Rahul Kulkarni

Abstract

In this article, the author argues that artificial intelligence should be viewed as a national capability requiring institutional infrastructure rather than merely advanced models or computing resources. It proposes a National Intelligence Framework with four interconnected pillars: Knowledge Infrastructure, Intelligence Economy, Intelligence Governance and Economic Participation. The framework proposes building systems that capture and organize expertise, applying the knowledge across the economy, establishing standards and oversight for AI, and creating mechanisms to reward people and institutions that contribute expertise. Together, these pillars are intended to strengthen India’s ability to translate existing technological, institutional and human capital into sustained productivity, innovation and economic growth.

The Intelligence Flywheel Framework 

Artificial intelligence is often discussed as a technology race: who builds the largest models, owns the most GPUs, or funds the next generation of startups. These are important questions, but they miss the larger transformation underway. A growing body of economic research suggests that AI is not simply another technology to be adopted, but a general-purpose capability that reshapes how production itself is organised. Brynjolfsson, Rock and Syverson argue that the largest productivity gains from AI come not from the models themselves, but from the complementary organisational changes that surround them. The IMF similarly identifies AI as a macroeconomic force capable of reshaping productivity, labour markets and long-term growth. Brookings places AI alongside electricity and the internet as a general-purpose technology whose economic impact depends less on the technology itself than on the infrastructure and institutions built around it. More recent work even proposes autonomous digital intelligence as an emerging factor of production alongside labour and capital.

Taken together, these perspectives point to a broader conclusion. The defining competitive advantage of the AI era will not be on who can build the best model by capability alone. Countries will increasingly compete on who builds the fastest and most trusted systems for capturing knowledge, deploying intelligence, measuring outcomes and reinvesting those gains back into the creation of new knowledge. These will be the intelligence flywheels of the future – self-reinforcing systems where every cycle of knowledge creation and deployment makes the next cycle faster and more efficient. Much like electricity grids and water pipelines, AI is becoming a form of national infrastructure whose value depends on the institutions that create, govern and sustain these intelligence flywheels.

India enters this transition with four extraordinary strategic assets. Three decades of IT services have grown into a USD 250+ billion industry, serving thousands of global enterprises while creating one of the world’s deepest repositories of codified enterprise knowledge and operational expertise. India hosts over 1,700 Global Capability Centres employing nearly two million professionals, making it the world’s largest GCC ecosystem and the operational backbone of many of the world’s leading enterprises. India’s Digital Public Infrastructure connects 1.4 billion Aadhaar identities, processes more than 18 billion UPI transactions each month, and demonstrates the country’s ability to build interoperable digital systems at population scale. Finally, India produces more than one million engineering graduates annually, hosts the world’s third-largest startup ecosystem, and has a workforce exceeding 560 million people, providing an unparalleled foundation for AI-driven productivity and innovation.

Yet these assets remain disconnected. Knowledge is documented but rarely becomes reusable intelligence infrastructure. Scientific discovery remains weakly connected to economic deployment. AI is largely deployed as isolated productivity tools rather than as a new production model. Governance focuses primarily on regulating models instead of measuring national intelligence capability. And the people whose expertise trains these systems seldom participate in the value they continue to create. India possesses many of the ingredients for an intelligence economy, but not yet the operating model that allows them to reinforce one another.

This essay argues that the next phase of AI is more an institutional challenge than a technological one. The objective is not simply to build smarter AI systems, but to build a national intelligence flywheel in which every stage continuously strengthens the next. Expertise needs to transform into reusable knowledge, knowledge into productive capability, productive capability into trusted public infrastructure, and economic value into incentives for creating even more knowledge. 

The framework introduced here, the National Intelligence Flywheel Framework, organises this challenge along two dimensions. The horizontal axis follows the economic lifecycle of intelligence, from the creation of knowledge to its deployment throughout the economy. The vertical axis distinguishes what a nation must build from how it must govern it. Infrastructure creates capability; institutions ensure that capability is trusted, measured and rewarded. Together, these two axes define the four foundational pillars of the National Intelligence Flywheel Framework. 

CREATE INTELLIGENCE DEPLOY INTELLIGENCE

Where intelligence is created Where intelligence is applied

INFRASTRUCTURE  Knowledge Infrastructure             Intelligence Economy

    • Knowledge representation            • Strategic positioning

                 • Agentic execution      • Monetisation models

                 • Human-AI orchestration • Intelligence-native enterprises

                 • Intelligence evaluation • Workforce augmentation

         • Knowledge discovery           

--------------------------------------------------------------------------------------------------------

INSTITUTIONS   Intelligence Governance             Economic Participation

   • AI assurance                       • Skill attribution

                 • National measures • Capability finance

                 • National standards

National intelligence compounds when knowledge is encoded, applied, verified, and rewarded.

