Introduction: The Collapse of Time
Imagine an Emperor of China seated at the apex of his vast dynasty. When he issues a proclamation, the imperial edict must travel by horseback across mountains, rivers, and provincial borders — a journey of days, weeks, sometimes months before the words of the sovereign reach the ears of a frontier garrison commander. The decision has been made. But the execution hangs suspended in time, hostage to geography and the stamina of a horse. Now travel forward to the First and Second World Wars. A general in Paris or London seeking to redirect troops along the Somme or coordinate a flanking maneuver near Monte Cassino relies on morse telegraph, field radio, or a dispatch rider weaving through artillery smoke. Communications still take hours. The delay between the awareness of a battlefield event and the authority to respond is measured not in milliseconds but in the anguish of waiting.
Now arrive at 2026. In the White House Situation Room, an AI system has already processed satellite feeds, parsed the electromagnetic signature of a distant naval movement, cross-referenced historical pattern libraries, and presented the Commander-in-Chief with a ranked menu of response options before the coffee in the briefing cup has had time to cool. In January 2026, the capture of a Venezuelan head of state was coordinated with AI-assisted intelligence tools operating at near-instantaneous speed. In February 2026, when the United States launched Operation Epic Fury against Iran, the Maven AI system — a Palantir-powered platform embedded across all U.S. combatant commands — enabled twenty soldiers to accomplish the target-identification workload that required two thousand intelligence analysts during the 2003 invasion of Iraq.
That transformation — from weeks to milliseconds, from battalions of analysts to algorithms, from deliberation to instantaneous recommendation — is what this paper names Decision Compression. Decision Compression is not merely about speed. It is about the structural collapse of the entire decision lifecycle: the time it takes to observe, analyze, recommend, approve, execute, and adapt. Artificial intelligence is compressing every ring of that cycle simultaneously, and the consequences — strategic, economic, political, and deeply human — are only beginning to be understood.
This paper moves through six interconnected movements. It begins with a historical account of how long decisions used to take — from emperors to generals to regulators — and then examines the internet era’s first wave of disruption, where companies that failed to compress their decision cycles fast enough were overtaken by faster, leaner competitors. It then turns to the mechanics of how AI compresses every phase of decision-making, the collapse of time into autonomous loops, the implications of this compression for power, fragility, and human agency, and finally what durable lessons we can extract from this accelerating pattern. The argument throughout is that Decision Compression is not a technical feature. It is a civilizational shift.
Section 1: A Brief History of How Long It Took To Decide
1.1 Emperors Had Weeks
The first great constraint on human power was not intelligence or will but the speed of communication. In the Han Dynasty, a military commander dispatched to the western frontier operated, for practical purposes, as a sovereign in his own right. Imperial edicts took weeks to arrive. By the time the Emperor’s will was known, battles had been won or lost, alliances formed or shattered. The decision cycle of imperial governance was measured in seasons. Julius Caesar, commanding his legions in Gaul, held near-total decision authority by necessity — not because Rome trusted him, but because Rome could not reach him in time to direct him otherwise.
What this meant, in practice, was that the margin for strategic error was enormous. A misjudgment at the frontier had weeks to compound before correction could arrive. The decision-making apparatus of ancient empires was not a system of fast commands cascading from center to periphery. It was a slow, probabilistic relay of authority through layers of human intermediaries, each adding delay, interpretation, and political friction.
1.2 Generals Had Days
By the time of the Napoleonic Wars, the compression had begun — but only modestly. The electric telegraph, introduced in the 1830s and widely deployed by the American Civil War, began to shrink the decision cycle for military commanders from weeks to days. Napoleon himself, operating before the telegraph, was famous for personally riding to different parts of the battlefield to maintain situational awareness — a recognition that in the absence of rapid communication, the commanding intelligence had to physically co-locate with the information.
By World War I, field telephones and wireless radio had reduced the transmission lag to hours, but the processing and interpretation of that information still occurred in the minds of exhausted human commanders working by candlelight in underground bunkers. The fog of war was not primarily atmospheric. It was temporal. Decisions lagged reality because the tools for observation and communication were slow. By World War II, radar gave commanders minutes of warning. Codebreaking at Bletchley Park allowed Allied strategists to anticipate German movements with unprecedented foresight — but the human analysis of intercepts still took time, and the institutional machinery of approval and deployment remained glacial by today’s standards.
1.3 CEOs Had Weeks
In the boardrooms of the mid-twentieth century, the decision cycle of a major corporation was measured in quarterly rhythms. Market research was conducted, tabulated, and presented over weeks. Consumer sentiment was assessed through surveys that took months to design, administer, and analyze. A product decision made by the CEO of General Motors or IBM in January might not reach full implementation before summer. The competitive environment changed slowly enough that this pace was not fatal. The decision cycle matched the speed of the market.
This equilibrium began to crack in the 1980s and shattered in the 1990s. The personal computer, the fax machine, and then the internet accelerated both the generation and transmission of market information. A CEO who in 1985 could afford to spend six weeks deliberating on a strategic pivot found in 1998 that six weeks was an eternity — long enough for a nimbler competitor to define the category, capture the customer, and build the switching costs that would make displacement nearly impossible.
1.4 Regulators Had Months
Regulatory institutions are, by design and by legal necessity, slow. The notice-and-comment rulemaking process mandated by the Administrative Procedure Act can take years from proposed rule to final implementation. Environmental impact assessments, antitrust reviews, pharmaceutical approvals — all operate on timelines of months to years, built around the assumption that careful deliberation, public input, and inter-agency coordination are worth the time they require. This was a reasonable assumption in a world where the objects of regulation also moved slowly.
Artificial intelligence has broken this calibration. A regulatory framework designed for a technology that evolves on annual timescales is structurally incapable of governing one that evolves on monthly ones. The European Union’s AI Act, the most comprehensive regulatory effort to date, took more than three years from initial proposal to entry into force — a period during which the AI landscape transformed multiple times. By the time the rules were written, the technology they described had largely moved on. This is not a failure of regulatory intent. It is a failure of regulatory tempo.
