1. The Scenario That Deserves Attention and the Assumption It Requires
In early 2026, a macroeconomic scenario published by Citrini Research – The 2028 Global Intelligence Crisis – spread rapidly across financial markets and technology circles. Written as a retrospective from June 2028, it outlined a future in which rapid advances in artificial intelligence triggered widespread automation of white-collar work, leading to collapsing consumer demand and a sharp repricing of global markets.
The mechanism Citrini proposed was both elegant and unsettling. As AI systems became capable of performing large portions of knowledge work, enterprises deployed them aggressively to improve productivity and reduce costs. Human labor displaced faster than new forms of employment emerged. Household income fell, consumption contracted, and an economy that appeared productive on paper began to hollow out in practice.
The report introduced a powerful term for this phenomenon: Ghost GDP – an economy where output continues to rise while the human purchasing power that sustains it quietly erodes.
The macroeconomic logic is coherent. If productivity gains accrue primarily to capital, particularly infrastructure providers and compute owners, rather than to labor, economic growth may become increasingly disconnected from broad-based prosperity. That is not science fiction. The velocity of money slowing as labor income contracts, consumer spending softening faster than GDP reflects it, mortgage books quietly deteriorating – these are plausible transmission mechanisms, not fantasy.
But the Citrini scenario rests on a critical assumption that deserves far closer scrutiny:
That enterprises can absorb and deploy artificial intelligence at roughly the same speed at which its capabilities improve.
That assumption does not withstand examination.
When one looks closely at how large organizations actually function, their data environments, governance structures, regulatory obligations, and coordination complexity, a very different picture emerges. The primary constraint on enterprise AI adoption is not intelligence. It is organizational coherence. And until that constraint is addressed, the path from AI capability to economic transformation will be far slower, far messier, and far more structurally constrained than rapid-displacement scenarios suggest.
This does not mean the economic risk Citrini identifies is imaginary. It means the shape of disruption will be fundamentally different from their version. It will not arrive as a cliff. It will arrive as a slow erosion, unevenly distributed, difficult to diagnose, and for that reason, harder to respond to.
2. Enterprises Do Not Move at the Speed of Technology
Enterprise AI adoption data already reflects the depth of this structural friction — and the numbers are more striking than most boardroom conversations acknowledge.
McKinsey’s 2025 State of AI survey found that 88% of enterprises now report regular AI use in at least one business function. Yet nearly two-thirds remain in the experimentation or piloting stage. Only 39% of organizations report any EBIT impact from AI and among those, most say AI accounts for less than 5% of enterprise-wide earnings. BCG finds that the average organization abandons 46% of AI proofs of concept before production. Kyndryl reports that 62% of organizations have not moved AI projects beyond the pilot stage. S&P Global reports that 42% of companies abandoned most AI initiatives in 2025, more than double the prior year’s rate.
Read these numbers together. Enterprises are spending aggressively. The capability is real. The intent is genuine. And yet full-scale deployment remains the exception. The gap between what AI can do and what enterprises can actually absorb is not a technology gap. It is an organizational one and it is far wider than rapid-displacement scenarios assume.
Large institutions are not optimized for speed. They are optimized for continuity, compliance and risk containment. The systems that govern decision-making inside banks, hospitals, insurers and global corporations exist precisely to prevent rapid and poorly understood change. Enterprise cloud adoption unfolded over more than a decade. ERP modernization initiatives regularly span five to ten years. Artificial intelligence does not bypass these dynamics. If anything, it amplifies them.
Deploying AI agents inside an enterprise is not merely a software upgrade. It requires redesigning decision workflows, validating probabilistic outputs under real operational pressure, aligning governance processes, and ensuring compliance across multiple regulatory regimes simultaneously. BCG’s analysis of organizations that have actually achieved AI transformation yields a striking finding they call the 10-20-70 rule: 10% of AI effort should go to algorithm design, 20% to technology and data infrastructure, and 70% to people, process redesign, governance and change management. Most organizations invert this ratio entirely; over-investing in models and under-investing in the organizational foundations that make those models work.
This is why the organizations capturing real value from AI today are rarely the fastest adopters. They are the ones that move deliberately – integrating systems, validating outcomes, and expanding only when their operational foundations are genuinely ready. Companies that solve integration and governance challenges first achieve four times faster AI deployment and three times higher value capture rates than those that rush.
The bottleneck is not intelligence. It is organizational coherence – and coherence cannot be purchased off the shelf.
