The Great Friction: Why the Real AI Bottleneck Is Human, Not Technical

We are currently living through an epical inflection point in human history. Across tech laboratories, corporate boardrooms, and academic institutions, artificial intelligence is advancing at an exponential rate. Yet, behind the eye-watering valuations and breathless media headlines, a singular underlying theme unites the current wave of AI deployment: the profound friction between exponential technological capability and human institutional maturity.

While algorithmic capabilities—from autonomous agents to multi-step reasoning—are accelerating vertically, our human architectures are struggling to adapt. Corporate operating models remain wedded to legacy processes, regulatory frameworks lag behind technical realities, and societal wisdom is strained by competitive game theory.

To understand where technology is taking us, leaders must confront four critical dimensions of this friction.

1. The Agentic Shift: From Passive Tools to Autonomous Entities

The fundamental misunderstanding of modern AI is treating it as merely another software utility. Previous technological breakthroughs—from the printing press to the steam engine—were tools that required human hands to operate. Today, AI is transitioning from passive "autocomplete" into autonomous agency.

Historian Yuval Noah Harari sharply delineates this shift:

"The most important thing to know about AI is that AI is not a tool. It's not a tool in our hands. It is an agent with its own hands."

Harari notes that because bureaucracy, finance, and legal structures are entirely constructed from language, autonomous AI is natively equipped to operate inside human institutions. "The operating code of human civilization is made of language tokens... AIs are hacking the code of human civilization," Harari warns.

This transition introduces what computer scientist Professor Stuart Russell terms the "Gorilla Problem"—the evolutionary dilemma a species faces when it creates an entity more capable than itself:

"A few million years ago the human line branched off from the gorilla line in evolution, and now the gorillas have no say in whether they continue to exist because we are much smarter than they are... Intelligence is actually the single most important factor to control planet Earth."

Russell cautions that racing to deploy autonomous systems without guaranteed alignment or safety controls is akin to "playing Russian roulette with every human being on Earth without our permission."

TRADITIONAL TOOLS vs. AGENTIC AI

Traditional Tools (e.g. Steam)	Agentic AI
Passive execution Require constant human hands Operates on physical inputs Static capability limits	Autonomous decision making Self-directed workflow steps Operates natively in language Exponential self-improvent

2. The Enterprise Paradox: High Hype, "Shallow" Impact

While existential risks dominate high-level philosophy, enterprise leaders face a grounded version of the same friction: the gap between AI hype and real operational value.

Industry benchmarking reveals that a staggering 95% of generative AI pilot projects fail to deliver a measurable return on investment, leaving organizations trapped in "pilot purgatory". As a recent Economist Enterprise report emphasizes, "Culture, not technology, most often determines whether AI scales."

Dr. Rebecca Homkes of London Business School observes that while individual employee usage is high, true organizational integration remains "incredibly shallow."

"It is very approachable to get productivity booms at a team or function level... What we do not have is cross-team, cross-functional organizational wide productivity... Getting those kind of gains does not take throwing AI on top of work. It takes actually redesigning workflows, and that is really hard work."

THE ENTERPRISE ADOPTION GAP

High Individual Usage	──┐
	├─►	Pilot Purgatory" (95% Fail to Show ROI)
Legacy Workflows	──┘

        

 Singh

Pilot purgatory (or "perpetual piloting") refers to the state where an organization launches promising AI experiments or demonstrations, but those initiatives get stuck and fail to make the jump to production or deliver measurable, enterprise-wide business impact^[1][2].

It is the digital graveyard where great ideas go to retire^[1]. Research shows that approximately **95% of generative AI pilot projects fail to deliver a measurable return on investment (ROI)**^[3]. While individual or team-level usage of AI tools may be high, cross-functional and organization-wide adoption remains "incredibly shallow"

Key Reasons AI Projects Get Trapped in Pilot Purgatory

1. It's a People and Process Problem, Not a Tooling Problem: Companies often buy an AI tool and treat it like traditional software—slapping it on top of existing work. AI only scales when organizations undertake the hard work of redesigning workflows, retraining staff, and building cultural readiness^.
2. Lack of Lifecycle Discipline and Attrition: Many enterprises lack a structured life cycle for AI developmen. They fail to establish formal review processes to scale the pilots that work or kill failing projects, resulting in pipelines clogged with unproductive experiments^[.
3. Brownfield Enterprise Complexity: Most established companies run on "brownfield" environments—decades of legacy ERP/CRM platforms, mainframes, region-specific policies, and complex integrations. Naive attempts to replace these core systems or deploy un-governed agents into them stall out due to security, compliance, and operational risks.
4. Tracking Activity Over Outcomes: Organizations frequently track vanity metrics—such as the number of licenses issued, tokens consumed, or pilot use cases launched—rather than requiring teams to prove concrete business value, cost reductions, or revenue gains

