AI Radicalism: Sangeet Paul Choudary's "Reshuffle"
It isn't about automation. It's about coordination.
“AI won’t take your job, but someone using AI will.”
An odious phrase, one we have all heard at conferences or training sessions, probably more than once. It puts a gun to the heads of every white-collar professional in the world. It also happens to be spectacularly wrong.
This is one of many fascinating and useful insights from Sangeet Paul Choudary’s new book, Reshuffle: Who Wins When AI Restacks the Knowledge Economy.
He aims his message at executives of incumbent enterprises across sectors. For veterans of the financial services industry still dealing with COBOL-based mainframes and siloed databases, Reshuffle will provide an excellent guide to thinking about how artificial intelligence will disrupt and how it can be harnessed.
Much of what we’ve heard about AI over the past few years, particularly since the advent of large-language-model generative AI, is misleading. I think Choudary’s work—a business book, not a pop-culture armchair read—provides a cogent and more practical world view about how AI’s impact on industries, organizations, and people.
For example, the discussion around AI is usually around automation. I recently moderated a panel at the Hong Kong Institute of Bankers’ annual event, with four seasoned COOs and business heads from major banks and insurers. And indeed, we talked a lot about productivity gains and the impact on their teams. This is fair for a manager or department chief who has to immediately deal with questions of tech and business strategy, human resources, KPIs, and RoI.
But it’s also a potential disaster, because Choudary believes these focus on the wrong things. The question, he insists, is not how smart is the machine? but what does our system look like once it adopts this new logic of the machine?
The distinction is critical. Financial incumbents have long embraced technology for efficiency: batch processing, reconciliations, compliance automation. But AI’s power lies in its ability to coordinate fragmented systems. This aligns decisions, workflows, and relationships in a way that was previously impossible.
It also means discussion of automation and up-skilling is not what will define success and failure. I’m not saying people can or should shrug these things off, but rather the onus for an organization’s survival has little to do with these small-beer initiatives.
Let’s get into Choudary’s arguments. He begins with the story of containerization. When standardized shipping containers transformed global trade, the revolution was not in the box itself but in the coordination it forced upon the world. AI, he suggests, is performing a similar function in knowledge economies. It reduces what he calls the “coordination tax”, the exponential costs of aligning decisions across teams, systems, and organizations. Where banks’ legacy architectures segregate deposit accounts from lending and risk functions, AI architectures learn to interlink them dynamically, generating a shared layer of understanding across incompatible domains.
For bank executives, this shift reframes AI from a feature in customer interfaces to infrastructure, an invisible layer enabling synchronized operations, compliance, and risk management across silos. Get this right, and many current woes due to legacy and ‘tech debt’ might be readily side-stepped.
Reshuffle presents the unbundling and rebundling framework as its central analytical axis: breaking down established systems into modular components and reassembling them into new forms of coordination and value. For financial incumbents, core banking modernization has often stalled because of path dependency: every ledger, payments engine, and KYC registry depends on another. Choudary’s insight is that AI effectively automates the rebundling itself.
Agentic systems—AI applications capable of interpreting goals and executing tasks autonomously—transform how work gets done. Rather than automating a teller’s form entry, AI may reconfigure the entire customer-journey architecture. Tasks once bundled into the role of a ‘relationship manager’ or ‘risk analyst’ are disassembled, optimized across systems, and reallocated, either to human specialists or autonomous agents.
In this context, ‘legacy’ becomes not merely a technology debt but a coordination bottleneck: a symptom of design logic that cannot exploit distributed intelligence.
Financial institutions that treat AI as a cost-reduction lever risk entrenching that bottleneck. Choudary warns of the “coordination paradox”: optimization within silos creates dysfunction at the interconnections. One team ramps up AI-driven lending approvals; another optimizes compliance validation; but absent a unified coordination logic, inconsistencies compound risk. The firms that win are those that rebuild around AI as connective tissue, reinforcing coordination without consensus.
Historically, banks accumulated advantage through ownership: of capital, infrastructure, or regulatory licenses. Today, Choudary observes, value accrues to those who align what they do not own into coherent systems, ie, orchestrators. In consumer banking, fintech challengers like Stripe exemplify this shift. Stripe doesn’t compete by offering payment terminals or merchant accounts (although it does do those things); it coordinates the fragmented world of payment instruments, currencies, and compliance obligations into a seamless interface.
This new logic challenges the core of incumbent financial institutions, whose power rests on proprietary systems and controlled channels. In AI-native ecosystems, control is derived less from regulation and more from coordination ability. For a broker-dealer or insurer, the ability to form API-level partnerships is no longer sufficient. What matters is the capacity to orchestrate: guiding data, decisions, and execution across domains where no single entity owns the entire process. AI becomes the invisible manager of the financial supply chain.
In one of the book’s most provocative arguments, Choudary redefines organizational knowledge as a form of capital: rentable, recombinable, and scalable. Once AI can read unstructured sources (meeting notes, audio calls, chat records) and generate shared summaries, institutional knowledge is disembedded from individual expertise and made reusable. In a financial context, imagine if knowledge of how to adjudicate a loan, price a derivative, or detect fraud became an accessible, queryable asset across the enterprise.