3. The Enterprise Entropy Constraint
In a previous essay on Data Entropy, I described a pattern that quietly shapes every enterprise technology environment. Over time, organizational systems accumulate disorder. Every merger introduces incompatible data architectures. Every SaaS application generates its own schemas and definitions. Regulatory requirements introduce reporting layers that rarely integrate cleanly with existing systems. Teams develop localized workflows that encode business logic outside formal enterprise platforms. The result is a steady, compounding expansion of semantic fragmentation across enterprise systems.
This dynamic can be expressed as what I would call the Enterprise Entropy Constraint:
As organizations scale, the rate at which their data ecosystems accumulate entropy grows faster than the rate at which the enterprise can reconcile it.
This is not merely an observation. It functions as a governing principle of organizational complexity, one with profound and underappreciated implications for artificial intelligence deployment.
AI systems require coherent inputs to produce reliable outputs. Yet the environments into which these systems are being introduced are rarely coherent. They are the accumulated residue of decades of technological layering, regulatory change, and organizational evolution. An AI agent that encounters two conflicting definitions of “active customer” across CRM, billing and support systems does not pause for clarification. It improvises. It bridges the gap probabilistically. And in doing so, it produces an output that is syntactically correct and semantically wrong. It does not fail loudly. It fails silently and confidently. In an enterprise context, that is the most dangerous failure state of all.
What makes the entropy constraint structurally intractable is that entropy is not primarily a technical problem. It is the direct byproduct of how enterprises are governed, grown and regulated.
Every new compliance requirement adds data fields, retention policies and reporting workflows typically bolted onto existing systems rather than architecturally integrated. Each bolt-on introduces semantic drift. Consider GDPR’s right-to-erasure requirement. When a customer requests deletion, an enterprise must purge data across CRM, billing, support, analytics, email archives and third-party integrations. To demonstrate compliance, organizations create audit tables and deletion logs, each with its own schema, its own definition of “deleted,” and its own maintenance burden. The enterprise becomes more compliant on paper and more fragmented in practice. An AI agent deployed six months later inherits the full disorder.
In 2024 alone, U.S. federal agencies introduced 59 AI-related regulations, more than double the prior year. The EU AI Act now mandates bias-tested datasets, exhaustive technical documentation and human-in-the-loop oversight for high-risk AI systems, precisely the domains where enterprise AI would have the most transformative impact: credit decisioning, healthcare diagnostics, employment screening, legal determinations. These requirements do not merely slow AI adoption. In certain domains, they make full automation legally impossible, period.
The deepest irony is this: the compliance infrastructure that constrains AI deployment is itself one of the primary engines of entropy generation. The enterprise tries to become more compliant and inadvertently becomes more entropic. Until leaders treat entropy as a measurable, manageable enterprise risk, with the same seriousness as cybersecurity or financial exposure, AI systems will continue to inherit disorder they were never designed to navigate.
4. Coordination: The Hidden Bottleneck of Automation
Even if enterprise data environments were perfectly structured, another constraint would remain – one that receives almost no attention in mainstream AI strategy discourse.
Enterprise work is not simply a collection of tasks. It is a network of coordinated decisions.
Consider a typical operational process. A credit decision involves risk analysis, regulatory compliance review and portfolio exposure management. A product launch requires legal approval, financial forecasting, marketing alignment and supply chain coordination. Customer service decisions often depend on historical context spread across multiple disconnected systems. Automation can replace isolated tasks with relative ease. Replacing the coordination layer is a fundamentally different challenge.
Organizations rely on negotiation, escalation paths, informal communication channels and human judgment to reconcile the edge cases that formal systems cannot anticipate. These are not inefficiencies waiting to be optimized away. They are the connective tissue of organizational decision-making – the mechanism by which exceptions get handled, ambiguities get resolved, and business logic that was never formally documented gets applied correctly.
Introducing AI agents into this landscape does not eliminate coordination requirements. In many cases, it increases them. Probabilistic outputs must be validated. Exceptions must be investigated. Decisions must remain auditable. Regulatory obligations require human oversight in high-impact domains. Gartner projects that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. That trajectory is credible. But it describes integration into existing coordination structures, not replacing them.
The constraint on enterprise automation is therefore not raw intelligence. It is the organizational coordination required to deploy intelligence safely and that coordination requirement does not diminish as AI improves. It evolves.
5. The Living Error-Correction Layer
For decades, experienced employees inside enterprises have performed a function that rarely appears in any formal job description.
They act as living error-correction layers against entropy.
The senior analyst who knows which internal dataset cannot be trusted. The underwriter who recognizes patterns that fall outside historical models. The operations manager who understands which exceptions the documented process never captured. These individuals do more than execute tasks. They continuously compensate for the entropy that accumulates inside complex organizational systems carrying institutional knowledge that was never encoded anywhere, applying judgment that no process map has ever fully articulated.