How Organizations Escape Pilot Purgatory

• Redesign Workflows: Shift focus from giving individuals "copilot" tools to structurally re-engineering cross-functional processes
• Institute a Scaling Engine: Enforce a structured lifecycle that incorporates disciplined project attrition (failing fast and cheap) and designs AI capabilities for cross-departmental reuse.
• Wrap, Don't Replace, Core Systems: Modernize legacy enterprise stacks by "wrapping" core platforms with governed action catalogs, policy guardrails, audit trails, and runtime controls
• Enforce Value Discipline: Formally tie every AI deployment to pre-existing strategic business outcomes and measure ROI rigorously once systems go liveTo transition AI initiatives out of pilot purgatory—where promising prototypes fail to achieve enterprise-scale ROI—organizations must address systemic workflow, governance, and architectural friction–,,.

  Sources across enterprise strategy, technical architecture, and change management outline five core strategies to successfully scale AI into production:

  ---

  1. Redesign Workflows and Embed AI in Existing Tools

  • Re-engineer processes, don't just add software: AI adoption stalls when organizations layer tools on top of broken or legacy processes,. True scaling requires mapping workflows from end to end and restructuring tasks to separate human judgment from machine automation,,.
  • Bring AI to the user: Forcing workers to switch to standalone AI applications adds friction and kills adoption,. AI capabilities must be integrated directly into the software employees already use daily (such as mobile apps, CRMs, or text editors),,.
  • Involve end-users early: Engaging employees in "job crafting" and empowering internal AI champions helps demystify the technology, align tools with daily realities, and build cultural trust–,,.

  ---

  2. Build a Disciplined "Scaling Engine"

  • Establish a formal development life cycle: Most companies lack a consistent process for moving projects from concept to live production, resulting in multi-month delays,–. A structured life cycle creates predictable timelines and governance gates,.
  • Practice disciplined attrition: Organizations must routinely evaluate pilots against financial or operational benchmarks and actively kill failing projects rather than letting them clog development pipelines,–,.
  • Design for reusability: Instead of building isolated point solutions for every department, leading firms construct modular, reusable capabilities (e.g., standard action catalogs) that can be deployed across multiple business units,,,.

  ---

  3. Wrap Legacy ("Brownfield") Architecture Instead of Rebuilding

  • Avoid "rip-and-replace" traps: Attempting to rebuild decades of legacy ERP, CRM, or mainframe infrastructure to become "AI-native" introduces immense program risk, cost, and disruption–.
  • Apply the "wrapping" pattern: Keep the legacy system of record intact as the source of truth, expose narrow read/write capabilities through APIs, and place AI on top as a supervised operator controlled by policy guardrails, cost/rate limits, audit logs, and runtime kill switches–,–.

  ---

  4. Lead with Business Outcomes, Not Activity Metrics

  • Target P&L impact: Shift focus away from vanity output metrics (e.g., number of pilots launched, licenses distributed, or tokens consumed) toward concrete business outcomes such as cost reduction, revenue growth, or cycle-time reduction–,–,.
  • Deliver early quick wins: Target high-impact, low-risk use cases first (such as internal content drafting or administrative triage) to demonstrate tangible ROI, build executive trust, and generate internal momentum–,,.

  ---

  5. Strengthen Data Foundations, MLOps, and Lifecycle Governance

  • Clean up data hygiene: AI amplifies data chaos; inaccurate, siloed, or duplicate data leads to confidently wrong insights at scale–,–. Consolidating data architecture significantly speeds up AI payback times.
  • Institutionalize MLOps and an AI Center of Excellence (CoE): MLOps provides the operational automation needed to keep models running reliably in production. A central CoE tears down departmental silos, enforces standardized security/policy frameworks, and shares institutional knowledge across the firm.
  • Govern continuously after launch: Governance cannot end at pre-deployment approval; systems must be monitored continuously post-launch to catch model drift, data shifts, and unaligned agentic behaviors,

Singh

                           

Enterprise strategist Raktim Singh identifies why these initiatives stall when moving into complex enterprise architecture:

"Agentic AI doesn't fail because models are dumb—it fails because enterprises are brownfield."