1.5 Legislators Had Years
The legislative process is the slowest decision machine in democratic governance. A bill introduced in the United States Congress faces years before passage, assuming it passes at all. The design of this slowness is deliberate — it was intended to prevent hasty, poorly considered law from being imposed on a diverse society before consensus could be built. But it also means that legislative responses to technological disruption consistently arrive after the disruption has already reshaped the social and economic terrain.
Today, in 2026, legislators are debating AI governance frameworks while AI systems are already embedded in the targeting chains of active military operations, in the credit decisions affecting millions of consumers, and in the content moderation systems that shape what billions of people see and believe. The legislative decision cycle has not compressed. The technology it seeks to govern has compressed dramatically. The gap between the speed of the governing and the speed of the governed is the defining governance crisis of our era.
Section 2: The Internet Era — When Companies Failed to Compress
The internet age introduced the first systematic evidence that Decision Compression is not merely a military or governmental phenomenon. It is the central competitive dynamic of the digital economy. Company after company, dominant in its moment, failed not because it lacked intelligence or resources, but because it could not compress its decision cycles fast enough to match the velocity of the market around it. The following cases are not merely business school anecdotes. They are proof-of-concept demonstrations that in the age of accelerating information, the speed of decision is the proximate cause of survival or extinction.
2.1 AOL and the Dial-Up Paradox: When Infrastructure Became a Prison
In the early 1990s, America Online was not merely an internet service provider. It was the internet, in the minds of millions of American households. By 1999, AOL had more than twenty-five million subscribers. Its CDs, mailed in such quantities that they became a cultural artifact, were the on-ramp to the digital world for an entire generation. But it had built its entire commercial architecture around a technology — narrowband dial-up access over telephone lines — that was already being made obsolete.
When broadband arrived, first through EarthLink (founded in Pasadena, California, in 1994) and later through cable providers and DSL infrastructure, AOL’s response was structurally compromised by its own installed base. The decision to pivot to broadband required destroying the economics of a dial-up business that was still generating enormous revenue. The executives who might have compressed that decision cycle were operating within an organization whose incentives, contracts, and investor expectations all argued for delay. By the time AOL merged with Time Warner in 2000 in what became one of the most catastrophic corporate combinations in history, the compression gap was already fatal. The lesson is not that AOL was unintelligent. The lesson is that its decision architecture could not compress fast enough to escape the gravity of its own legacy.
2.2 BlackBerry: The Keyboard That Became a Tombstone
Research In Motion’s BlackBerry was not simply a popular device. In the early 2000s, it was infrastructure. Governments, law firms, and investment banks depended on it with near-religious intensity. At its peak in 2009, RIM held over 20% of the global smartphone market. The decision failure that followed is among the most instructive in corporate history. When Steve Jobs unveiled the iPhone in January 2007, RIM’s co-CEO Mike Lazaridis was reportedly unimpressed, questioning how Apple could possibly build a device that lasted more than a few hours on a battery. He was right about the technical constraint. He was catastrophically wrong about what customers would decide to accept.
There’s no chance that the iPhone is going to get any significant market share. No chance.
— Steve Ballmer, CEO of Microsoft, January 2007 [1]
Within three years, the iPhone had redefined what a smartphone was. Android followed in 2008. RIM, facing a decision about whether to abandon its keyboard-centric architecture and its enterprise-first philosophy, chose to iterate incrementally rather than compress the decision cycle dramatically. It released touchscreen devices that felt like compromises, delayed its platform overhaul, and watched its market share collapse from over 20% to under 1% within five years. That delay was not ignorance. It was organizational decision friction.
2.3 Intel and the Microprocessor Throne: Disrupted from Below
Intel’s Pentium microprocessor, introduced in 1993, was the engine of the personal computing revolution. Through the 1990s and 2000s, Intel’s x86 architecture was so deeply embedded in the PC ecosystem — in software compatibility, in developer toolchains, in manufacturing partnerships — that the company’s dominance seemed structural rather than contingent. But the decisions that would shape Intel’s future were being made elsewhere, by different companies, around different architectures, for different use cases.
Apple’s decision in 2020 to design its own M-series silicon demonstrated that the vertical integration of hardware and software could produce performance and efficiency gains that Intel’s roadmap could not match. More consequentially, the emergence of GPU-accelerated computing — pioneered by Nvidia, founded in 1993 and initially focused on graphics processing for gaming — revealed that the dominant computational architecture for artificial intelligence was not the CPU that Intel had perfected, but the massively parallel GPU. Nvidia’s compressed decision to pivot its CUDA platform toward machine learning, made years before the AI boom was commercially obvious, proved to be among the most consequential strategic calls in technology history.
By Q1 fiscal year 2027 (February–April 2026), Nvidia reported record revenue of $81.6 billion — up 85% year over year — driven by Data Center revenue of $75.2 billion. [2] Jensen Huang, Nvidia’s founder and CEO, characterized this moment with characteristic sweep:
The buildout of AI factories — the largest infrastructure expansion in human history — is accelerating at extraordinary speed.
— Jensen Huang, CEO of NVIDIA, May 20, 2026 [3]
Intel, which had every resource and nearly every advantage in the semiconductor industry, had not made Nvidia’s decision. The compression gap — between recognizing a structural shift in computational demand and reorganizing an entire company’s product strategy to meet it — was the difference between a $5.4 trillion market capitalization and a much humbler trajectory.
2.4 Yahoo, Google, and the Search for Decision Speed
The story of Yahoo’s displacement by Google is sometimes told as a story of superior algorithm design. But the deeper story is about decision velocity. Yahoo in the late 1990s was a portal company — a directory, an email provider, a news aggregator. Search was one feature among many. When Google made the decision to make search the singular obsession of its product and engineering organization, Yahoo chose to remain a destination. That decision, made slowly and by consensus across a vast organization, compounded over years into an insurmountable gap. By the time Yahoo recognized the full strategic importance of search, Google had already built the advertising infrastructure — AdWords, AdSense — that turned search intent into a printing press for revenue. Yahoo eventually offered to acquire Google in its early years for one million dollars. Google declined. Yahoo later rejected a $44.6 billion acquisition offer from Microsoft in 2008 and was eventually sold to Verizon for approximately $4.5 billion in 2016.
