December 22, 2025

For a century, we organised companies on industrial-age assumptions. We divided work into discrete tasks, delegated them to specialised departments, documented every step, escalated issues and hoped for alignment. This model achieved remarkable success – it built railways, airlines, global supply chains and even the foundations of the internet. However, it relied on the assumption that humans were the sole interpreters of context. When that assumption no longer holds, everything shifts. Artificial intelligence does not merely automate routine tasks; it is a much more complex system. It comprehends context, retains past events and identifies inconsistencies that humans might overlook. It integrates data from multiple systems in seconds and provides insights without requiring manual data analysis. For enterprises today, this is not merely another tool but a paradigm shift in the nature of work. A New Production Factor: Cognitive Infrastructure Consider any company we admire: each has mastered a foundational resource before others. Toyota mastered the flow of value in its factories; Amazon excelled in lightning-fast logistics; Netflix captured our attention. The next generation of leaders will master a different resource: cognitive infrastructure. This is the system that carries knowledge, context, memory and reasoning across an entire organisation. Imagine it as providing your company with a brain. For the first time an enterprise can truly “think” alongside you. Not with a single isolated bot but with a comprehensive suite of intelligence working in concert. Consider today’s AI-powered toolset, which may include a large reasoning model (such as Google’s Gemini), an AI-powered workflow builder (Google Workspace Studio), a live code-documentation engine (CodeWiki), a persistent research notebook (NotebookLM) and even on-device assistants (Gemini Nano). These are not standalone applications but collectively form a nervous system for your business. This is a new kind of infrastructure that industrial-age companies never imagined. From Process-Driven to Intelligence-Driven Organisations Many organisations have undertaken digital transformation projects that have yielded limited results. This is because most companies update tools rather than their underlying assumptions. Historically, employees were expected to perform numerous tasks simultaneously, including memorising information, locating documents, interpreting data, coordinating activities, attending meetings and manually documenting their work. These assumptions become unsustainable at scale. Artificial intelligence (AI) offers a transformative approach. An AI-augmented organisation can: – Retain information by tracking decisions and history. – Interpret data through the analysis of reports, conversations and data. – Summarise complex information into clear insights. – Monitor projects for potential issues or delays. – Connect disparate data points. – Filter and prioritise information. This shift from asking how to improve processes to what is possible when context is preserved is where true transformation begins. Knowledge Work Leaks — AI Provides a Solution Organisations often experience a knowledge leak, losing more information than they create. For example, departing employees take decades of expertise with them, key decisions are lost in chat logs and outdated documents become obsolete quickly. New hires can spend months catching up. Furthermore, critical system details are often confined to a single engineer’s memory. This creates a slow drain in collective knowledge. While new information is constantly being added, it is not always retained. AI does not eliminate all knowledge gaps but provides a second brain for the enterprise that: – Captures context in real-time. – Continuously updates itself. – Links decisions to their underlying reasons. – Explains code and policies in accessible language. – Preserves historical data without overwhelming users. For example, consider a tool like CodeWiki that generates not only static documentation but continuously tracks the evolving state of a codebase. Alternatively, Workspace Studio could treat every meeting, chat, document and task as structured signals for learning rather than isolated artefacts. Even a research assistant like NotebookLM could become a collective memory anchor, not a place for note-taking but a repository of the team’s knowledge. This is not magic; it is the result of information being retained. Agentic Workflows: Work That Moves On Its Own What depletes teams is not the creative aspect of the work but the friction surrounding it. Individuals express dissatisfaction not with problem-solving but with the pursuit of context. Consider the repetitive tasks: locating the correct file version, seeking clarifications, tracking who is obstructing whom, switching between multiple applications, copying updates between tools, requesting approvals, coordinating opinions and filling in gaps of missing context. This is what causes fatigue, not the work itself. AI agents transform the landscape. They do not eliminate work but eliminate context friction. Imagine instructing the system to: “Prepare a compliance summary using the last three customer escalations, compare it against our risk matrix and draft remediation steps. Alert the security team only if the severity exceeds threshold B.” Observe the system’s response. It gathers emails, documents and data, synthesises the relevant information, compiles the summary and only engages human experts when necessary. Humans provide judgement, not drudgery. This is not merely automation (which follows fixed rules); it is interpretation—the computer comprehends your requirements and executes them. Your enterprise becomes an orchestration engine: people define intent, AI handles the routine tasks. where do humans fit? Higher up—not out. Yes, where do humans fit? Higher up—not out. When AI takes on the heavy lifting of context, where do people go? Up. We rise to focus on what machines still struggle with: purpose, ethics, creativity, and connection. Three big roles emerge: Intent Setters. These are the leaders and visionaries who define direction with clarity and nuance. They set the goals and context — not by micromanaging tasks, but by communicating purpose. Ambiguity Navigators. These humans handle what AI can’t: empathy, ethics, conflict resolution, negotiation, and decisions under incomplete information. They interpret and adapt when the situation is fuzzy. System Shapers. These are the architects, designers, product thinkers, and transformation leads who decide how humans and AI interact. They define the guidelines, design the interfaces, and ensure the system as a whole evolves smoothly. As AI gets smarter, human judgment doesn’t become less valuable — it becomes more precious. We move from brawn to brains. Engineering Organizations: The Deepest Earthquake For engineering teams, this shift feels like a seismic event.