Figure 1. The National Intelligence Flywheel Framework

The four pillars of the National Intelligence Flywheel Framework are complementary capabilities that together power the national intelligence flywheel. Knowledge Infrastructure transforms expertise into reusable intelligence through knowledge representation, agentic systems, human-AI orchestration, evaluation and discovery. The Intelligence Economy deploys that intelligence across enterprises, startups, and the workforce with different business models to create new forms of productivity and economic value. Intelligence Governance establishes the standards, assurance and national measurement systems that make intelligence trustworthy at scale. Economic Participation ensures that the wealth created by intelligent systems flows back to the individuals, institutions and communities whose knowledge made those systems possible, aligning long-term incentives for continuous contribution and innovation.

Unlike traditional technology roadmaps, the framework is designed as a reinforcing system rather than a sequence of policy interventions. Better knowledge infrastructure creates more capable intelligent systems. Wider deployment generates richer data, stronger benchmarks and new knowledge. Better governance builds trust, accelerating adoption across the economy. Fair participation rewards those who contribute expertise, encouraging further knowledge creation. Each cycle leaves the nation with a larger knowledge base, stronger institutions and a more productive economy than before. This reinforcing system creates an intelligence flywheel that helps intelligence compound. The nations that build the fastest intelligence flywheels will build the fastest intelligence economies.

We now explore each one of the four pillars that are part of the Intelligence Flywheel framework. 

Knowledge Infrastructure

Knowledge Infrastructure defines how expertise is transformed into reusable intelligence. While foundation models provide general reasoning capability, durable competitive advantage increasingly depends on domain-specific knowledge: how organisations make decisions, execute processes, resolve exceptions and continuously improve performance, in the context of their domain. The strategic challenge is therefore not solely the development of better models, but the construction of systems that systematically capture, represent, evaluate and expand domain specific knowledge.

It must be noted that Knowledge and Data are not the same. Data records observations. Knowledge captures procedures, decision logic and causal relationships that enable action based on that data. As intelligent systems become more capable, the representation of knowledge becomes a national capability analogous to software engineering in the digital economy. Countries that encode expertise into reusable and interoperable forms accumulate an intelligence asset that improves with every deployment.

Knowledge infrastructure consists of five mutually reinforcing capabilities: knowledge representation, agentic systems, Human-AI orchestration, intelligence evaluation and knowledge discovery. Together they define how tacit expertise becomes executable intelligence and how each generation of intelligent systems contributes to the next.

Knowledge Representation

Before an intelligent system can perform a task, the underlying expertise must be documented in a form that is complete, auditable and reusable. Knowledge representation is therefore the first engineering problem in building knowledge infrastructure. 

India possesses a distinctive advantage in this respect. Over three decades, the IT services industry developed systematic processes for converting tacit operational and domain expertise into explicit organisational knowledge. The Knowledge Transfer (KT) session, developed to preserve continuity across projects, required experienced practitioners to document processes, decision criteria, assumptions, exception handling and operational context before responsibility changed hands. TCS scaled this practice across hundreds of thousands of professionals. Infosys institutionalised it through structured training programmes at Mysore. Wipro embedded similar practices within domain-specific Centres of Excellence. Delivery playbooks, operating procedures, process documentation, decision trees and CMMI artefacts were created to improve software delivery and organisational quality across the IT services companies. Viewed retrospectively, these organisations were constructing one of the world’s largest repositories of explicit enterprise domain knowledge. 

This changes the economic interpretation of India’s services industry. Enterprise domain knowledge representation is no longer simply a mechanism for organisational continuity. It becomes a reusable intelligence asset whose value increases as it is encoded into deployable agentic systems. The principal strategic question for India’s technology sector is therefore whether this accumulated knowledge becomes domestic intelligence infrastructure or remains in isolated files and minds of IT services professionals working in this sector.

Agentic Systems

To transform these enterprise domain knowledge representations into actionable intelligence, we must move beyond the deterministic workflows of traditional software. Modern AI-based agentic systems provide the necessary architecture for this shift, as they are capable of goal-oriented task breakdown, autonomous decision-making, and execution based on the very knowledge we captured in the previous section.

Conventional applications encode deterministic workflows defined during software development. Agentic systems instead combine general reasoning models with explicit representations of domain knowledge, external tools and continuous feedback mechanisms. The quality of the resulting system depends as much on these knowledge representations passed as context to the model, as on model capability.

This introduces a new unit of software construction: the executable skill, a reusable mechanism for agent skills to encapsulate this context. Rather than describing procedures through documentation intended for human interpretation, organisations increasingly represent expertise through structured skill definitions specifying objectives, decision boundaries, required tools, inputs, outputs, escalation conditions and evaluation criteria. Emerging format specifications such as skills.md illustrate this transition by treating organisational expertise as version-controlled software artefacts rather than static documents.