When automation initiatives remove these individuals prematurely; before their institutional knowledge has been captured, validated and encoded into the systems that replace them; the enterprise does not merely lose labor capacity. It loses its informal stabilization mechanism.
AI systems deployed in their place inherit the full entropy load of the organization’s data and processes, without the contextual awareness required to navigate it. The errors that follow rarely appear immediately. They propagate quietly through interconnected workflows, producing outputs that appear technically correct while gradually diverging from operational reality. The organization moves faster toward outcomes it did not intend and cannot easily diagnose.
This is the precise mechanism by which misdeployment compounds. And it is why the workers most at risk from poorly sequenced AI adoption are not necessarily the least skilled. They are the ones whose judgment was serving as the last line of defense against a data environment that no formal system fully understood.
Displacing your error-correction layer before encoding what it knows does not accelerate transformation. It accelerates entropy.
6. Misdeployment, Not Displacement – The Real Enterprise Risk
Here is a headline most people remember: in 2024, Klarna announced its AI agent was doing the work of 700 customer service representatives. It was cited widely as evidence that Citrini’s scenario was already beginning to unfold. What received far less attention was that by mid-2025, Klarna’s CEO Sebastian Siemiatkowski publicly acknowledged the company had gone too far. “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” Klarna began rehiring human agents.
The important detail is not that the AI failed technically. The system handled two-thirds of customer interactions and reduced average resolution time significantly. By narrow operational metrics, it performed. What failed was the sequencing. The institutional knowledge held by experienced support staff who understood edge cases, customer nuance and the exceptions that formal process never fully anticipated walked out the door before it was captured or codified. The gaps only became visible under the full weight of live operational pressure.
Klarna is not an outlier. It is the most visible data point in a pervasive pattern. BCG finds that only 6% of enterprises qualify as genuine AI high performers, defined as organizations reporting at least 5% EBIT impact from AI. The remaining 94% are investing at scale and capturing value at the margins, if at all. More than 80% of organizations report no meaningful enterprise-wide EBIT impact from generative AI despite years of experimentation and hundreds of billions in cumulative investment.
This brings us to the economic argument and here it diverges most sharply from Citrini’s version.
Citrini’s economic thesis rests on mass displacement creating a consumption collapse. It is a coherent transmission mechanism. But the actual economic risk unfolding in enterprises today is different in character. It is not mass displacement. It is massive, misallocated investment. Global AI spending surpassed $250 billion in 2024. Yet Nobel laureate Daron Acemoglu’s MIT study projects just a 0.5% productivity increase attributable to AI over the next decade. Apollo’s chief economist captured the gap precisely: AI is everywhere except in the macroeconomic data.
The investment flows to compute owners and infrastructure vendors. The productivity gains do not follow at the pace or scale the investment implies. Capital is not lost in a dramatic crash; it is quietly stranded in infrastructure that was never ready to support what was built on top of it.
That is the slow version of Ghost GDP. Not mass unemployment. Not a market crash. A compounding divergence between what enterprises appear to be building and what they are actually delivering, sustained by board optimism, vendor incentives, and the organizational difficulty of admitting that the foundation was never solid enough to build on.
7. Addressing the Critics
Several counterarguments naturally arise when discussing enterprise constraints on AI adoption. They deserve direct responses.
“AI-native startups will simply bypass these limitations.” This is partially true and it is the counterargument that deserves the most serious engagement, because the data behind it is real and growing. Menlo Ventures’ 2025 State of Generative AI in the Enterprise found that AI-native startups now capture 63% of the enterprise AI application market, up from 36% the prior year, earning nearly $2 in revenue for every $1 earned by incumbents. In highly regulated domains like finance, AI-native challengers are building next-generation alternatives to legacy platforms precisely because incumbents cannot move fast enough to deliver a credible product.
This is the competitive pressure argument at its sharpest, and enterprise leaders should not dismiss it. The disruption channel here is not internal replacement, it is external revenue compression. Incumbents may be partially protected by their own friction from rapid internal displacement. They are not protected from competitors who operate without that friction and are steadily capturing the customers and use cases they are too slow to serve.
However, the picture is more nuanced than a simple “startups win” narrative. The same Menlo data shows incumbents holding 56% of the infrastructure layer, where reliability and integration depth outweigh speed. Bessemer Venture Partners observes that the second act of AI disruption is already underway with incumbents responding through aggressive M&A, acquiring AI-native capabilities they cannot build fast enough internally. The organizations most at risk are not those moving cautiously on solid foundations. They are those moving neither fast enough externally to stay competitive, nor deliberately enough internally to build the governance and data foundations that durable transformation requires.