2.5 Internet Explorer and the Browser Wars
At its peak in the early 2000s, Microsoft’s Internet Explorer held approximately 90% of the global browser market. It came pre-installed on every Windows PC, and bundling tied the browser’s release cycle to the operating system’s — meaning browser updates happened on a timeline of years, not months. Google Chrome, launched in 2008, was built around a different decision architecture: rapid iteration, independent release cycles, and a focus on speed above all else. Chrome’s decision cycle was weeks. IE’s was years. By 2012, Chrome had surpassed IE in global market share. By 2016, IE was in structural decline. Microsoft eventually rebuilt its browser entirely on the Chromium engine — an acknowledgment that the decision architecture of the legacy product was not reformable.
2.6 Webvan, Amazon Fresh, and the Grocery Delivery Misjudgment
Webvan was among the most ambitious failures of the first dot-com era. Founded in 1996, it raised over $375 million in its IPO and committed to capital-intensive warehouse infrastructure and proprietary logistics software before establishing whether customers would actually change their grocery shopping behavior at scale. The decision to build first and validate second proved fatal. Webvan filed for bankruptcy in 2001, having spent over $800 million. The same market was eventually captured by Amazon Fresh and Instacart, companies that built on a foundation of validated consumer behavior — accelerated dramatically by COVID-19 — and data-driven decision loops that allowed continuous refinement of the customer experience. The difference between Webvan and Amazon Fresh is not vision. Both saw the same future. The difference is the sequencing and speed of the decisions that built toward that future.
2.7 Postmates, DoorDash, and the Pandemic Compression
Postmates launched in 2011 as a pioneer of urban on-demand delivery. It created a market, trained consumers on the behavior, and built a courier network in major cities. But by 2020, it was DoorDash — better funded, more aggressive in suburban expansion, and more sophisticated in its algorithm-driven matching of supply and demand — that had compressed its decision cycles sufficiently to dominate the COVID-19 delivery surge. When restaurants closed their dining rooms in March 2020, the company best positioned to capture that demand was the one whose pricing algorithms, driver allocation, and restaurant partner onboarding could respond in real time. DoorDash went public in December 2020 at a valuation of $39 billion. Postmates was acquired by Uber for $2.65 billion the same year. The delta between those outcomes traces directly to the differential speed of organizational decision-making.
2.8 Skype, Zoom, and the Conference Call That Changed Everything
Skype was, for much of the 2000s and early 2010s, synonymous with video calling. But its acquisition by Microsoft in 2011 embedded it in an organization whose decision cycles were calibrated to enterprise software timescales. When COVID-19 arrived and the global economy moved to remote work within a matter of weeks, Zoom — founded in 2011 by former Cisco WebEx engineer Eric Yuan — was positioned to absorb the demand. Zoom’s single-minded focus on ease of use, its decision to optimize for reliability at scale, and its freemium model had been compression decisions made years earlier. In Q1 2020, Zoom’s revenue grew 169% year-over-year. By the end of 2020, its market capitalization exceeded $100 billion. Skype’s market share collapsed. The pandemic was not the cause of Zoom’s victory. It was the accelerant that revealed a compression gap that already existed.
2.9 MySpace to Facebook to TikTok: The Compression Cascade
The trajectory from MySpace to Facebook to TikTok is a story of iterating decision cycles at accelerating velocity. MySpace was outcompeted by Facebook because Facebook made faster, better-calibrated decisions about the user experience, the identity layer, and the developer ecosystem. Facebook’s News Feed, introduced in 2006, was a decision made against the loud objections of users — a decision that Facebook compressed through and implemented anyway, correctly recognizing that behavioral engagement would override initial resistance.
TikTok’s displacement of Facebook among younger demographics tells a similar story at higher velocity. ByteDance’s recommendation algorithm — trained on behavioral signals at a scale that its competitors did not initially match — made the decision of what to show each user faster, more accurately, and more compulsively than Facebook’s News Feed. The product itself was a decision-compression engine. Every swipe was a data point. Every pause was a signal. The algorithm refined its understanding of individual preference in near-real time, creating a feedback loop structurally faster than any human editorial decision could be.
2.10 MSN Messenger, Slack, and the Corporate Chat Revolution
Microsoft’s messaging products — MSN Messenger, Windows Live Messenger, Skype for Business — underwent so many rebranding cycles that the confusion itself became an organizational biography. Each rebrand was a decision to respond to competitive pressure without making the harder structural decision to build a category-defining product from scratch. Slack, founded in 2013, made one decision clearly and well: to build a messaging platform designed specifically for team collaboration, with threading, integrations, and search calibrated to the cognitive workflows of knowledge workers. That clarity of purpose, maintained through rapid product iteration, allowed Slack to grow from zero to a $7.1 billion acquisition by Salesforce in 2020, despite Microsoft’s enormous distribution advantage.
2.11 ChatGPT, Google Gemini, and the Anthropic Ascent
The AI application market offers perhaps the most compressed version of the disruption cycle yet observed. OpenAI launched ChatGPT on November 30, 2022. Within two months, it had reached 100 million users — the fastest consumer application adoption in history to that point. Google, whose researchers had invented the transformer architecture on which ChatGPT was built, found itself in the position of the disrupted incumbent, forced to accelerate its own AI product timelines dramatically. Its response — Bard, later renamed Gemini — was a compressed decision to ship a product before it was fully ready, leading to a factual error in the first public demonstration that briefly wiped $100 billion from Alphabet’s market capitalization.
Anthropic, founded in 2021 by Dario Amodei, his sister Daniela Amodei, and a cohort of researchers who left OpenAI over concerns about safety and organizational direction, pursued a different compression strategy. Rather than racing to the consumer market, Anthropic compressed its enterprise decision cycle — building Claude as the AI of choice for Fortune 500 companies across banking, healthcare, legal, and technology sectors. The results, as of June 2026, are extraordinary: Anthropic reported Q1 2026 revenue of $4.8 billion and projected Q2 2026 revenue of $10.9 billion — more than doubling in a single quarter and exceeding its entire 2025 annual revenue in three months. [4] On June 1, 2026, Anthropic filed confidentially for an IPO at a valuation of $965 billion, making it the most valuable private AI company in the world, surpassing OpenAI’s reported $852 billion valuation. [5]
I’m deeply uncomfortable with these decisions being made by a few companies, by a few people.