The following four architectural properties distinguish these agentic systems from the deterministic workflows of the past and make them highly valuable as knowledge infrastructure:

1. Explicit knowledge representation. Agentic systems require domain expertise to exist outside individual practitioners. Converting tacit knowledge into structured skills as skills.md frequently exposes undocumented assumptions, inconsistent decision processes and missing operational criteria. The process of representation therefore improves organisational understanding independently of automation.
2. Continuous refinement. Every deployment generates execution traces, evaluation results, human interventions and performance metrics. These observations provide training data for improving the underlying skill definitions. Competitive advantage consequently shifts from static capability to learning velocity. Organisations capable of rapidly refining deployed skills accumulate performance advantages that cannot be replicated simply by adopting larger foundation models.
3. Interoperable execution. Agent ecosystems increasingly depend on standardised interfaces for capability discovery and tool invocation. Protocols such as the Model Context Protocol (MCP) allow intelligent systems to invoke capabilities developed by independent organisations without requiring bespoke integration. The long-term significance lies less in any individual protocol than in the emergence of an interoperable ecosystem in which intelligence becomes composable infrastructure.
4. Composable architectures. Enterprise systems evolve from monolithic applications toward collections of specialised agents coordinated through orchestration layers. Individual skills may evolve independently while contributing to larger workflows, allowing organisations to improve capabilities incrementally rather than through complete system replacement.

Taken together, these properties alter the basis of competition. Model capability establishes an initial level of performance, but the quality of knowledge representation, the speed of refinement and the ability to compose specialised capabilities into increasingly sophisticated systems is what provides long term advantage. 

Human-AI Orchestration

As organisations adopt agentic systems, the role of human expertise changes fundamentally. The objective is no longer to optimise how humans perform repetitive knowledge work, but to optimise how humans and intelligent systems collaborate. Human-AI orchestration therefore becomes the third engineering problem in building knowledge infrastructure.

The dominant narrative surrounding AI has focused on replacement: humans versus machines. In practice, enterprise deployments increasingly demonstrate a different pattern. The highest-performing systems are neither fully autonomous nor entirely human-driven. They are collaborative systems in which humans contribute judgment, context and oversight, while agents contribute execution, consistency and scale.

This transition gives rise to a new organisational role: the skill architect. Unlike a software architect, who designs applications, or a prompt engineer, who optimises model interactions, the Skill Architect is responsible for designing organisational intelligence itself. This includes identifying reusable expertise, defining executable skills, specifying decision boundaries, selecting appropriate tools, designing evaluation criteria and continuously improving deployed skills based on operational feedback.

The focus of expertise therefore shifts from repeatedly solving the same problem to encoding the solution once and continuously refining it thereafter. Knowledge workers increasingly become knowledge designers, responsible for improving the capabilities of intelligent systems rather than directly executing every task. Their contribution compounds across every subsequent deployment of the skills they create.

Human-AI orchestration also requires explicit allocation of responsibility. Some decisions remain fully automated, others require human approval, while high-consequence decisions continue to rely on human judgment supported by intelligent systems. Designing these interaction boundaries becomes a core systems engineering problem rather than an organisational afterthought. Poorly defined orchestration leads either to excessive automation, where humans lose meaningful oversight, or excessive intervention, where intelligent systems provide little productivity benefit.

The engineering objective is therefore not autonomous agents operating independently of people, but well-designed human-agent systems whose combined capability exceeds that of either working alone. Organisations that systematically develop this orchestration capability will accumulate organisational intelligence that becomes increasingly difficult for competitors to replicate.

Intelligence Evaluation

Engineering disciplines mature only when they become measurable. Software engineering introduced testing frameworks, code coverage, continuous integration and performance benchmarking. Knowledge infrastructure requires an equivalent discipline for evaluating intelligent systems.

Traditional software validation focuses primarily on functional correctness. Agentic systems introduce additional dimensions of evaluation because outputs are probabilistic, reasoning processes evolve over time and system performance depends upon both model capability and knowledge representation. Evaluation therefore becomes a continuous engineering activity rather than a final testing phase.

Most publicly reported AI benchmarks measure only outcome correctness. Benchmarks such as GSM8K, MMLU and HotpotQA compare whether a model produces the correct answer on standardised datasets. While valuable for measuring general reasoning capability, they provide only a partial assessment of production-grade intelligent systems.

Enterprise deployments require evaluation across three complementary dimensions.

1. The first is outcome quality. Does the system consistently produce correct and reliable results?
2. The second is execution quality. How efficiently does the system arrive at those results? This includes reasoning quality, tool selection, latency, inference cost, robustness to changing conditions and the frequency of human intervention. An agent producing correct answers through brittle reasoning or prohibitively expensive execution cannot be considered production ready.
3. The third is learning capability. Does the system improve as it is deployed? Every execution produces traces, corrections, user feedback and evaluation data. Organisations capable of systematically incorporating this feedback into their executable skills develop intelligence that compounds over time, while static systems gradually lose competitiveness.

The proposed 3×3 Agent Evaluation Framework organises these dimensions across multiple levels of abstraction, allowing organisations to evaluate not only models but complete human-agent systems. This shifts evaluation from measuring isolated AI performance toward measuring organisational intelligence.

Figure 2. 3x3 Evaluation Framework

National competitiveness increasingly depends upon domain-specific evaluation rather than generic benchmarks. Global datasets rarely evaluate multilingual agricultural advisory, GST compliance, district hospital workflows, Indian contract law or public service delivery. A model achieving state-of-the-art performance on international leaderboards but failing within India’s institutional context provides limited strategic value. 