When Jensen Huang mapped the modern AI ecosystem as a “five-layer cake”—energy, chips, infrastructure, models, applications—he offered more than a metaphor. He gave the field a grammar. It allowed people to speak about AI in terms of physics and production rather than mystique. For once, the conversation returned to ground truth: megawatts, fabrication plants, fibre, training cycles. The framework travelled quickly. Policymakers adopted it. Investors reorganised theses around it. Enterprises borrowed it for strategy decks. It now circulates as a near-official diagram of the global AI race. Its clarity, however, conceals a blind spot. Huang’s layers describe how AI becomes real. They say little about why it ought to exist, for whom it is built, or under what constraints it should operate. Between the physical substrate and the applications we touch lies a missing stratum: the human architecture of AI — the intentions, judgments and institutional choices that shape every outcome but rarely appear in technical diagrams. This is not the usual refrain about “keeping humans in the loop.” It is something more structural: the role of investors, business leaders, technologists and operators as intent-setters, ambiguity navigators and system designers. Without that architecture, the AI stack becomes directionless horsepower. What the Five Layers Show — and What They Hide Huang’s model remains a useful map: Read at face value, it implies a self-contained machine: feed in capital and engineering talent, and intelligence will flow outward like electricity through a grid. But history rarely cooperates with neat diagrams. General-purpose technologies behave more like tectonic plates than appliances. They disrupt institutions, labour markets, regulatory regimes, and notions of human competence. They redistribute power before anyone is ready. The missing layer is where real decisions reside: Who sets direction? Who imagines constraints? Who absorbs uncertainty? Walk the stack from bottom to top, and the imprint of human architecture becomes obvious. Energy: Not Supply, but Civilisational Intention AI’s appetite for energy is not incremental. It is civilisational. Training frontier models and serving them at scale turn electricity into a strategic commodity, not a utility. This layer appears technical but begins with human intention. Someone decides whether a nation commits to massive build-outs of solar, wind, SMRs or grid-scale storage. Someone determines how much resilience is non-negotiable. Someone judges the trade-offs between industrial policy, climate goals and computational ambition. None of these are engineering judgments. They are institutional choices about what kind of computational civilisation we intend to build. Ambiguity enters immediately: climate volatility, regulatory shifts, geopolitical disruptions, wild variance in workload forecasts. Here the human layer becomes visible: the people who interpret uncertainty, decide which risks to tolerate, and shape energy governance accordingly. Physics offers constraints; institutions choose what to do with them. Chips: The Bottleneck Where Power Concentrates Chips are the narrowest part of the funnel — and the most political. Talk often centres on benchmarks and efficiency. But the real contest is over control of the bottleneck of intelligence production. Intentions set the trajectory: Which fabrication ecosystems to commit to for a decade? Which architectures to bet on? Which workloads to optimise for? Which failure modes are tolerable? Ambiguity is constant: export controls, vendor theatrics, supply shocks, heterogeneous fleets. Decisions must be made while the ground shifts beneath them. System shaping then determines whether firms build coherent compute governance—fair allocation, lifecycle standards, tooling that abstracts vendor differences—or collapse under fragmentation. A country may win fabs and still lose the AI century if it fails to cultivate people who can architect around uncertainty. Infrastructure: Where Ambition Meets the Real World Data centres are where plans collide with land rights, zoning laws, community sentiment, cooling limits, sovereignty regulations and budget ceilings. Huang’s contrast between three-year American build cycles and China’s overnight construction is not a boast. It is a commentary on execution culture. Again, intention precedes engineering: Where to build? How large? Owned or leased? What availability and latency promises to make? What reliability envelope to enforce? Ambiguity lurks everywhere: planning permissions, community pushback, evolving security expectations, volatile workload patterns. System shaping determines whether organisations develop mature incident reviews, coherent chargeback models, and infrastructure patterns that prevent operational drift. Infrastructure is not only concrete and fibre; it is an exam in institutional competence. Models: The Arena of Negotiated Judgment Large models are often presented as artifacts—weights, benchmarks, licensing deals. But their real nature is negotiated. Intent defines objective functions, data regimes, risk thresholds, deployment criteria. Ambiguity surrounds every evaluation: noisy benchmarks, hyped claims, emergent capabilities, domain drift, hallucinations. System shaping becomes the decisive factor: evaluation governance, red-teaming, rollback authority, documentation practices, feedback pipelines. A model is not simply a trained network. It is a social contract about acceptable behaviour under uncertainty. Applications: Where Power Touches People The top layer is where AI meets lived experience. Here the consequences of bad intent or weak governance become tangible—job displacement, degraded service quality, loss of agency, or conversely, genuine augmentation. Intent determines whether companies build tools that elevate workers or simply chase marginal cost savings. Ambiguity emerges in adoption patterns, trust deficits, unexpected failure cascades, misaligned incentives. System shaping requires rethinking workflows, guardrails, escalation paths, training regimes, and the institutional mechanisms through which users can resist harmful deployments. If this layer fails, everything below it ultimately collapses into public backlash. The Real Competition: Human Architecture, Not Compute Volume Once we acknowledge the missing layer, the global AI race looks different. China builds faster. The United States designs better chips. Europe asserts regulatory leverage. The Middle East commands energy. But the decisive question is quieter: Which societies are investing in people who can think across systems? Who is training the next generation of intent-setters, ambiguity navigators and system architects? The country with the most coders may not lead the century. The country with the most systems thinkers might. In India, the opportunity is substantial but incomplete. We produce engineers at scale. We do not yet produce enough people who can think across layers—energy to chips, chips to infrastructure, infrastructure to models, models to applications, applications to institutions. That gap is not technical. It is educational and cultural. It is the missing layer. Conclusion: Restoring Direction to the