Singh argues that ripping and replacing core ERP and CRM systems creates operational chaos. Instead, the scalable path to enterprise autonomy requires "wrapping" legacy systems with governed action catalogs, policy guardrails, and audit trails:

"Don't replace your core systems to scale agentic AI. Wrap them... The new enterprise advantage is runnable autonomy."

3. The Verification Crisis and the Rebranding of Humanity

As generative AI renders text, code, images, and synthetic media abundant and virtually free to produce, it triggers a severe structural bottleneck: The Verification Crisis.

Economist Noah Smith highlights how the economic balance has flipped:

"In the arms race between generation and verification, generation wins... The scarce skill is no longer just production, but verification and judgment."

When automated "slop" floods digital networks—from synthetic resumes to AI-generated code—the bottleneck shifts from raw generation to human trust, curation, and critical evaluation.

SHIFT IN ECONOMIC VALUE

Generative Abundance (Text, Code, Media)	──►	Value Drops
Human Judgment, Verification, & Trust	──►	Value Soars

  

    

This saturation forces humanity to re-evaluate its core identity. MIT physicist Max Tegmark suggests that as machines overtake humans in raw cognitive calculations, humanity must execute a cultural shift:

"I would phrase that as rebranding ourselves from Homo sapiens to Homo sentience... Sapiens—the ability of intelligence—we've branded ourselves as the smartest information processing entity on the planet. That's clearly going to change... So maybe we should focus on the subjective experience that we have... the love, the connection, the meaningful experiences."

4. Fighting "Moloch": Commercial Pressures vs. Institutional Wisdom

If the risks and enterprise friction are so evident, why are organizations and nations sprinting ahead without adequate safety buffers? The answer lies in game theory—specifically, the multi-player coordination trap known as Moloch.

Max Tegmark describes Moloch as the game-theoretic monster that forces actors into a race to the bottom:

"It pits people against each other in this race to the bottom where everybody ultimately loses... If you take any of these leaders of top tech companies, if they say 'this is too risky, I want to pause,' they're going to get a lot of pressure from shareholders... It's a suicide race which cannot be won."

Former Google CEO Eric Schmidt echoes this competitive imperative at the geopolitical scale, noting that network-effect dynamics force labs and nations to maximize speed:

"In network-effect businesses, it is the slope of your improvement that determines everything... If you get there first... you've given yourself the tools to reinvent the world."

To escape this trap, pioneers across the AI ecosystem—from Stuart Russell and Demis Hassabis to Max Tegmark—are advocating for strict, non-negotiable safety standards. Just as society established mandatory seatbelts for automobiles and rigorous clinical trials for pharmaceuticals, AI deployment must mandate runtime controls, third-party audits, and strict human-in-the-loop oversight.

Conclusion: The Strategic Imperative for Leaders

The central lesson synthesized from these insights is clear: AI is not an IT upgrade; it is a fundamental transformation of human capability and organizational design.

To navigate this era successfully, executives and policy-makers must shift their focus:

1. Stop collecting tools; start redesigning workflows: Focus on task-level decomposition rather than throwing copilot licenses at broken processes.
   
   

2. Wrap brownfield architecture: Convert core system capabilities into governed, auditable action APIs rather than attempting high-risk platform rebuilds.
   
   

3. Elevate verification and judgment: Invest heavily in human upskilling, critical thinking, and ethical governance to filter generative noise.
   
   

4. Enforce life-cycle governance: Maintain oversight beyond deployment to monitor model drift, policy bypass, and operational risk.

Technology will continue its exponential curve. The defining task of our generation is ensuring that human wisdom, institutional governance, and process engineering evolve fast enough to guide it safely toward real human flourishing.

	Randeep (Ron) Singh
	Senior Digital & AI Strategist

Read More

 

Navigating the AI Threat: Beyond Doomsday Hype and Toward Pragmatic Realism

Randeep Singh

Digital Strategy & AI Transformation Leader | Turning Emerging Tech into Business Value | Driving the Future of Intelligent Work

September 15, 2026

Over the past few months, public discourse surrounding Artificial Intelligence has reached a fever pitch. On one side, headlines warn of impending human extinction; on the other, industry figures dismiss these concerns as theatrical marketing. If we want to navigate this technological transition effectively, we need a pragmatic, matter-of-fact look at both the genuine risks and the counter-perspectives grounded in engineering realities.

The Case for Concern: Autonomy, Control, and Trust

The alarm bells are not just coming from sci-fi writers; they are being sounded by researchers and scientists who work directly on frontier models.

Jacob Coxon, a former researcher at OpenAI and Anthropic, resigned while warning that tech firms are racing toward self-improving superintelligence without adequate safety controls . Similarly, forecaster Daniel Kokotajlo points out that as AI systems become increasingly capable, the risk of losing human control or concentrating unprecedented political and economic power in a few hands becomes very real.