India therefore requires a National AI Benchmarking Initiative comprising standardised evaluation datasets across Indian languages, regulatory environments and sector-specific domains. Maintained by leading academic institutions and integrated into public procurement, these benchmarks would establish common measures of quality while directing innovation toward nationally relevant capabilities. Evaluation standards ultimately become procurement standards, and procurement standards shape the direction of technological investment.

Knowledge Discovery

Knowledge infrastructure should not merely preserve existing expertise. It should continuously expand the frontier of knowledge itself. Knowledge discovery therefore represents the final capability in the intelligence lifecycle and closes the feedback loop through which intelligence compounds.

Scientific discovery has historically been constrained by human cognitive limits. Researchers spend considerable effort identifying relevant literature, synthesising prior work, formulating hypotheses and designing experiments before new knowledge can be created. Agentic AI systems fundamentally change this process by combining large-scale knowledge retrieval, reasoning, planning and iterative experimentation into a continuous discovery loop. Rather than serving only as search engines or summarisation tools, these systems increasingly participate in the scientific method itself by generating hypotheses, proposing experiments, interpreting results and refining subsequent lines of inquiry.

This transition is already visible at the scientific frontier. Google’s DeepMind demonstrated that AI could solve problems previously considered intractable through AlphaFold, which predicted the three-dimensional structures of more than 200 million proteins and fundamentally accelerated structural biology, work recognised through the 2024 Nobel Prize in Chemistry awarded to DeepMind CEO Demis Hassabis and team. More recently, autonomous research systems such as The AI Scientist demonstrate how agentic systems can iteratively survey literature, formulate hypotheses, design experiments, execute computational workflows and generate complete scientific manuscripts with minimal human intervention. In parallel, autonomous laboratory platforms at organisations such as Recursion Pharmaceuticals and Insilico Medicine already combine AI-generated hypotheses with robotic experimentation, creating closed-loop systems in which hypothesis generation, experimentation and learning continuously reinforce one another.

These agentic systems should not be understood as replacements for scientists, but as force multipliers that dramatically expand the scope of scientific exploration. A researcher may read hundreds of papers during a doctoral programme; an agentic research system can synthesise hundreds of thousands, identify previously disconnected concepts across disciplines and continuously incorporate new experimental evidence. The limiting factor for scientific discovery therefore shifts from human capacity to read and reason toward the quality of the underlying knowledge infrastructure and experimental feedback loops.

In India, the opportunity extends beyond contemporary scientific literature. We possess one of the world’s largest repositories of historical technical knowledge, including Ayurveda, Sanskrit scientific literature, agricultural practices and traditional ecological knowledge. Much of this knowledge remains inaccessible to computational analysis despite representing centuries of empirical observation. Combining these historical knowledge systems with modern genomics, systems biology, computational chemistry and AI-driven hypothesis generation creates opportunities for entirely new forms of scientific discovery. Rather than treating traditional knowledge as cultural heritage alone, it can become an active component of contemporary research programmes whose findings are validated using modern scientific methods.

Knowledge discovery therefore completes the Knowledge Infrastructure pillar. Knowledge representation captures existing expertise. Agentic systems operationalise that expertise. Human-AI orchestration continuously improves it. Intelligence evaluation measures its quality. Knowledge discovery expands the underlying corpus from which future intelligent systems are built. Each deployment contributes new observations. Each evaluation refines existing knowledge. Each scientific discovery enlarges the national intelligence base. The output of one generation of intelligent systems becomes the knowledge infrastructure for the next. This is the mechanism through which national intelligence compounds.

The Intelligence Economy

Knowledge infrastructure defines how intelligence is created. In this section we discuss the intelligence economy, which defines how intelligence is transformed into productive capacity and economic value.

The industrial economy organised production around physical capital, labour and energy. The digital economy reorganised production around software and networks. The intelligence economy introduces a further transition: decision-making itself becomes programmable. Intelligent systems increasingly perform cognitive tasks that previously required trained professionals, while humans concentrate on judgment, oversight and the creation of new knowledge.

This transition affects every sector of the economy. Artificial intelligence is not another industry alongside manufacturing, healthcare or financial services. It is a general-purpose production technology that changes how every industry creates value, organises work and competes. The strategic challenge for nations is therefore not simply accelerating AI adoption but building markets that continuously convert intelligence into economic capability.

To effectively operationalise this transition, organisations must navigate four dimensions of structural change: how they secure competitive advantage (strategic positioning), structure value delivery (outcome-based licensing), design intelligence-native operations (intelligence-native enterprises), and augment human capability (workforce augmentation). 

Strategic Positioning

The first strategic decision facing every organisation is not whether to adopt AI, but where intelligence creates durable competitive advantage. Not every business activity benefits equally from intelligent systems. Some activities become commodities as foundation models improve and capabilities diffuse across competitors. Others become increasingly differentiated because they accumulate proprietary knowledge, deployment experience and continuously improving skills. Enterprise strategy therefore shifts from identifying where labour creates value to identifying where organisational intelligence compounds.

Figure 3 categorises organisational strategy along two dimensions: the depth of intelligence embedded within the organisation and the extent to which intelligence reshapes the underlying business model.

