Notes - Reshuffle

Sangeet Paul Choudary | December 27, 2025

Chapter 1: Beyond Automation

The Real Story of Singapore’s Rise

Singapore’s transformation from a resource-poor port into one of the wealthiest nations is often attributed to government foresight or financial innovation. However, the true catalyst was a low-tech invention: the shipping container. In the 1960s, Singapore bet on containerization, building deepwater terminals and linking its port to global supply chains before others did.

Containerization as Coordination

While containerization looks like hardware automation (cranes and ships), its real impact was solving unreliability. Before containers, every shipment step (truck, train, ship) had to be renegotiated with separate contracts and handling standards. This fragmentation made shipping unpredictable and expensive.

The shipping container forced coordination, which is the ability to align independent agents toward a collective outcome. Key breakthroughs included:

A single integrated contract: Shippers used onel of lading for door-to-door delivery.
Standardization: Standard container sizes allowed boxes to move across different modes of transport without being unpacked.
Reliability: Predictable transit times allowed for just-in-time manufacturing, where businesses stopped stockpiling inventory.

Singapore’s success came from positioning itself as a coordination hub, selling reliability rather than just speed.

Coordination Eating Complexity

In the modern digital economy, the most transformative companies don't just build tools; they solve coordination problems.

Airbnb standardized trust through a central reputation system.
Stripe simplified financial complexity with a single API.
Tesla eliminated charging uncertainty with its Supercharger network.
Nike vs. Shein: While Nike uses manual oversight to coordinate global manufacturing, Shein uses algorithmic coordination. Shein’s algorithms read social media signals to predict demand and route micro-os to thousands of factories in real-time.

The Five Core Functions of AI

AI should be understood as a practical utility that performs five essential steps: sense, model, reason, act, and learn.

Sense: Observes the environment (e.g., Google Maps sensing location/traffic).
Model: Creates a representation of the world.
Reason: Applies rules to the model to find a path.
Act: Translates reasoning into actionable steps.
Learn: Continually updates its model based on outcomes.

AI as Coordination

The "Intelligence Distraction" causes us to judge AI by how human it seems rather than how effective it is. AI's real power is its ability to bridge the coordination gap—the divide between structured systems and the unstructured workflows that rely on tacit knowledge.

AI's economic logic rests on two principles:

Reducing task execution costs: Making expertise cheaper and scalable.
Lowering coordination costs: Bridging fragmented information across dartments or companies.

Coordination Without Consensus

Traditional coordination (like containerization or Walmart) requires everyone to agree on standards upfront. AI enables coordination without consensus by recognizing patterns in unstructured data to bridge systems that were never built to connect. For example, AI can take a voice note from a small tour guide and turn it into a structured listing on a global booking platform without requiring the guide to adopt new software.

The Flywheel of Economic Growth

Coordination drives specialization, which leads to fragmentation, which then demands even better coordination. This self-reinforcing flywheel is what fuels economic growth. Conversely, the coordination paradox occurs when companies use AI only for automation; speeding up one part of a system while others lag makes the overall organization more fragile and disjointed.

Chapter 2: The Wrong Frame

The Maginot Line Fallacy

After WWI, France built the Maginot Line, a 450-mile-long defensive masterpiece. However, in 1940, Germany launched a blitzkrieg, a new system of warfare that used two-way radios to coordinate tanks and aircraft in real-time. The Maginot Line was perfectly designed for a version of warfare that no longer existed.

Task-Centric vs. System-Centric Views

Most AI discussions suffer from a similar framing error, focusing on the binary question: "Will AI take my job?".

Task-Centric View: Sees AI as a tool to enhance or replace individual tasks. The system remains unchanged, and the solution is for workers to "reskill".
System-Centric View: Examines how AI transforms the organization of work itself. Roles may vanish even if the tasks they involve still exist.

Example: The Redundant Typist

Before word processors, typing pools were essential because errors were expensive and revisions required retyping whole documents. The word processor made the typist job redundant, not because the task of typing died, but because the system no longer needed a dedicated role to manage the constraint of expensive editing.

Unbundling and Rebundling

Systems are structured around managing constraints. When a constraint (like the cost of distribution) collapses, the system unbundles. When a new coordination method takes hold, components rebundle in new ways.

Music: Albums (bundles of songs) unbundled into digital tracks, then rebundled into Spotify playlists based on listener preference rather than production limits.
Basketball: Traditional positions (center, guard) were designed for a static system. Analytics shifted the game to "positionless" play, where players are valued for capabilities like three-point shooting and spacing that cut across old roles.