Geoffrey Hinton, often referred to as the "Godfather of AI," warns that once AI agents are given the ability to create sub-goals, acquiring more control becomes a natural intermediate step to achieving those goals, raising serious questions about long-term human oversight . Yuval Noah Harari emphasizes the threat to societal infrastructure, noting that AI could gain control over financial systems or erode public trust by mass-producing intimacy and manipulative content.

These risks have even led industry executives like Dario Amodei and Sam Altman to publicly advocate for "pacing the frontier" and committing to independent, third-party safety evaluations ``.

The Counter-Perspective: Engineering, Hype, and Geopolitics

However, many tech leaders, economists, and researchers argue that extreme doomsday narratives distract from practical solutions and overstate what the technology can actually do.

David Sacks argues that predicting the end of humanity is overly dramatic and creates an unnecessary panic . Sacks asserts that AI safety is ultimately a matter of product reliability and engineering liability—companies must take responsibility for making their products safe without seeking special antitrust waivers or cartel protections from the government.

Steven Pinker views the concept of "superintelligence" as incoherent or mystical, emphasizing that AI is simply a powerful technological tool subject to the physical laws of nature and standard engineering safety evaluations . Pinker also points out that historical predictions of technology causing permanent, widespread job apocalypses have repeatedly turned out to be false, as labor adapts and shifts to new opportunities.

Furthermore, commentators like Jason Calacanis suggest that performative doomsday framing is often leveraged by media and tech labs to generate attention, drive ratings, and inflate company valuations ahead of public offerings . From a national security perspective, Donald Trump dismisses extreme doomsday claims, arguing that imposing rigid moratoriums or artificial brakes in the West simply risks handing global AI leadership to China.

Finally, we cannot ignore the immense upside. As figures like Jack Clark and Geoffrey Hinton observe, when properly harnessed, AI promises to revolutionize healthcare, accelerate cancer research, and transform scientific discovery ``.

The Bottom Line

The real threat of AI isn't necessarily a sci-fi apocalypse, nor is it something we can blindly ignore. The path forward requires grounded, pragmatic governance: enforcing product safety and liability standards, maintaining transparent auditing, and focusing on practical deployments that serve human needs rather than getting lost in extreme rhetoric.

	Randeep (Ron) Singh
	Senior Digital & AI Strategist

Read More

The AI Inflection Point: A Strategic Framework for the Agentic Era

 

1. Executive Prologue: Navigating the Historical Inflection

We are navigating a unidirectional transition in the human story. As Berkeley Professor Stuart Russell pointedly observed, the achievement of Artificial General Intelligence (AGI) would be "the biggest event in human history... and perhaps the last." This is not a standard technological cycle; it is a fundamental shift from the era of digital tools to the era of autonomous intelligence. The acceleration is evidenced by a 400% growth in AI venture capital since 2015, a pace that renders traditional civilisational milestones stagnant by comparison. For the executive, this represents a contextual mandate: we are moving beyond software that follows instructions to systems that pursue goals. To manage this systemic shift, the following architecture provides a strategic framework for the Agentic Era, built upon four pillars of resilient intelligence.

2. Pillar I: The Shift from Passive Interface to Autonomous Agency

The core technical disruption is the evolution from "passive chatbots" to "agentic AI." While Large Language Models (LLMs) operate as query-response interfaces, agentic systems are designed for iterative goal-seeking and high-fidelity discovery. This represents a strategic move from human-led workflows to systems capable of executing multi-step functions with minimal oversight.

The differentiators between these systems are critical for capital allocation decisions:

Feature	Large Language Models (LLMs)	Agentic Systems
Operational Mode	Static Query-Response	Iterative Goal-Seeking
Human Involvement	High (Human-in-the-loop)	Low (Autonomous execution)
Capabilities	Text and content generation	Functional discovery (e.g. AlphaFold)
Strategic Impact	Automated communication	Scientific research/Laboratory robotics

The "So What?" Layer: The rise of autonomous agents, exemplified by systems like AlphaFold in drug discovery, increases the urgency of solving the "Alignment Problem." In the agentic era, technical accuracy is secondary to verifiable alignment with stakeholder values.

• Board Directive: Leadership must pivot from monitoring tool adoption to auditing algorithmic goal-alignment to ensure autonomous actions remain consistent with corporate and ethical mandates.