— Dario Amodei, CEO of Anthropic, CBS 60 Minutes, November 2025 [6]
The Anthropic story is the most recent and most dramatic iteration of the pattern this paper describes. In the AI application market, the compression cycle has accelerated to the point where a company founded in 2021 can reach a near-trillion-dollar valuation by 2026 while a technology giant founded in the 1970s scrambles to respond. The gap between the fastest and the slowest decision-makers has become so large that it no longer fits within the vocabulary of conventional competitive strategy. It requires a new vocabulary. That vocabulary is Decision Compression.
Section 3: The Mechanics of Decision Compression
To understand why Decision Compression is structurally different from ordinary speed improvement, it is necessary to examine the anatomy of a decision cycle. Every significant decision — whether military, commercial, regulatory, or political — moves through a recognizable sequence of phases: observation, analysis, recommendation, approval, execution, and adaptation. In a pre-AI world, each phase was bounded by human cognitive limits and institutional friction. AI does not merely accelerate individual phases. It simultaneously compresses all of them, eliminates many of the handoffs between them, and in some cases merges them into a single continuous loop.
3.1 Observation Cycles: From Weeks to Instantaneous
The first phase of any decision is observation — the gathering of relevant information about the state of the world. For most of human history, this was a bottleneck of physical presence. You had to be there, or have someone there, to observe what was happening. Satellites expanded the scope of observation dramatically — but the imagery they produced still required human analysts to interpret, and the interpretation process was slow.
AI-powered observation systems have collapsed this cycle to near-zero latency. Computer vision systems can process satellite imagery in seconds, identifying military movements, infrastructure changes, crop patterns, or crowd density with accuracy that exceeds trained human analysts in controlled conditions. Natural language processing systems continuously monitor social media, news feeds, financial filings, and corporate communications in real time, surfacing signals that would take human teams weeks to synthesize. The observation cycle, which once took days or weeks to complete, now completes in seconds — and does so continuously, without fatigue, across every relevant domain simultaneously.
3.2 Analysis Cycles: From Battalions to Algorithms
The observation-to-analysis gap has historically been among the most significant bottlenecks in decision-making. The CIA’s Directorate of Intelligence employs thousands of analysts whose primary function is to transform raw intelligence into actionable insight. Financial institutions employ armies of research analysts to make sense of market data. Each of these human analytical processes is subject to cognitive limits, political pressures, and the irreducible constraint that human attention can only be focused in one direction at a time. The Maven AI system used by the U.S. military in Operation Epic Fury allowed twenty soldiers to accomplish the analytical workload of two thousand. [7] This is not a metaphor. It is a documented operational fact, and it represents a compression ratio of 100:1 in the analysis cycle.
MIT Professor Eric So, who launched the AI in Financial Markets and Decision-Making research group at MIT’s Initiative on the Digital Economy in October 2025, has noted that AI can automate tasks that formerly required enormous human effort — but that the compression of analytical cycles raises equally significant questions about whether the diversity of analytical perspectives, which historically stabilized markets and institutions, is being dangerously narrowed. [8]
3.3 Recommendation Cycles: From Committees to Ranked Options
In a traditional organizational decision process, the move from analysis to recommendation involves a committee, a working group, a senior review — institutionalized forms of deliberative synthesis. These structures exist to aggregate expertise, surface dissenting views, and build organizational consensus needed for implementation. They are also slow. A strategic recommendation that has moved through a consulting engagement, a board committee review, and a senior management discussion has been refined and socialized — but it has also been delayed by months.
AI recommendation systems compress this cycle to near-instantaneous output. A large language model given relevant context and a well-specified decision problem can generate a ranked set of options, with supporting analysis and risk assessment, in seconds. In the military context, Maven generates target prioritization recommendations faster than any human review committee could convene. In the financial context, algorithmic trading systems generate and execute buy/sell recommendations within microseconds of observing a market signal. In the corporate context, AI-powered strategy tools are beginning to synthesize competitive intelligence, financial projections, and scenario analyses into board-ready recommendations that previously took consulting firms months to prepare.
3.4 Approval Cycles: The Human in the Loop and the Illusion of Control
The approval cycle — the moment at which human authority formally ratifies or rejects a recommendation before execution — is the phase most directly implicated in debates about AI safety and accountability. In theory, the human in the loop represents an irreducible check on AI system outputs. In practice, as the Al Habtoor Research Centre observed in March 2026, when a human commander has only minutes to review an AI-generated strike recommendation backed by synthesized data they could not have assembled themselves, the decision is less an independent judgment and more an approval. [9] The compression of all preceding cycles has structurally pre-loaded the approval.
This is Decision Compression’s most philosophically significant implication. The human approval does not introduce the margin that separates thoughtful deliberation from rubber-stamping. When the observation, analysis, and recommendation cycles have been completed by AI in seconds, and when the human approver has neither the time nor the independent informational resources to meaningfully contest the recommendation, the approval cycle has been functionally collapsed into the recommendation cycle. The human remains formally in the loop. Whether the human is substantively in the loop is a different question entirely.
3.5 Execution Cycles: From Orders to Autonomous Action
Execution — the translation of an approved decision into physical or digital action — has historically been a separate cycle with its own latency. A military order, once approved, still had to be transmitted to units, interpreted by commanders, and implemented through coordinated movement. AI-enabled execution has compressed this cycle into microseconds in domains where the execution is digital, and into minutes or hours in domains where physical action is involved. In financial markets, approximately 70% of U.S. equity trading volume was executed by AI algorithmic systems as of 2021, with that percentage increasing annually. [10] In military operations, autonomous drone systems operating in GPS-denied, signal-jammed environments — such as the LUCAS drone deployed in the Iran conflict of 2026 — execute attack missions without requiring real-time human command inputs. [11] The execution cycle, once a bottleneck of human coordination, is being absorbed into the automated loop of the AI system itself.