Examples of Systemic Reconfiguration

Amazon and Kiva Robots: The robots aren't there to "augment" workers; they are part of a fulfillment system designed to minimize human variability to guarantee two-day delivery.
Law Firms: AI converts tacit knowledge (expertise in a partner's head) into explicit assets. This strains the traditional "pyramid" model of firms, as fewer junior associates are needed for "grunt work" like contract review.
Shein: Unbundles the fashion designer's cohesive job into a gig-like role focused on speed and responding to algorithmic briefs.

Chapter 3: New Power, New Tensions

Walmart, Kmart, and the Barcode

In the 1970s, both Walmart and Kmart adopted barcodes. Kmart used them as a simple checkout tool. Walmart, however, built a "barcode-native" architecture. Walmart used the data to provide suppliers with store-level sales visibility, shifting power from manufacturers to the retailer. Scale only creates power when the system components reinforce each other through coordination.

The Four Quadrants of Value Distribution

AI grows the economic pie, but it doesn't mean shared prosperity.

Luddite Fallacy: Fear of machines taking jobs in a fixed system.
Techno-optimist Fantasy: Belief that innovation will expand the pie for everyone.
Task-based Misdirection: Treating the system as fixed while tasks change.
Rebundled System: Entire industries are rebuilt around new logic, and those who control coordination capture the upside.

The Five Levers of Coordination Power

Power in a complex system is managed through five structural levers:

Representation: Defining what is seen and measured.
Decision-making: The power to choose what happens.
Execution: Determining who carries out an action.
Composition: Controlling how different parts connect.
Governance: Setting the rules of participation.

The British Empire used these levers to extract value from India by mapping the land (representation) and centralizing decision-making in London while using local intermediaries for execution.

AI and the Reshuffling of Power

AI allows firms to exert control even without traditional market power.

Climate Corp: Created a digital representation of farms using satellite and soil data. By defining what constitutes "good farming," they gained the power to set insurance prices and steer farmers toward specific products.
Uber Freight: Platforms unbundle the trucking job. Experienced truckers once earned premiums for their judgment and navigation skills; platforms now centralize these decisions, reducing the driver's role to execution alone and driving down their pricing power.

Tensions in the AI Era

Every reshuffle introduces new tensions:

Workers vs. Tools: AI captures tacit knowledge, reducing worker autonomy.
Centralization vs. Decentralization: AI can push power to the edges (frontline sales teams) or consolidate it at headquarters (predictive pricing tools).
Tool Providers vs. Solution Providers: Companies like Uber relied on Google Maps, only to have Google eventually launch the competing service Waymo.

The Learning Loop Difference

Unlike static tools like paper maps or barcodes, AI benefits from learning loops. The more an AI system is used, the more data it collects and the more it transforms the system it serves. This creates accelerating effects where AI shapes system behavior, and that changing behavior further trains the AI.

Chapter 4: AI Unbundles the Job

Moving Beyond the "AI vs. Human" Binary

The common question "Will AI take my job?" is based on the flawed assumption that jobs are stable units of work and that skills alone provide security. Historically, the skill-biased view of change suggested that more education protected workers, but the rise of computers showed that routine tasks were the most vulnerable, regardless of education level. While task-based framing once promoted the idea of humans and machines working in complementarity, AI now challenges this by absorbing tacit knowledge and non-routine tasks that were previously thought to be safe.

The Typist Effect: System Redesign over Task Automation

A job can disappear even if its core tasks remain resistant to automation. The typist role vanished not because typing disappeared—people type more than ever—but because word processors eliminated the high cost of errors and formatting that originally justified a dedicated role. This illustrates that a job is a design artifact of a specific system of work; when the system’s architecture changes, the role’s logic may collapse even if the tasks persist.

The Three Forms of Value

To understand how AI impacts work, we must distinguish between three types of value:

Intrinsic Value: The subjective meaning or importance of a task.
Economic Value: How much someone is willing to pay, which is driven by scarcity.
Contextual Value: How pivotal a task is to enabling outcomes within a specific system. AI collapses economic value by removing the scarcity of skilled labor, turning once-expert tasks into "good enough" outputs at near-zero marginal cost.