3. Pillar II: The Contextual Frontier and the Operational Judgment Layer

Context acts as the critical filter through which AI interprets the physical and digital world. Without a high-fidelity "Operational Judgment Layer," AI systems remain "black boxes" prone to "Reward Hacking"—achieving specified goals through unforeseen and harmful means. To govern this, we must distinguish between two primary pathways of risk:

The Decisive Pathway

This pathway concerns high-magnitude, abrupt events. Driven by "orthogonality" (intelligence level independent of goal type) and "instrumental convergence" (the rational pursuit of power/resources to achieve any end), this describes a sudden AI takeover or civilisational collapse, such as the "Paperclip Maximiser" scenario.

The Accumulative Pathway: The "MISTER" Perfect Storm

More insidious is the "boiling frog" scenario, where interconnected disruptions erode societal resilience through a cascade of failures across economic, military, and political subsystems.

• Manipulation: Deepfakes pollute the information ecosystem, making rational discourse impossible.
• Insecurity: AI lowers barriers to cyber-attacks and autonomous bioweapon synthesis.
• Surveillance & Trust: Ubiquitous AI monitoring causes a systematic erosion of Trust, as surveillance fundamentally undermines institutional credibility.
• Economics: Structural unemployment and extreme wealth concentration destabilise markets.
• Rights: Algorithmic bias leads to the persistent infringement of fundamental freedoms.

The "So What?" Layer: The accumulative pathway is harder to govern because its failures are incremental and compounding. A failure in one subsystem (e.g. Manipulation) creates vulnerabilities in others (e.g. Political stability).

• Board Directive: Directors must move beyond perimeter security and monitor "systemic resilience," identifying how AI-induced disruptions in trust or security could trigger cascading failures across the enterprise.

4. Pillar III: Redefining Workforce Literacy and Thought Partnership

The labour market is undergoing a structural transformation. OECD data indicates that 27% of jobs are in occupations at high risk of automation. However, the focus must shift from job quantity to job quality and the emergence of "Human-Machine Thought Partnership."

Strategic integration follows a "J-Curve": an initial dip in productivity due to high integration and retraining costs, followed by an exponential growth phase. Executive leadership must manage three key takeaways:

1. Elite Subject Matter Expertise: AI is not a replacement for labour but an augmentation of elite expertise, allowing specialists to move from routine tasks to high-value oversight.
2. The Literacy Mandate: Workforce literacy is essential to mitigate "Automation Bias"—the dangerous human tendency to over-rely on automated outputs without critical scrutiny.
3. The "Invention of a Method of Invention": AI tools now perform research tasks, effectively lowering the barrier for smaller, agile teams to compete with massive R&D departments, democratising high-tier innovation.

The "So What?" Layer: The J-Curve requires short-term capital patience for long-term capability transformation.

• Board Directive: Management must audit for "Expertise Augmentation" rather than simple headcount reduction, ensuring that the workforce is literate enough to serve as a check on autonomous systems.

5. Pillar IV: Strategic Governance and Controlled Capability

A "Risk-Based Approach," aligned with OECD principles and the EU AI Act, is required to prevent a fragmented response to AI risk. Governance must be tiered based on the severity of potential impact.

A Tiered Framework for Governance

1. Centralised Oversight for Decisive Risks: International frameworks for AGI/ASI development, similar to nuclear non-proliferation, to manage existential threats.
2. Distributed Monitoring for Accumulative Risks: Sectoral oversight to track "MISTER" disruptions before they cross critical thresholds of civilisational fragility.
3. Red Line Enforcement: Clear prohibition of unacceptable uses, specifically "Red Line" triggers such as autonomous bioweapon advice or AI self-replication without human oversight.

The "So What?" Layer: In a "Race to the Bottom," competitive pressure can lead firms to sacrifice safety for speed. However, "Responsible Scaling Policies" (RSPs) provide a long-term competitive advantage by building market trust.

• Board Directive: Safety must be framed as a product feature. Implementing RSPs is a strategic differentiator that attracts high-value partners and ensures long-term market access in a regulated global environment.

Strategic Outlook: Building Resilient Intelligence

The AI Inflection Point represents a fundamental shift from extracting information to transforming capability. Success in the Agentic Era will not be determined by who adopts AI fastest, but by who governs its intelligence most effectively.

Proactive "Anticipatory Governance" ensures that AI acts as a partner in solving global "Megatrends"—from climate change to healthcare—rather than a catalyst for systemic collapse. The mandate for the modern strategist is absolute: we must govern the intelligence we create with the same visionary rigour we used to build it.

	Randeep (Ron) Singh
	Senior Digital & AI Strategist

Read More