3.6 Adaptation Cycles: Continuous Learning Without Rest
The final phase of any decision cycle is adaptation — the assessment of outcomes and the incorporation of lessons into future decisions. In pre-AI organizations, this happened slowly: through after-action reviews, quarterly business reviews, annual strategic planning processes. The adaptation cycle was measured in months or years. Machine learning systems adapt continuously and automatically, incorporating outcome feedback into their next recommendation within the same operational window. A trading algorithm that executes a losing trade at 9:01 AM has already incorporated that outcome into its model by 9:01:01 AM. The adaptation cycle has been compressed from months to milliseconds, closing the loop between decision and learning in ways that fundamentally change the nature of organizational intelligence.
Section 4: AI Collapses Time — Minutes, Seconds, and Autonomous Loops
Having examined how AI compresses each individual phase of the decision cycle, it is necessary to step back and observe what happens when all six phases are compressed simultaneously. The result is not merely a faster version of the same decision process. It is a qualitatively different phenomenon: the collapse of the decision cycle into a continuous autonomous loop that operates at a speed and scale entirely beyond human cognitive reach.
4.1 The Compression to Minutes: Strategic-Level AI
The first tier of Decision Compression — the compression of strategic decisions from weeks to minutes — is already operational. When a Chinese naval vessel moves into unexpected proximity to the Long Beach seaport, the AI systems embedded in U.S. military and intelligence infrastructure do not wait for a scheduled briefing. They begin immediately: processing satellite imagery, correlating historical naval movement patterns, running escalation scenario models, assessing probable intent, drafting military response options, calculating oil market implications, and alerting allied governments. This entire process — which in the Cold War era would have taken days of inter-agency deliberation — completes in minutes. [12]
In the corporate domain, strategic decisions that once required months of market research, competitive analysis, and financial modeling are being compressed by AI systems that continuously monitor competitive positioning, synthesize earnings calls and regulatory filings, model scenario outcomes, and present strategic options in real time. The CEO of a major financial institution in 2026 has access to a continuous stream of AI-synthesized strategic intelligence that no team of human analysts could have produced on any timescale consistent with its strategic relevance.
4.2 The Compression to Seconds: Operational and Tactical AI
The second tier — compression to seconds — is the domain of operational and tactical AI. High-frequency trading systems execute thousands of trades per second based on market signals that humans cannot perceive in the time available. [13] AI-powered cybersecurity systems detect, classify, and begin responding to intrusion attempts in seconds. In military operations, AI-powered sensor fusion systems integrate radar, acoustic, infrared, and signals intelligence into a unified operational picture faster than any human analyst can read a single report.
It is at this tier that the alignment between human intention and AI execution becomes most critical and most difficult to maintain. When an AI system is executing thousands of decisions per second, any systematic error in its objective function or training data propagates not as a single mistake but as a cascade. The 2010 Flash Crash, in which AI trading systems interacting without adequate circuit breakers caused the Dow Jones Industrial Average to fall nearly 1,000 points in minutes before partially recovering, demonstrated the systemic fragility that emerges when compressed decision loops interact without sufficient human oversight infrastructure.
4.3 The Compression to Autonomous Loops: Beyond Human Oversight
The third and most consequential tier of Decision Compression is the autonomous loop — a decision cycle that operates continuously without requiring human initiation, approval, or oversight at the individual decision level. Algorithmic trading systems, content recommendation engines, autonomous weapons systems operating within pre-defined rules of engagement, and AI-powered supply chain optimization systems all operate in forms of autonomous loops. The Stanford HAI 2026 AI Index Report, released April 14, 2026, is unsparing in its assessment: compiled by a steering committee chaired by Yolanda Gil of the University of Southern California, the 423-page report found that AI now scales faster than the institutions built to govern it. [14] Generative AI reached 53% population adoption in three years — faster than the personal computer or the internet. [15] The gap between what AI systems can do and how prepared institutions are to manage them is widening, not narrowing.
The data reveals a field scaling faster than the systems around it can adapt.
— Yolanda Gil & Raymond Perrault, Stanford HAI 2026 AI Index Report, April 2026 [16]
In the autonomous loop, the human decision-maker is not in the loop in any meaningful operational sense. The human has made a prior decision — to deploy the system, to set its parameters, to define its objectives — and the system executes continuously within that prior authorization. This is the architecture of modern financial markets, modern content platforms, and, increasingly, modern military operations. It raises a question that no technical advancement can resolve: when autonomous loops make consequential decisions at superhuman speed, where does accountability reside, and what governance structures are adequate to the task?
Section 5: Implications — Strategic Advantage, Systemic Fragility, and Human Irrelevance
5.1 Strategic Advantage: The Compression Premium
The most immediate implication of Decision Compression is competitive: those who compress faster win. This is true across military, commercial, and political domains. In military terms, the United States’ deployment of AI-powered targeting and logistics systems in Operation Epic Fury allowed it to strike 1,000 targets in the first 24 hours of the conflict and 5,000 targets over 10 days — a tempo of action that no adversary relying on conventional intelligence and targeting processes could match. [17]
In commercial terms, the Anthropic ascent from a 2021 startup to a near-trillion-dollar entity by June 2026 represents the most compressed value creation cycle in corporate history. The IMF, in its April 2026 assessment of AI’s macrofinancial implications, projected that AI will boost global GDP by approximately 0.5% annually between 2025 and 2030, with high-adoption scenarios potentially delivering up to 1.8% global TFP growth within five years. [18] But these aggregate numbers mask a critical distributional reality: the strategic advantage from Decision Compression accrues overwhelmingly to entities that have already achieved compressed decision cycles, widening the gap between them and those who have not.
5.2 Systemic Fragility: The Flash Crash Problem, Generalized
The same compression that creates strategic advantage also creates systemic fragility. When decision cycles are measured in months or years, errors propagate slowly. There is time to observe an error, identify its cause, and intervene before it cascades. When decision cycles are measured in milliseconds, errors propagate at the speed of the system itself — faster than any human institution can observe, diagnose, and correct. The Stanford 2026 AI Index reported 362 documented harmful AI incidents in 2025 — a 56% increase over 2024. [19] More troubling than the count is the pattern: the incidents are growing in severity and systemic scope as AI systems become more deeply embedded in critical infrastructure. The Al Habtoor Research Centre noted that prior AI targeting systems demonstrated error rates of approximately ten percent in adversarial environments — meaning that in a campaign targeting 5,000 targets, hundreds may have been flagged in error. [20]
The financial system offers the clearest laboratory for observing this fragility dynamic. As Matthew Lyberg, global head of AI at Manulife Wealth & Asset Management, observed in early 2026, when interpretive diversity narrows — whether through uniform regulation or algorithmic similarity — systemic fragility increases. When multiple AI systems trained on similar data, optimized for similar objectives, and operating on similar architectures make correlated decisions at millisecond speed, the conditions for a cascade failure are structurally present. The 2010 Flash Crash was a preview. The next version may be harder to reverse.