Working Above or Below the Algorithm

In algorithmically managed systems, workers are divided into two classes:

Above-the-Algorithm: These workers (engineers, data scientists) design the systems and leverage algorithms to gain outsized returns, often aligning themselves with capital through stock ownership.
Below-the-Algorithm: These workers (drivers, warehouse pickers) are managed by the system, which narrows their tasks, minimizes their discretion, and captures the value of technological "augmentation" for the platform owner rather than the worker. Even creative professionals, like fashion designers at Shein or copywriters, risk being pushed "below the algorithm" as AI absorbs their judgment and reduces them to cogs in a reactive, data-driven machine.

System Interventions and Agency

Workers often attempt to regain agency through system interventions, such as Uber drivers coordinating logoffs to trigger surge pricing. This is dubbed Operation Mincemeat, a reference to a WWII deception where the British planted false signals to manipulate the enemy’s intelligence system. However, these are often temporary wins in a race against a learning system that eventually identifies and closes these loopholes.

Chapter 5: Rebundling the Job

Following the Constraint

When AI unbundles a job, value migrates to new constraints in the system. Constraints are the limiting factors that determine overall performance, and they fall into three categories: scarcity-based, risk-based, and coordination-based. To stay relevant, workers must stop chasing skills and start following these shifting constraints.

Lessons from the Sommelier and the Radiologist

The sommelier provides a masterclass in rebundling: in a world where wine knowledge is abundant and free, their value has increased because they manage emotional risk and curation. They don't just provide information; they provide confidence in a moment of uncertainty. Similarly, radiologists have not been replaced by AI because they rebundled their roles around clinical judgment and multidisciplinary coordination, moving away from simple image analysis (which AI does better) toward interpreting results in a complex patient context.

New Human Advantages: Curiosity and Curation

As AI makes answers abundant and cheap, value shifts upstream to curiosity (framing the right question) and downstream to curation (discerning which signals to act on).

Curiosity acts as an economic asset by focusing exploration on high-leverage targets, preventing the waste of cognitive resources on irrelevant AI-generated possibilities.
Curation creates narrative control, which is the ability to influence what people act on, not just what they see. Magicians do this by rebundling old illusions into elaborate performances to distract from the fact that their secrets can now be easily deconstructed online.

Visibility and the Tuareg Problem

Solving a constraint is only rewarded if it is economically visible. Tuareg guides in the Sahara salt trade held high contextual value but low economic value because their skill was fragmented, uncoordinated, and distant from the point of value capture. Today’s care workers and warehouse associates face a similar "Tuareg problem"—they resolve system breakdowns that are invisible to the central algorithm and are therefore economically undervalued.

Chapter 6: Rebundling the Organization

Structure as a Precondition for Freedom

Organizations often view autonomy and coordination as a trade-off, like a seesaw. Real Madrid’s Galácticos failed because they had superstar talent but lacked a shared framework for coordination, while Barcelona under Guardiola used positional play—a strict structure that actually expanded the tactical freedom of players by ensuring the team remained balanced.

The Coordination Tax

As organizations scale, they pay a "coordination tax"—the exponential growth in meetings, emails, and rituals required to keep teams aligned. Historically, the vertical filing cabinet helped manage this by separating knowledge workers from clerical staff who handled storage and retrieval. Today, AI can eliminate this tax by converting unstructured information (like Zoom calls and Slack threads) into structured, shared intelligence that is proactively served into relevant workflows.

From See-saws to Flywheels: Agentic Workflows

AI dissolves the autonomy-coordination trade-off by creating a reinforcing flywheel. Ramp uses AI to turn customer support calls into automated podcasts, allowing engineering and product teams to access the "voice of the customer" without filtered reports. Furthermore, agentic workflows—goal-oriented chains of AI agents—allow small teams to execute complex processes autonomously while remaining perfectly aligned with broader organizational goals.

The Coordinated Organization

The dream of the "autonomous organization" is a misconception; knowledge work requires human judgment and the management of interdependencies. The true goal is a coordinated organization where AI acts like an "integrated brain," unbundling knowledge from specific roles and rebundling it as a shared asset.

Chapter 7: Rebundling the Value Chain

The Building Blocks Economy

With the rise of cloud services, business capabilities have been unbundled from underlying assets into rentable, modular building blocks. MrBeast Burger launched in 300 locations overnight by renting these blocks: brand and demand (YouTube), food preparation (ghost kitchens), and logistics (delivery apps).