Figure 3. The AI Landscape: Where Organisations Play and Compete

Most Indian enterprises currently operate within the productivity quadrant. AI is applied to software development, customer support, document processing and operational optimisation. These deployments improve efficiency while preserving existing organisational structures. Productivity gains create immediate economic value but rarely establish durable competitive advantage because similar capabilities rapidly diffuse across competitors.

The platform quadrant represents a fundamentally different strategy. Organisations redesign products, workflows and operating models around intelligent systems from the outset. Intelligence is no longer embedded within existing processes; it becomes the mechanism through which the organisation itself operates. Competitive advantage derives from continuously improving organisational intelligence rather than reducing operating cost.

At the national level, this progression extends further. India’s long-term opportunity lies in intelligent public infrastructure capable of coordinating markets at population scale. Digital Public Infrastructure demonstrated this principle through identity, payments and document exchange. Similar intelligence infrastructure for agriculture, healthcare, logistics, education and justice could coordinate knowledge across entire sectors rather than within individual organisations.

The engineering quadrant, encompassing frontier model development, semiconductor ecosystems and foundational AI research, remains strategically important. However, these capabilities require sustained investment in research institutions, specialised talent and patient capital accumulated over decades. India’s comparative advantage lies first in operationalising intelligence across its existing economic strengths while progressively expanding frontier research capability.

Monetisation Models

The traditional model of the services industry links revenue to labour, focusing on practitioner hours and project duration. In contrast, intelligence-native models prioritise the delivery of measurable results – such as fraud detection, clinical pathway optimisation, or production efficiency – aligning commercial incentives with the actual value generated for the client. Organisations such as Palantir exemplify this shift, delivering high-stakes decision-making and operational optimisation rather than relying on conventional software subscriptions or hourly consulting. In this paradigm, competitive advantage stems from the reliability and continuous refinement of the underlying intelligence rather than the size of the workforce.

This transition reshapes the internal composition of the enterprise. Engineers can now spend less time executing standardised processes and more time designing, evaluating, and refining executable skills. Organisations can increasingly build repositories of reusable knowledge and evaluation datasets, treating them as primary productive assets. India’s Global Capability Centres (GCCs) are uniquely positioned to leverage this shift. Having historically served as execution centres for global enterprises, GCCs can evolve into intellectual property creators. By encoding deep domain expertise into proprietary agent ecosystems and licensing these solutions back to parent organisations, GCCs transform from cost centres into drivers of national intelligence.

Intelligence-Native Enterprises

The intelligence economy creates new organisational forms whose primary asset is accumulated knowledge rather than software or labour. Each represents a different layer within the emerging intelligence stack.

1. Vertical intelligence companies develop proprietary expertise within narrowly defined domains such as GST compliance, agricultural lending, medical diagnostics or legal reasoning. Their competitive advantage derives from increasingly sophisticated knowledge assets, deployment experience and evaluation data rather than larger foundation models.
2. Orchestration platforms coordinate networks of specialised agents. They decompose complex business objectives into constituent tasks, identify appropriate capabilities, integrate results and provide transparent reasoning to human decision-makers. Their primary value lies in system composition rather than domain specialisation.
3. Skill marketplaces establish markets for reusable expertise. Domain specialists encode accumulated knowledge into executable skills that preserve attribution, version history and continuous improvement. Expertise becomes a productive digital asset capable of repeated licensing independently of the expert’s time.
4. Rural intelligence platforms extend intelligent capabilities to underserved regions through multilingual and voice-first interfaces. Agricultural advisory, market intelligence, financial planning, regulatory compliance and business operations become accessible to first-generation entrepreneurs, farmers and small enterprises without requiring specialist knowledge or English-language proficiency.

These organisational forms share a common characteristic – they accumulate proprietary intelligence rather than labour. Their competitive advantage compounds through learning rather than through scale alone.

Workforce Augmentation

The long-term success of the intelligence economy depends upon how intelligence affects human capability. Much of the current discussion surrounding AI focuses on automation and labour displacement. While these effects are real, they provide only a partial description of the economic transition underway. Intelligent systems also increase the capability available to every worker by expanding access to expertise, reducing information asymmetries and lowering the cost of high-quality decision support.

For India, this distinction is particularly significant. More than half a billion workers participate in sectors where access to specialised knowledge remains limited. A farmer using a multilingual agricultural agent gains access to agronomic expertise previously available only through extension officers. Construction workers can verify skills, improve safety and obtain digital certification. Small businesses gain access to legal, financial and operational expertise previously affordable only to large enterprises. The productivity gains arise not from replacing workers but from increasing the capability available to every worker. It is important to recognise that AI augments expertise but cannot resolve physical or systemic impediments, such as agricultural climate risks or complex administrative hurdles.

The fiscal implications are equally important. Large-scale labour replacement reduces wage income, payroll taxes and welfare contributions while concentrating productivity gains within a relatively small number of firms. Workforce augmentation produces a different outcome by increasing labour productivity while preserving labour force participation and expanding taxable economic activity.

Public policy should therefore distinguish clearly between automation and augmentation. Incentives for AI deployment should favour systems that increase human capability, improve labour productivity and expand participation within the formal economy. Procurement frameworks, tax incentives and national evaluation criteria should explicitly reward augmentation rather than labour substitution.