5.3 Human Irrelevance at Certain Speeds: The Agency Question
The deepest implication of Decision Compression is philosophical, and it was anticipated with unusual clarity by Henry Kissinger, who spent the final years of his life thinking seriously about what AI means for human agency and governance. In his final book, Genesis: Artificial Intelligence, Hope, and the Human Spirit, co-authored with Eric Schmidt and Craig Mundie, Kissinger raised the question of what it means for human society when decisions that shape the lives of millions are made by systems whose reasoning cannot be fully explained, at speeds that preclude meaningful human review.
What will become of human consciousness if its own explanatory power is surpassed by AI, and societies are no longer able to interpret the world they inhabit in terms that are meaningful to them?
— Henry Kissinger, ‘The Age of AI and Our Human Future’, 2021 [21]
This is not a rhetorical question. It is a structural one. When the AI systems embedded in financial markets make millions of decisions per second that collectively determine whether pension funds grow or shrink, whether credit is available to small businesses, or whether commodity prices spike in ways that affect food security in developing nations — and when no human being can observe, understand, or intervene in those decisions in real time — the category of ‘human irrelevance at certain speeds’ has moved from philosophical abstraction to operational reality.
Stanford’s Fei-Fei Li, Sequoia Professor of Computer Science and co-director of Stanford’s Human-Centered AI Institute, has argued consistently that the answer to this challenge is not to slow AI development but to ensure that human agency is embedded in the design of AI systems from the beginning. In Senate testimony and in her public writing, she has maintained: “While AI, like most technologies, promises to solve many problems for the common good, it can also be misused to cause harm. It falls upon the U.S. government to spearhead the ethical procurement and deployment of these systems.” [22]
The governance gap is not merely a problem of regulatory speed, though it is that too. It is a problem of institutional legitimacy. The Stanford 2026 AI Index found that among surveyed countries, the United States showed the lowest trust in its own government’s ability to regulate AI, and that globally, trust in institutions to manage AI effectively is declining even as AI adoption accelerates. A technology that is compressing every decision cycle in the economy is simultaneously compressing the time available for the democratic deliberation that gives governance its legitimacy.
Section 6: What Have We Learned? Five Pillars for The Compressed Age
The historical arc traced in this paper — from imperial horse-riders to autonomous kill chains, from dial-up modems to near-trillion-dollar AI companies founded five years ago — yields not a set of prescriptions but a set of durable structural insights. These are the pillars that any serious engagement with the age of Decision Compression must rest upon. They apply to military commanders, corporate strategists, regulators, legislators, and citizens in equal measure.
Pillar 1: Speed Without Wisdom Is Not an Advantage — It Is a Liability
The first and most counterintuitive lesson of the compressed age is that acceleration without epistemic discipline is not a competitive advantage — it is a mechanism for rapidly scaling mistakes. Every one of the corporate failures documented in Section 2 of this paper involved a form of decision speed. AOL moved quickly to sign dial-up subscribers at a time when broadband was emerging. Webvan moved quickly to build warehouse infrastructure before validating customer behavior. BlackBerry moved quickly to defend its keyboard architecture at the moment when the market was moving decisively toward touchscreen. In each case, the speed of the wrong decision was part of what made it catastrophic.
AI does not automatically produce the right decision faster. It produces decisions faster. If the objective function is miscalibrated, if the training data embeds historical bias, if the system is optimizing for a proxy metric that diverges from the actual goal — the errors propagate at the same speed as the successes. The wisdom question — what are we actually trying to optimize for, and are we measuring the right things — remains irreducibly human. Organizations and governments that mistake speed for intelligence will use Decision Compression to scale their errors with unprecedented efficiency.
Pillar 2: The Compression Gap Is the New Competitive Moat
The differential between compressed and uncompressed decision cycles has become the primary source of competitive advantage in the digital economy, and this differential is widening faster than most institutions recognize. The companies that have achieved the greatest compression — not just in product development but in every phase of their operational and strategic decision cycles — are pulling away from their competitors at a rate that conventional competitive strategy cannot explain or address. Dario Amodei’s characterization of Anthropic’s growth trajectory as a “radical acceleration” [23] that is catching all of humanity off guard is not marketing hyperbole. It describes a structural reality: the exponential scaling of AI capability, combined with organizational design choices that allow rapid deployment of that capability, is creating compression moats that are deeper and more defensible than any network effects or brand advantages that defined competitive moats in the pre-AI era.
For companies seeking to build compression moats, the lesson is that the investment must be made before the competitive pressure is obvious. Nvidia’s decision to build the CUDA parallel computing platform, made when machine learning was still primarily an academic discipline, is the archetype. The compression moat must be built in the absence of competitive urgency, because by the time the urgency is apparent, the moat has already been built by someone else.
Pillar 3: Institutional Governance Must Compress or Become Irrelevant
The third pillar addresses the most significant governance challenge of our era: democratic institutions, regulatory bodies, and legislative processes designed for a world of slow decisions are becoming structurally incapable of governing a world of fast ones. Julian Nyarko, a law professor and Stanford HAI Associate Director, observed as 2025 drew to a close that firms and courts must stop asking ‘Can it write?’ and instead start asking ‘How well, on what, and at what risk?’ [24] This is a compression challenge: the relevant questions are becoming more specific, more technical, and more time-sensitive faster than legal and regulatory institutions can adapt their frameworks to address them.
The UN resolution on artificial intelligence in the military domain, passed in December 2025, is a step toward compressing international governance frameworks — but a small one, given that the technology it seeks to address was already operational in an active conflict before the resolution’s first multilateral discussion was even scheduled. [25] Governance compression does not mean governance speed-for-its-own-sake. It means redesigning institutional decision processes so that the lag between technological deployment and institutional response is measured in months rather than decades. That will require new regulatory architectures, new legislative tools, and a fundamental reconception of what democratic deliberation looks like when the objects of deliberation are moving at machine speed.