The honest answer is this: organizational friction slows internal adoption but does not slow external competitive pressure equally. Enterprise leaders face a genuine dual challenge – maintaining the sequencing discipline that prevents internal misdeployment, while simultaneously moving fast enough externally to avoid ceding ground to AI-native challengers. These are not the same problem and conflating them produces either reckless speed or paralytic caution. The leaders who navigate this decade successfully will be those who can hold both imperatives simultaneously.
“AI will resolve enterprise data problems by cleaning and reconciling fragmented datasets.” This misunderstands the nature of entropy. Enterprise data disorder is not merely a formatting issue. It reflects institutional context that was never formally documented, decision logic that exists only in the experience of the people who built and operated those systems. AI systems cannot infer what was never recorded. They can accelerate reconciliation once humans have codified the underlying logic. They cannot replace the codification itself.
“Gradual automation will ultimately produce the same economic outcome Citrini describes, just more slowly.” Perhaps. But the timeline matters enormously. A two-year automation shock produces systemic instability in labor markets, financial systems, and the policy frameworks designed to respond. A multi-decade transition allows institutions, labor markets and governance frameworks to adapt. The difference between those two paths is not merely pace. It is the difference between crisis and transformation.
“Enterprises that proceed cautiously will simply be outcompeted by those that don’t.” This is the most seductive argument and the one most likely to produce the very misdeployment risks this piece has described. Speed without foundation is not competitive advantage. It is fragility that has not yet been tested. The organizations that will define the next decade of enterprise performance are not those who moved first. They are those who moved with the most sequencing discipline.
8. Strategic Readiness – What Enterprise Leaders Must Do Now
The organizations that capture durable value from artificial intelligence will not necessarily be those that deploy it first. They will be the ones that deploy it on coherent foundations.
That distinction has a specific operational meaning. Before deploying agents capable of making decisions at scale, enterprise leaders must confront a set of structural questions that rarely appear in AI strategy discussions but that determine whether AI becomes a source of transformation or a source of compounding fragility.
Where does entropy exist inside your organization? Which systems contain conflicting definitions of critical business concepts? Which workflows depend on institutional knowledge that has never been formally encoded? Which processes rely on human judgment that your organization has never needed to articulate explicitly because a person was always there to exercise it?
These are not technical questions. They are organizational ones. And answering them honestly requires a different kind of readiness audit than most enterprises have undertaken.
The research on what separates organizations that succeed is consistent. Accenture’s Making Reinvention Real with Gen AI report, drawing on over 2,000 client engagements and insights from more than 3,000 executives, finds that organizations with responsible AI principles and governance in place across the full AI lifecycle are 2.7 times more likely to create enterprise-level value from AI. As BCG’s 10-20-70 rule makes clear, the investment ratio that predicts success is the inverse of what most organizations currently practice and getting that ratio right is a leadership decision, not a technical one.
If you are looking for a concrete place to start, identify one high-value workflow where AI has been proposed or piloted, and map every data source that workflow touches. For each source, ask these three questions – who owns the definition of the core concepts in that data? When was it last validated against operational reality? And is the same concept defined consistently across every system an AI agent would need to consult? The answers will tell you more about your AI readiness than any benchmark survey. They will also tell you exactly where to invest first.
The practical sequence beyond that is demanding but clear. Establish systems of record and data contracts before deploying agents at scale. Invest in knowledge transfer, capturing institutional judgment from experienced employees before restructuring the roles that hold it. Build governance architecture that makes AI decisions auditable, explainable and correctable. Deploy in high-coherence domains first. Validate rigorously. Expand only when the foundations hold.
The economic stakes of getting this right extend beyond the individual firm. When enterprises deploy AI on solid foundations – with governance in place, knowledge properly transferred, and workforce evolution rather than abrupt displacement – productivity gains circulate more broadly. Value is created, not merely extracted. That is the path away from the Ghost GDP that Citrini rightly fears, even if the route there looks nothing like their two-year cliff.
Intelligence is no longer the scarce resource inside enterprises. Organizational coherence is. The companies that recognize this early will not simply avoid costly missteps; they will build the operational architecture required to turn abundant intelligence into durable competitive advantage.
The disruption is real. The timeline is longer than the scenario suggests. The shape is messier. And the stakes, for leaders who misread either the urgency or the sequence, are precisely as serious as Citrini warns.
Prepare strategically. Build on solid ground. And before you deploy the intelligence, make sure you understand exactly what it is being built on.
About the Author
Debiprasad Banerjee is the CEO of Acxelin Agentix, an AI-first company helping enterprises reimagine customer engagement through intelligent automation and digital experiences. Through his writing, he shares practical insights on AI, innovation, and the future of business. Explore more articles from the Acxelin Agentix team for expert perspectives on enterprise AI and digital transformation.