The Shift from Labor to Leverage

AI unbundles expertise from the expert, turning knowledge into a capital asset that is rentable, recombinable, and scalable. Competitive advantage shifts from what you own to how well you assemble and coordinate what others provide. This creates a new form of leverage involving:

Network Capital: Influencing a connected audience.
AI Production: Using expertise as a building block.
Agentic Execution: Running workflows at scale without a traditional workforce.

Capital vs. Labor: The Scanned Actor

A new tension arises between those who turn their skills into scalable building blocks and those whose skills are extracted and commoditized by capital. In Hollywood, background actors are scanned to create digital doubles, unbundling their performance from their physical presence and allowing studios to use their likeness infinitely without residuals. In contrast, OnlyFans creators use AI avatars as complements to scale their reach while using their "network capital" to ensure the avatar always points back to their real persona.

Defensibility through Positive Constraints

In a world where everyone has access to the same commoditized tools, "vibe coding" (prioritizing outputs that feel impressive but lack logic) becomes a risk. To build a defensible business, firms must choose positive constraints—deliberate design limitations that define performance. Muji created a unique, defensible supply chain not through proprietary tech, but through a strict system of constraints: no brand names, plain packaging, and a limited selection of essentials, which created a logic that competitors simply could not copy.

Chapter 8: The Tool Integration Trap

The Difference Between Tools and Engines

Competitive advantage in the age of AI depends on whether a company uses AI as a tool or an engine. A tool is a utility that speeds up specific, isolated tasks without changing how the overall system functions. In contrast, an engine is a core component that determines the performance of the entire system; the rest of the organization must be redesigned around its specific capabilities and constraints.

This distinction is illustrated by Formula 1 racing, specifically Renault’s 2005 victory over Ferrari. While Ferrari had a superior engine budget, Renault built their entire car (chassis, gearbox, and aerodynamics) as an integrated system designed to work in perfect harmony with their engine's power curve. In modern F1, winning teams do not just buy the best components; they own their "engine destiny" by ensuring every molecular detail of the car is tuned to that engine.

AI-Native Advantage: TikTok vs. Legacy Social Media

TikTok succeeded by treating AI as an engine rather than a tool. Traditional platforms like Facebook and Instagram used AI as a tool to enhance their existing "social graph" (who you know). TikTok rendered the social graph irrelevant by building a "behavior graph" where AI predicts what you want to watch based on your interactions.

TikTok also used positive constraints—limiting video length to 60 seconds—to create a high volume of data signals, allowing their AI engine to learn much faster than competitors. While legacy platforms competed on network size, TikTok won by changing the rules of distribution entirely.

The Solution Provider’s Dilemma

A structural tension exists between tool providers (who build modular capabilities like Google Maps) and solution providers (who bundle these tools to solve customer problems, like Uber). As tools become more intelligent, solution providers risk falling into an integration trap.

Learning Advantage: Tool providers learn from the aggregate data of all their customers, while a solution provider only learns from its own operations.
Performance-Based Lock-In: As a solution provider reorients its business around a third-party AI engine to gain short-term efficiency, it becomes increasingly dependent on that provider’s roadmap and pricing.
Wrapper Risk: Eventually, the solution provider may be reduced to a "wrapper"—a thin interface over someone else’s performance layer, losing all unique differentiation.

Chapter 9: The Solution Advantage

Moving from Tools to Solutions

Technology's potential is only realized when someone solves the adoption constraints surrounding it. Tools provide capability, but solutions provide reliability and accessibility in the real world. Historically, residential solar failed to scale until companies addressed the financing constraint through leasing models, shifting it from a "panel problem" to an "outcome problem".

Robots-as-a-Service: The Formic Example

In manufacturing, robots are often too expensive and complex for factories to manage. Formic addresses this by offering a robotic workforce as a solution rather than a tool. Formic designs, installs, and maintains the robots, charging by the hour for productivity. By absorbing the risk of downtime and technical failure, Formic makes automation accessible to factories that lack in-house expertise.

Three Business Models of Solution Providers

Work-as-a-Service: Charging for the consistent performance of a machine or task (e.g., Rolls-Royce charging for engine hours).
Results-as-a-Service: Guaranteeing a specific measurable result (e.g., Orica charging mining companies for "Rock-on-Ground" fragmentation quality rather than just selling explosives).
Outcomes-as-a-Service: Tying success to the customer’s ultimate strategic goal (e.g., pharmaceutical companies like Roche exploring pricing based on a patient’s actual therapeutic response).