Pillar 4: The Human Must Be Genuinely in the Loop, Not Ceremonially In It
The fourth pillar addresses the most dangerous form of the human-oversight failure: the scenario in which a human is formally present in an approval process but has been structurally stripped of any meaningful capacity to exercise independent judgment. The Al Habtoor Research Centre’s analysis of AI in military targeting captured this precisely: when a human commander has only minutes to review a recommendation backed by data they could not have assembled independently, their decision is less an independent judgment and more an approval.
Genuine human oversight of compressed decision systems requires three things that are rarely present simultaneously. First, the human must have access to information independent of the AI system’s synthesis — a second channel that preserves the possibility of divergent assessment. Second, the human must have the authority and organizational support to override the AI recommendation without penalty, even when the recommendation appears well-supported. Third, the human must have enough time to exercise genuine deliberation — which means that the design of the compressed decision system must include deliberate slowdowns at the approval stage that are treated not as failures of efficiency but as features of accountability.
Pillar 5: Decision Compression Is a Civilizational Choice, Not a Technological Inevitability
The fifth and final pillar is the most important, and the most difficult to hold onto in the face of competitive pressure. Decision Compression is not happening to us. It is being chosen — by companies that deploy AI systems, by governments that fund AI research and procurement, by consumers who adopt AI-powered products, and by investors who allocate capital toward the companies building compressed decision infrastructure. At every point in this chain, human beings are making choices. The compression is not a natural force. It is the aggregate of human decisions.
This means that the trajectory of Decision Compression is not fixed. The pace of compression in military targeting, in financial markets, in content recommendation, in corporate strategy — all of these are subject to design choices, regulatory frameworks, procurement standards, and social norms that human institutions can shape. Kissinger was right that this is the most difficult question of the age. But Fei-Fei Li is also right that we should not give up our agency. The compression can be disciplined without being stopped. It can be directed toward human flourishing rather than human irrelevance. But that requires treating Decision Compression as what it is: a civilizational choice that deserves the full weight of democratic deliberation, even if — especially if — that deliberation must itself become faster.
Conclusion: Why I Named This Paper Decision Compression
I named this paper Decision Compression because the phrase captures something that no alternative formulation quite achieves. It is not about AI speed alone, because speed is a feature of many technologies that have not transformed the structure of power as profoundly as AI is now doing. It is not about AI intelligence, because intelligence without the compression of decision cycles does not produce the competitive, military, and political consequences this paper has described. It is the compression of the entire decision cycle — observation, analysis, recommendation, approval, execution, and adaptation — simultaneously and in interaction, that defines the transformation we are living through.
The historical argument of this paper is that Decision Compression has always been a source of advantage. Empires that could communicate faster across their territories held strategic advantages. Generals who could receive and act on battlefield intelligence faster won battles their slower-moving adversaries lost. Companies that could iterate their products and strategies faster captured markets that incumbents could not defend. But the compression has always had limits imposed by human cognitive capacity and institutional friction. AI has removed those limits. The compression now approaches the speed of machine processing, and the consequences of that removal are only beginning to be fully understood.
The corporate examples of Section 2 demonstrate that Decision Compression is not merely a theoretical framework — it is a documented causal mechanism that explains the rise and fall of companies across three decades of digital economic history. AOL, BlackBerry, Yahoo, Internet Explorer, Webvan, Postmates, Skype, MySpace, and MSN Messenger all failed for a common reason: their decision cycles were too slow relative to the competitive environment around them. The companies that displaced them — Google, Apple, Amazon, DoorDash, Zoom, TikTok, Slack — all succeeded in part because they had compressed some or all phases of their decision cycles more effectively. Anthropic’s current trajectory, achieving near-trillion-dollar valuation within five years of founding, is the most compressed version of this dynamic yet recorded.
The military argument of this paper demonstrates that Decision Compression is not merely a commercial phenomenon. The 2026 US-Iran conflict — characterized as the first AI war — shows that the compression of military decision cycles to near-real-time speeds is already operational and already consequential. The question of whether the human commander who approves an AI-generated target recommendation has made a genuine decision or performed a bureaucratic ratification of a machine output is not abstract. It is being answered, imperfectly and under fire, in active operations today.
The governance argument demonstrates that the institutions designed to govern human societies — legislatures, regulatory agencies, international bodies — are operating on decision timescales structurally incompatible with the speed at which the technologies they seek to govern are changing. The Stanford 2026 AI Index’s finding that AI scales faster than the institutions built to govern it is perhaps the most consequential empirical observation in contemporary policy discourse. It is not an argument for abandoning governance. It is an argument for compressing governance itself — for redesigning the institutional decision cycles of democratic societies to match the velocity of the technologies that shape them.
What have we learned? We have learned that the advantage of compressed decisions is real, measurable, and growing. We have learned that the compression creates systemic fragility as well as strategic advantage, and that organizations and governments that treat compression as an unambiguous good will eventually pay the price of the errors they scale. We have learned that genuine human agency in compressed decision systems requires deliberate structural design, not ceremonial presence. We have learned that the governance gap between technological capability and institutional response is widening, and that closing it requires compressing governance itself without destroying its democratic character. And we have learned — perhaps most importantly — that Decision Compression is a choice, not a fate.
The Emperor’s horse-rider is gone. The general’s telegraph is long silent. The CEO’s quarterly retreat is becoming an anachronism. What replaces them will be determined not by the technology alone, but by the quality of the human decisions made about how that technology is built, deployed, governed, and constrained. Decision Compression, in the end, is the compression of the interval between problem and response, between awareness and action, between perception and consequence. We are living in that interval now. What we do with it will define the century.