One possible extension of this principle is Universal Basic Augmentation: ensuring that every citizen has access to domain-specific intelligent systems that improve learning, productivity and decision-making. Unlike income transfers, capability provision compounds over time. Better-equipped workers generate higher-quality knowledge and richer operational data. That knowledge strengthens future intelligent systems, increasing the capability available to subsequent generations.

The intelligence economy therefore completes the second stage of the National Intelligence Flywheel Framework. Knowledge infrastructure encodes expertise. The intelligence economy operationalises that expertise across enterprises, markets and society. The resulting economic activity generates capital, experience and new knowledge that reinforce the national intelligence base. Intelligence becomes not only a technological capability, but an engine of sustained economic growth.

Intelligence Governance

Knowledge Infrastructure defines how intelligence is created. The intelligence economy defines how it is deployed. Intelligence governance defines how it is evaluated, trusted and aligned with national priorities.

Every major technological transition has required new institutions to enable this. Financial markets required accounting standards, central banks and securities regulation. Aviation required certification authorities and air traffic control. The internet required protocols, cybersecurity standards and digital identity. Intelligent systems introduce an equivalent institutional challenge. As AI increasingly participates in high-consequence automated decisions, governance becomes part of the infrastructure itself rather than an external regulatory function.

This transition toward embedded, infrastructure-level governance is critical to create the stability required for mass adoption. Rather than constraining innovation, the objective of intelligence governance is to reduce uncertainty; organisations will invest more readily, governments will procure more confidently, and citizens will adopt more rapidly when intelligent systems can be evaluated consistently, compared objectively, and governed transparently.

To address this, we must organise governance around three complementary functions: AI assurance, which validates technical capability and viability of the system; national measures, which provides the visibility and data necessary for evidence-based policy; and national standards, which establish the foundations for cross-system coordination and public trust. Moving beyond high-level objectives toward verifiable impact requires us to translate these functions into specific engineering practices, which we explore across the following three domains.

AI Assurance

The first governance challenge is establishing whether an intelligent system is genuinely capable of performing the task it claims. As organisations progress from isolated productivity tools to mission-critical deployments, distinguishing substantive capability from marketing claims becomes increasingly important. Investors require objective criteria for allocating capital. Governments require consistent procurement standards. Regulators require mechanisms for assessing risk. Enterprises require confidence that deployed systems will perform reliably under operational conditions.

Figure 4. Assessing Agentic AI Viability

Figure 4 proposes a three-stage framework for assessing the viability of agentic systems.

1. The first stage evaluates technical capability (Is it really AI?). An intelligent system should demonstrate autonomous decision-making within clearly defined operating boundaries, adapt through repeated interaction and improve over time. Systems that merely execute predetermined workflows or static decision trees should be classified as conventional software regardless of the terminology used to describe them. Technical capability alone, however, provides little indication of economic value.
2. The second stage evaluates commercial viability (Is it viable?). A technically sophisticated system remains unsuitable for production if it cannot generate positive economic outcomes. Viability therefore depends upon measurable performance, sustainable unit economics and organisational readiness. Evaluation methods such as Human-in-the-Loop assessment and LLM-as-a-Judge frameworks provide quantitative evidence of quality, while inference costs, latency and operational reliability determine whether deployment is commercially sustainable. Equally important, organisations must redesign workflows around Human-AI teams rather than inserting intelligent systems into existing processes without broader organisational change.
3. The final stage evaluates operational assurance (Is it future proof?). Production systems require explicit observability, comprehensive audit trails, security controls, reproducible execution and clearly defined accountability mechanisms. High-consequence applications should include predefined escalation pathways and break-glass procedures that immediately restore human control whenever system behaviour falls outside acceptable operating boundaries. Governance therefore extends beyond model evaluation to encompass the entire operational lifecycle of intelligent systems and alignment with human values or constitutional principles.

The progression from technical capability to commercial viability and finally operational assurance provides a common language through which governments, enterprises and investors can evaluate intelligent systems before large-scale deployment.

National Measures

Governments optimise what they choose to measure. Gross Domestic Product transformed twentieth-century economic policy because it provided a common measure of industrial output. It remains one of the most influential statistical innovations in economic history. However, GDP was designed to measure transactions within an industrial economy rather than the accumulation and deployment of intelligence. Several limitations become increasingly significant in an AI-driven economy. GDP measures production but not capability. It records expenditure on intelligent systems but provides little indication of whether those systems improve national productivity, scientific capability or institutional effectiveness. It records labour displacement and profit generation but cannot distinguish between economies that broadly augment human capability and those where the benefits of intelligence become increasingly concentrated. This distinction becomes more important as the marginal cost of intelligence continues to decline.

National strategy therefore requires complementary measures that evaluate intelligence directly rather than inferring it through economic output. To address this, we must evolve our national dashboard: if GDP measures the industrial output of yesterday, we require metrics that track the health and vitality of the intelligence flywheel itself. To bridge this measurement gap, we propose two complementary indices – each targeting a critical lever of national intelligence.