This dissolves traditional boundaries between labor and capital. What once demanded specialized teams—trade settlement exceptions, AML alerts, customer onboarding—becomes decomposed into modular workflows, executable by AI agents across distributed systems. Firms historically prized for their human expertise may find that expertise captured, codified, and leased back to them by technology vendors. For financial professionals, this implies a destabilization of the skill-premium economy. Value, Choudary says, shifts from knowing what to do, to knowing how to design the systems that decide what to do.
This judgmental capacity is where the ‘human in the loop’ remains vital. As AI commoditizes cognitive labor, the scarce resource is no longer information but discernment. Judgment, as he defines it, is the ability to act under uncertainty, while being accountable for outcomes. This has direct resonance for the financial sector. In markets, regulatory actions, and credit decisions, uncertainty is endemic. The banker’s or regulator’s role, then, is not to compete with AI’s speed of pattern recognition but to frame questions, set guardrails, and interpret ambiguous signals.
Two human abilities, curiosity and curation, retain enduring value. Curiosity frames the right questions; curation determines which answers deserve elevation. In financial reporting, the parallels are obvious: data analytics can generate thousands of risk correlations, but only good storytelling, the art of curating which correlations matter, creates clarity for boards and regulators. The narrative, Choudary implies, is the last bastion of human advantage in a machine-mediated system.
Choudary’s notion of the “coordination tax” deserves special attention from incumbents with sprawling legacy architectures. As firms scale, knowledge fragments. Teams duplicate effort, data diverges, and decisions delay. AI’s ability to traverse unstructured data means that it can act as a real-time coordination engine, learning the organization’s internal logic: how it frames trade-offs and interprets exceptions.
This turns the enterprise inside out. In past modernizations, knowledge remained embedded in human workflows. AI unbundles it, reconstituting it as shared capital. For a regional bank running multiple COBOL cores, this implies a powerful transition: instead of reengineering each subsystem, AI can abstract across them, interpreting behavior and outputs consistently. The coordination tax shrinks not because technology replaces labor, but because it reorganizes how labor, capital, and knowledge interact.
In the later chapters, Choudary examines a perennial technology dilemma, whether to build proprietary AI or rely on third-party tools. His test is simple: what happens if you lose access to the tool? If your business merely suffers inconvenience, buy. If it’s crippled, build. For financial incumbents, this distinction marks the line between competitive necessity and operational enhancement. Fraud detection, for instance, may thrive on off-the-shelf models. But risk modeling, compliance surveillance, and predictive liquidity management strike me as areas that define a firm’s strategic identity. These functions, or at least aspects of them, need to be retained as owned capabilities.
This logic reframes AI adoption strategy not as procurement but as existential design. AI-native firms, Choudary writes, don’t bolt AI onto existing structures; they rebuild the business model around it.
I suggest that for banks, we are no longer talking about digital transformation (modernizing what exists). We are entering a new phase, of AI reformation (rethinking what should exist). This means a lot of the tools in the banker’s toolbox, like applied programming interfaces (APIs, the connective tissue between software), could become out of date.
One of Reshuffle’s most striking contributions is its theory of “coordination without consensus.” Traditional software integration required compliance with standardized schemas; APIs provided limited, rule-based connections. AI agents, by contrast, learn across heterogeneity. They observe how disparate teams operate, infer intentions, and generate shared representations. This capacity to align systems without predefined consensus is precisely what legacy financial ecosystems lack.
In payment networks, correspondent banking, or insurance claims processing, heterogeneity (ie, ‘exceptions processing’), has been treated as an unavoidable cost. AI-driven coordination dissolves that constraint, through its ability to interpret disparate data sources, rather than through standardization. Each node remains autonomous, yet the system functions as if integrated. For cross-border clearinghouses or distributed finance consortia, that is a seismic shift.
For leaders in financial services, Choudary’s framework offers sobering and practical lessons:
AI is not an upgrade but a governance shift. It changes how control is exercised, from top-down compliance to bottom-up feedback. Governance emerges not through rules but through continuous observation.
Legacy integration is a coordination challenge, not a software one. AI exposes the limits of point solutions by making evident the frictions between them.
The new competition operates above the algorithm. Those who own the models may not capture the most value; those who orchestrate decisions and distribution around them will.
Traditional productivity metrics are misleading. Efficiency inside a department can destroy coherence across the enterprise. The metric to optimize is coordination quality.
In this context, financial institutions face a choice: remain custodians of static infrastructure or become orchestrators of dynamic ecosystems. The former optimizes for compliance; the latter competes for relevance.
In this new world of AI, the standard corporate playbook for digital transformation no longer works. AI is not a laboratory initiative, built around innovation labs, agile working, sprints. These may remain helpful at a tactical level, but they do nothing to reconsider how the basis of value creation is changing, nor sustain an incumbent’s advantages of trust, compliance, and scale. If someone else can better coordinate all the disparate elements, banks or insurers focused just on digital transformation will lose.
The good news is that there’s no reason why financial institutions cannot leverage AI to remain leaders in their field, or even set the rules for the next cycle. But AI requires an even more radical approach than digitalization. Choudary’s book is a good place to start.
Choudary, Sangeet Paul, Reshuffle: Who Wins When AI Restacks the Knowledge Economy, Future-First Books, 2025.