Footnotes and Endnotes
[1] Steve Ballmer, CEO of Microsoft. Quoted in USA Today interview, January 2007, on the iPhone launch. https://usatoday30.usatoday.com/money/companies/management/2007-04-30-ballmer-ceo-forum-usat_N.htm
[2] NVIDIA Corporation (SEC Filing). Form 8-K, Q1 Fiscal Year 2027 Results, May 20, 2026. Revenue $81.6B, up 85% YoY. https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000051/q1fy27pr.htm
[3] Jensen Huang, CEO of NVIDIA. Earnings statement Q1 FY2027, May 20, 2026. As reported by Gulf News and Business Chief. https://businesschief.com/news/nvidia-q1-2026-revenue-exceeds-expectations-amid-ai-boom
[4] The Statesman / Univest. Anthropic IPO filed at $965B valuation; Q1 2026 revenue $4.8B; Q2 projected $10.9B, June 2026. https://www.thestatesman.com/technology/anthropic-ipo-openai-race-valuation-2026-explained-1503600868.html
[5] CNBC. Anthropic tops OpenAI as most valuable AI startup, nears $1 trillion valuation, May 28, 2026. https://www.cnbc.com/2026/05/28/anthropic-open-ai-startup-value.html
[6] Dario Amodei, CEO of Anthropic. CBS 60 Minutes, November 2025. Also cited in Aiifi.ai, ‘Best Dario Amodei Quotes on AI,’ April 2026. https://www.aiifi.ai/post/dario-amodei-quotes
[7] Al Habtoor Research Centre. AI in War: What the Iran War Reveals About the Pentagon’s Algorithms, March 8, 2026. (20 troops vs. 2,000 analysts.) https://www.habtoorresearch.com/programmes/ai-war-pentagon-algorithms/
[8] MIT Initiative on the Digital Economy. New Research Group Examines Whether AI Will Lead to Better Finance Decisions. Profile of Prof. Eric So, MIT Sloan, October 13, 2025. https://ide.mit.edu/insights/new-researcher-examines-whether-ai-will-leads-to-better-finance-decisions/
[9] Al Habtoor Research Centre. AI in War: What the Iran War Reveals About the Pentagon’s Algorithms, March 8, 2026. On approval as ratification. https://www.habtoorresearch.com/programmes/ai-war-pentagon-algorithms/
[10] Michigan Journal of Economics. The Use of AI and AI Algorithms in Financial Markets. 70% U.S. equity trading volume via AI algorithms (2021). https://sites.lsa.umich.edu/mje/2025/03/09/the-use-of-ai-and-ai-algorithms-in-financial-markets/
[11] PlusClouds Blog. Artificial Intelligence in the 2026 Iran War. LUCAS autonomous drone system, March 6, 2026. https://plusclouds.com/us/blogs/2026-iran-savasinda-yapay-zeka-askeri-yapay-zekanin-gelecegi
[12] Chatham House. The Iran War Highlights the Creeping Use of AI in Warfare, March 27, 2026. Maven AI system capabilities. https://www.chathamhouse.org/2026/03/iran-war-highlights-creeping-use-ai-warfare
[13] Financial Modeling Prep / PMC. Artificial Intelligence in Algorithmic Trading — millisecond execution in high-frequency trading (HFT). https://site.financialmodelingprep.com/education/financial-analysis/Artificial-Intelligence-in-Algorithmic-Trading-The-Future-of-Finance
[14] Stanford HAI / ComplexDiscovery. Stanford’s 2026 AI Index Highlights Rapid Growth and Widening Governance Gaps, April 14–24, 2026. https://complexdiscovery.com/stanfords-2026-ai-index-highlights-rapid-growth-and-widening-governance-gaps/
[15] Stanford HAI / i-scoop.eu. 2026 AI Index Report: Generative AI reached 53% population adoption in 3 years, faster than PC or internet. https://www.i-scoop.eu/the-2026-ai-index-report-from-stanford-and-what-it-says-about-ai-right-now/
[16] Yolanda Gil & Raymond Perrault, Stanford HAI. Opening message, 2026 AI Index Report, April 14, 2026 — ‘a field scaling faster than the systems around it can adapt.’ https://complexdiscovery.com/stanfords-2026-ai-index-highlights-rapid-growth-and-widening-governance-gaps/
[17] Wikipedia — AI Warfare. Maven Smart System: 1,000 targets struck in first 24 hours, 5,000 over 10 days, Operation Epic Fury 2026. https://en.wikipedia.org/wiki/AI_warfare
[18] International Monetary Fund (IMF). Global Economic and Financial Implications of Artificial Intelligence, IMF Note, April 2026. GDP +0.5% annually; high-TFP scenario +1.8% in 5 years. https://www.imf.org/-/media/files/publications/imf-notes/2026/english/insea2026002.pdf
[19] Stanford HAI — Responsible AI Chapter. 2026 AI Index Report: 362 documented harmful AI incidents in 2025, up from 233 in 2024. https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-ai
[20] Al Habtoor Research Centre. AI in War: What the Iran War Reveals About the Pentagon’s Algorithms, March 8, 2026. ~10% error rates in prior AI targeting systems. https://www.habtoorresearch.com/programmes/ai-war-pentagon-algorithms/
[21] Henry Kissinger. Essay on AI, 2018, as quoted on PBS Firing Line with Fei-Fei Li, August 15, 2025. Also in ‘The Age of AI and Our Human Future,’ 2021. https://www.pbs.org/video/fei-fei-li-onhkvs/
[22] Fei-Fei Li, Stanford University. Senate testimony, 2023; quoted in PBS Firing Line, May 23, 2025. https://www.pbs.org/wnet/firing-line/?p=9291
[23] Dario Amodei, CEO of Anthropic. Morgan Stanley TMT Annual Conference, March 3, 2026. Summarized in 36Kr English, ‘Scaling Knows No Bounds.’ https://eu.36kr.com/en/p/3709346127933831
[24] Julian Nyarko, Stanford HAI / Stanford Law. Stanford AI Experts Say the Hype Ends in 2026, But ROI Will Get Real. Stanford Law School, December 16, 2025. https://law.stanford.edu/press/stanford-ai-experts-say-the-hype-ends-in-2026-but-roi-will-get-real/
[25] Chatham House. The Iran War Highlights the Creeping Use of AI in Warfare, March 27, 2026. On UN December 2025 resolution on AI in the military domain. https://www.chathamhouse.org/2026/03/iran-war-highlights-creeping-use-ai-warfare