1. The first, the India Augmentation Index, measures the extent to which intelligent systems increase national capability that may include dimensions such as learning velocity, access equity, research productivity, technological sovereignty, and quality of life. By tracking these dimensions, the index answers a single policy question: Is intelligence becoming broadly available across the economy?
2. The second, the Contribution Economy Index, measures whether the economic value generated by intelligent systems remains aligned with the creators of that knowledge. By evaluating workforce augmentation, expert royalties, and wealth distribution, this index determines whether intelligence expands participation or concentrates economic returns.

Neither index is intended to replace GDP. Rather, they complement traditional economic statistics by measuring dimensions of national capability that industrial-era metrics were never designed to capture.

National Standards

Markets scale when participants share common standards. Every major technological transformation has depended upon institutions that define common rules for interoperability, quality and trust. The Bureau of Indian Standards (BIS) enables manufacturers across thousands of firms to produce interoperable products against common specifications. The International Organisation for Standardisation (ISO) allows organisations operating in different countries to certify consistent management systems and quality processes. The National Institute of Standards and Technology (NIST) provides measurement science and risk management frameworks that shape global cybersecurity and AI governance. The Reserve Bank of India (RBI) establishes standards governing payments, settlement and financial infrastructure, enabling banks and fintech companies to innovate within a trusted and interoperable ecosystem. These institutions rarely determine which products succeed, but they establish the conditions under which markets can function. 

Artificial intelligence requires an equivalent standards ecosystem with three clear requirements:

1. The first requirement is technical interoperability. Agentic systems increasingly exchange context, invoke external tools and coordinate with other intelligent systems across organisational boundaries. Open standards such as the Model Context Protocol (MCP), emerging Agent-to-Agent (A2A) protocols and standardised skill definitions allow intelligence to move between organisations rather than remaining locked within proprietary platforms. Countries that actively participate in defining these standards influence the architecture of future intelligence ecosystems.
2. The second requirement is operational assurance. Organisations require standardised methods for documenting capabilities, measuring performance, auditing behaviour and certifying compliance. Emerging standards such as ISO/IEC 42001, the NIST AI Risk Management Framework, and the OECD AI Principles provide common governance frameworks through which intelligent systems can be evaluated consistently across sectors.
3. The third requirement is public procurement. Governments are often the largest purchasers of technology within an economy. Procurement standards therefore determine which capabilities organisations choose to develop. Rather than prescribing individual technologies, governments should specify measurable requirements for interoperability, observability, security, explainability and domain-specific performance. Vendors remain free to innovate while competing against transparent and nationally consistent standards.

National standards perform an important economic function. They reduce integration costs, increase market competition, prevent vendor lock-in and create predictable conditions for long-term investment. More importantly, they enable governments to procure interoperable intelligence infrastructure rather than isolated AI applications.

Intelligence Governance therefore extends beyond regulation. It establishes the institutional frameworks through which intelligence becomes trusted, interoperable and scalable public infrastructure. Just as BIS enabled industrial manufacturing, ISO enabled global quality systems, NIST established trusted measurement frameworks and RBI created interoperable financial infrastructure, India’s AI standards institutions can shape how intelligence is built, exchanged and governed across the economy. Countries that define these standards increasingly shape the markets that follow.

Economic Participation

Knowledge Infrastructure facilitates the systematic generation of intelligence, while the Intelligence Economy serves as the primary vehicle for its productive deployment. Intelligence Governance provides the essential framework for institutional trust and operational alignment. Finally, economic participation completes the strategic lifecycle, ensuring that the economic value generated by AI systems is reinvested into the knowledge base that helped create them in the first place.

Every economic system develops mechanisms for rewarding the factors of production. Industrial economies compensated labour through wages, capital through investment returns and innovation through intellectual property. The intelligence economy introduces a new productive asset: accumulated expertise. Unlike physical capital or software, expertise is distributed across individuals, institutions and communities. It improves through use, but only if contributors continue to participate in the value it creates. Economic Participation therefore addresses a fundamental question. How should an economy reward the creation of intelligence rather than simply its deployment?

This pillar consists of two complementary mechanisms: skill attribution and capability finance.

Skill Attribution

Current AI systems contain a structural market failure. Experts are compensated for performing work. They are rarely compensated for teaching intelligent systems how to perform that work. A physician contributing decades of diagnostic reasoning to a clinical agent may improve millions of future diagnoses. A tax specialist whose expertise informs a GST compliance agent may increase productivity across thousands of businesses. A manufacturing engineer whose operational knowledge becomes embedded within industrial agents contributes value that persists long after the original engagement has ended.

Current commercial arrangements typically compensate for these contributions once. The resulting intelligent systems continue generating economic value indefinitely. This creates a weak incentive for systematic knowledge sharing. As organisations increasingly recognise the long-term value of domain expertise, contributors may become reluctant to provide their highest-value knowledge unless mechanisms exist for preserving attribution and enabling continued participation in future value creation. Sustainable intelligence economies require incentives that encourage knowledge contribution rather than discourage it.