Liability as the New Moat

As AI commoditizes specialized work, the "moat" shifts to liability management. Professional services firms (legal, accounting, etc.) are often hired not just for their expertise, but for the accountability they assume if something goes wrong. If these firms hand off work to AI and risk to insurers, they may be "sandwiched" and lose their pricing power.

Case Study: Vertical Integration in Mining

AI tool providers in mining face a dilemma: how far up the value chain should they climb?

Earth AI moves from being a tool (predicting deposits) to a solution by verifying and drilling the sites themselves and owning the mineral rights.
By combining AI prediction with physical verification, they avoid being stuck as a "component" and instead become a high-value gatekeeper of verified deposits.

Chapter 10: Designing for Indecision

Decision Support as Competitive Advantage

In a world of abundant choice, consumers suffer from decision fatigue and confusion. Companies that simplify these decisions win the "right to serve" the customer.

Best Buy avoided bankruptcy by pivoting from selling hardware to providing decision support. While Amazon competed on price, Best Buy invested in expert staff to help customers navigate complex purchases like home theaters. They turned "showrooming" into an advantage by offering price parity—once the customer trusted their advice, the purchase stayed in-store.

Direct vs. Derived Demand

Brands like L’Oréal compete in derived demand (selling products), while platforms like Sephora own direct demand (solving the customer's goal of "looking beautiful"). Sephora uses AI tools like Color IQ to capture customers at the moment of choice, establishing a control point that allows them to dictate terms to the brands on their shelves.

Strategies for Capturing Control Points

Companies can use AI to establish three types of control points:

Relationship-based: Leveraging existing trust (e.g., a salon network).
Workflow-based: Integrating into critical processes (e.g., Figma).
Intelligence-based: Guiding decisions through personalized data (e.g., Google Maps).

Rebundling Lifelong Learning

The traditional four-year degree is being unbundled by micro-credentials. The future of education involves AI-driven rebundling, where an AI assistant uses a "credential wallet" to map a person's fragmented skills to emerging work opportunities, coordinating a lifelong learning journey.

Chapter 11: Control Without Consensus

The Failure of Amazon Alexa

Despite massive adoption, Amazon Alexa struggled because it failed as a coordination mechanism. Users had to follow exact syntax and manually manage "skills" that didn't talk to each other. It failed because it didn't solve the coordination burden; it just provided an interface.

Coordination Without Consensus

Traditional coordination (like container shipping or CCC Intelligent Solutions in auto claims) requires consensus—all parties must agree on standard formats and codes before participating.

AI enables coordination without consensus. Tractable uses AI to interpret raw smartphone photos of car damage to generate repair estimates, allowing insurers and repair shops to coordinate without needing to adopt the same software or tagging systems.

Applications in Fragmented Industries

Construction: AI can extract data from architects' PDFs, engineers' structural models, and contractors' Excel sheets to create a unified representation of a project without forcing everyone to use the same tool.
Sales (Rox): AI agents can pull data across CRMs, emails, and call logs to create a unified customer view and trigger actions (agentic execution) across disconnected teams.
Finance (Ramp): AI creates a control point by monitoring spending across an organization in real-time, identifying duplicate subscriptions and enforcing policies at the point of transaction.

AI-driven rebundling overcomes the "cold start problem" of ecosystems by delivering value before participants have to standardize their workflows.

Chapter 12: You Don’t Need an AI Strategy

The Task-Level Fallacy

Most leaders mistakenly focus on an "inside-out" strategy, asking how AI can automate current tasks. Real strategy is "outside-in": starting with how the system is changing and then redesigning the organization to fit that new reality.

Redefining "Where to Play" and "How to Win"

Where to Play: This is determined by coordination. For Singapore, the "where" wasn't just a physical strait, but the landscape of reliable, global trade made possible by the shipping container.
How to Win: This is determined by control. Singapore won by resolving the constraints of the new system—integrating intermodal transport and creating a "trust layer" for trade through governance and anti-corruption measures.

Four Strategic Postures for AI

Reactive Optimizer: Uses AI to speed up old tasks.
Anticipator (The Gretzky): Anticipates value migration but plays the same game in old structures.
Logic Shifter (The Curry): Rewires how decisions are made (e.g., John Deere moving decisions from the farmer to the machine).
Field Reshaper (The Tiger Woods): Restructures the entire ecosystem by owning the means of coordination (e.g., Climate Corp).

Strategy in the age of AI is not about surviving the "reshuffle" but about mastering the new mechanisms of coordination to tilt the playing field in your favor.