configurations, rank, calibrated band, distance gap, tensions, exclusions, influences, and roadmap.

Before you start — three rules of administration

One unit at a time

Answer for a single business unit or workflow family, not the whole firm. Fit varies by task — that is Part 2's entire argument. A bank's regulatory-reporting unit and its research desk should get different answers.

Someone who sees the work

The respondent should see the work directly, not only its reporting. Better still, have two roles answer independently — where they disagree, you have just measured equivocality (Part 2) rather than opinion.

It's a control setting

Re-run quarterly or after any meta-loop change; a configuration is a setting, not an identity. And log every "none of these fit" — clusters of misfits are how the preset library earns its next archetype.

The instrument

The configurator opens in its own clinical styling — deliberately: it is an instrument, not a lecture. It scrolls independently below. Open it standalone ↗ to share with a respondent.

Engine v0.2 — treat rank, calibrated band, raw-distance gap, tensions, exclusions, and roadmap as the payload. The instrument's generated design spec and known limits live in ai-org-configurator.md alongside it.

Reading the results as a portfolio

Run it for two or three more units and compare. The result is rarely one organization: expect a portfolio — a governed automation core in operations, an expert-copilot federation on the judgment desks, an adhocracy in incubation — with the center's real job being the shared substrate (memory, evals, governance) those configurations plug into. This is the modern successor to Mintzberg's divisionalized form: divisions differed by market; configurations differ by the nature of the work itself.