The intelligence economy therefore requires a new category of digital infrastructure: a Skill Attribution Protocol. The protocol establishes persistent provenance for knowledge contributions throughout the lifecycle of intelligent systems. Every reusable skill, evaluation dataset, workflow definition and expert contribution receives a verifiable identity that survives subsequent refinement, composition and deployment. Attribution becomes an intrinsic property of the knowledge asset rather than a contractual afterthought.

This enables several new economic relationships. Experts license knowledge rather than time. Organisations contribute reusable capabilities rather than static documentation. Communities develop shared intelligence assets while preserving collective ownership. Institutions receive continuing returns from expertise that remains valuable across multiple deployments. 

The construct of a Skills Attribution Protocol draws on version control systems like Git, which successfully established a global standard for authorship and traceability. However, Skill Attribution must go beyond simple provenance; it must function as a financial backbone. By establishing a persistent identity for every knowledge contribution – linking every skill, dataset, or workflow to its creator – the protocol transforms static traceability into active compensation. This creates a verifiable record that enables automated, multi-party revenue sharing, effectively ensuring that contributors participate in the value their expertise generates long after it was first codified. While version control tracks the who, the Skill Attribution Protocol creates the economic mechanism to pay the who.

Capability Finance

Markets allocate capital toward assets they understand. Industrial economies developed financial systems capable of valuing factories, machinery and physical infrastructure. The digital economy developed investment models for software, platforms and network effects. The intelligence economy requires financial institutions capable of recognising knowledge infrastructure as a productive asset.

Many of the capabilities described throughout this paper cannot be financed effectively through conventional venture capital. National agricultural intelligence, multilingual educational agents, judicial assistance systems and public health infrastructure generate substantial social returns while producing commercial returns only over extended time horizons. These characteristics resemble public infrastructure more closely than traditional software businesses.

India therefore requires financial mechanisms designed specifically for long-term intelligence capability.

1. Sovereign Capability Funds invest in strategic intelligence infrastructure whose primary returns accrue across the national economy. Agricultural intelligence, public health systems, multilingual education and judicial assistance represent "mission-oriented" capabilities that benefit every sector while remaining underinvested by purely commercial markets. India’s experience with ISRO, NPCI and Digital Public Infrastructure demonstrates that long-term public investment in shared infrastructure can create entirely new economic ecosystems.
2. Institutional Skill Endowments provide universities, research laboratories and public institutions with mechanisms for converting accumulated expertise into sustainable sources of revenue. Rather than depending exclusively upon grants, institutions develop domain-specific intelligent systems that can be licensed while remaining publicly governed. Research organisations therefore accumulate long-term capability instead of repeatedly rebuilding expertise through project-based funding.
3. Outcome-Linked Capability Bonds connect long-term capital to measurable improvements in national intelligence capability. Returns become linked to outcomes such as workforce augmentation, scientific productivity, language coverage or improvements in the India Augmentation Index rather than short-term commercial performance. Financial markets therefore become participants in national capability building rather than observers of it.

Finally, organisations require standardised methods for valuing intelligence assets. Current accounting standards recognise software, patents and certain forms of intellectual property, yet rarely capture the value of proprietary agent ecosystems, executable skills or continuously improving knowledge repositories. This limits the ability of investors to distinguish organisations building durable intelligence assets from those integrating commoditised AI services.

Developing recognised methods for Skill Asset Valuation would allow capital markets to reward long-term investment in organisational intelligence. Regulatory bodies such as SEBI could establish reporting standards for knowledge assets analogous to existing standards for financial and intellectual capital. Visible valuation frameworks would encourage sustained investment in organisational learning rather than short-term AI adoption.

Capability finance therefore extends beyond funding innovation. It determines which forms of intelligence receive sustained investment and which remain economically invisible.

Conclusion: Building India’s Intelligence Flywheel

India’s strategic objective should therefore not be to build the largest models, nor merely to accelerate AI adoption. It should be to build the institutions that allow intelligence itself to compound. Every Knowledge Transfer session that becomes an executable skill, every enterprise that licenses outcomes rather than labour, every benchmark that improves public procurement, and every expert who shares in the value they create increases the velocity of the national intelligence flywheel.

History suggests that countries rarely dominate because they invent every technology first. They lead because they build the infrastructure and institutions that allow those technologies to improve continuously and diffuse throughout the economy. The intelligence economy will be no different.

The defining competition of the AI era will not be over who builds the most intelligence. It will be over who builds the fastest intelligence flywheel.

Every generation builds the infrastructure that defines the next century. The twentieth century built institutions for industrialisation, scientific research and digital public infrastructure. The twenty-first century must build institutions through which intelligence itself compounds.

A National Knowledge & Skills Institute would convert enterprise and scientific expertise into reusable intelligence. An India Intelligence Deployment Mission would accelerate deployment across industry, research and public services. A National AI Benchmarking & Assurance Authority would establish the standards and evaluation frameworks needed for trusted AI, and a National Skill Attribution & Capability Finance Corporation would reward knowledge creators while financing long-term intelligence infrastructure. These institutions need not all be new organisations; many could evolve from existing ones. What matters is not their form, but their enduring function: to ensure that intelligence continuously compounds as a national capability.