By Sachin Dev Duggal, Founder & CEO, SekondBrain.ai
We spend a lot of time talking about artificial intelligence, but I increasingly think the real question is not about artificial intelligence at all. It is about human intelligence, human intent and whether countries like India choose to compete in yesterday’s AI race or define the next one.
The instinct, of course, is to build a bigger model. A better Indian model. A “Sarvam 3.0”, a “Sarvam 4.0”, or whatever the next national benchmark becomes. I say this with enormous respect for what Sarvam and others are doing. Building world-class Indian language capability is important, and frankly long overdue. India cannot be digitally sovereign if hundreds of millions of people are forced to interact with technology through someone else’s language, idiom or cultural lens.
But if India defines AI leadership simply as building a larger version of what frontier labs have already built, we may spend billions arriving late to a race whose rules were written by someone else. The frontier model race is capital-intensive, chip-intensive and increasingly shaped by geopolitics. India should participate, but it should not confuse participation with strategy.
The frontier itself is also changing. The last few years were about larger models and datasets. Now, much of the frontier is moving towards reasoning systems that deliberate for longer, break tasks into steps and spend more inference-time compute trying to reach better answers.
That is useful. But it is not enough.
A longer chain of reasoning over ungrounded information is still ungrounded. Making a machine deliberate for longer does not automatically make it know anything true. It may simply become better at navigating uncertainty, sounding plausible and constructing a path through a fog it still cannot see.
The way I think about it is this: everyone is racing to build a better navigator, the AI equivalent of a more intelligent GPS that can reason, plan, route and reroute. But a navigator is only as good as the map underneath it. If the map is incomplete, outdated or hallucinated, no amount of clever navigation will reliably get you to the truth.
Nobody has really built the library.
By library, I do not mean a static database or another document store. I mean a grounded, versioned, trustworthy substrate of knowledge that an AI system can reason over. A living structure where concepts are connected, facts have provenance, contradictions are visible, time matters and confidence is represented. A system that knows not just what it is saying, but why it believes it to be true.
This is where I think India has a genuinely different opportunity. The next AI race may not be won by the country that builds the largest model. It may be won by the country that builds the most trusted knowledge substrate.
For India, this matters at three levels: education, enterprise and sovereignty.
Start with education. For the last two hundred years, education systems were largely built for a world of scarcity. Information was scarce, expertise was scarce and execution was expensive. So we trained people to remember, specialise and perform a craft.
AI changes that equation. Software can be written faster. Analysis can be generated faster. Research can be summarised faster. Content can be produced faster. Increasingly, execution itself becomes abundant.
So what becomes scarce?
I suspect the answer is intent.
Knowing what problem to solve. Knowing which question to ask. Knowing which trade-offs matter. Knowing what should exist in the first place. The deeper danger is that humans stop developing intent because the craft becomes so easy.
This has profound implications for India. If we continue educating children for a world where success is defined by memorisation and standardised output, we may be preparing them for a workplace that is already disappearing.
The future may require less narrow specialisation and more systems thinking. Less rote memorisation and more judgment. More comfort at the intersection of technology, philosophy, design, economics, psychology and communication.
In a strange way, AI may push us back towards the age of the polymath. The industrial era rewarded specialisation because work could be divided into repeatable parts. The AI era may reward people who can connect those parts back into a whole.
India does not simply need more people who can use AI tools. It needs people who can define intent for AI systems, challenge their assumptions, understand their limits and apply them in culturally grounded ways.
The second opportunity is enterprise knowledge.
Every organisation has years, sometimes decades, of accumulated expertise: documents, emails, tickets, code, contracts, presentations, meeting notes, customer conversations and decisions. Yet ask a company why something was decided, what failed before or who knows the answer, and suddenly the organisation becomes strangely forgetful.
This is not because the information does not exist. It is because it was stored as words, not meaning.
Humans do not fundamentally think in documents. We think in concepts, relationships, experiences and memories. Language is the compression format we use to transmit those concepts to another brain. The mistake is assuming that mastering the compression format means mastering the underlying concept.
That is why simply putting a language model on top of enterprise documents often disappoints. It can summarise, search and generate. But without a deeper structure of memory and meaning, it does not really know where it is.
The next generation of enterprise AI will not just retrieve documents. It will help organisations remember what they already know. A customer complaint should connect to a product decision, which connects to a design trade-off, a previous meeting, a regulatory concern and perhaps a piece of code written three years ago.
That is not search. That is institutional memory.
The third opportunity is sovereignty.
AI is not just another software layer. It is becoming a cognitive infrastructure. If a nation’s businesses, schools, courts, hospitals and public services become dependent on systems they do not control, sovereignty is no longer only about borders, energy or defence. It is also about memory, values and meaning.
Every sovereign will eventually ask: Whose memory is this system built on? Whose values does it reflect? Whose interpretation of history, law, culture and society does it understand? Who decides when access is granted, restricted or withdrawn?
This is why the AI sovereignty debate cannot stop at chips and models. Computing matters enormously, but computation is not the whole story. A nation also needs sovereignty over context. It needs trusted knowledge systems, cultural grounding and the ability to decide what truth means in its own institutional setting.
India is unusually well placed here, not because India is simple, but because it is complicated. It operates across languages, regions, cultures, economic backgrounds and legal realities at a scale few countries can truly understand. Meaning changes with language, geography, history and lived experience.
That complexity is not a weakness. It may be India’s greatest AI advantage.
The mistake would be to believe that the only path to leadership is building a bigger version of someone else’s model. The better question is: what can India build that the rest of the world has not yet understood it needs?
I think the answer is the library.
A living, time-versioned, trustworthy graph of concepts. A system where knowledge is not just stored, but grounded. Where facts carry provenance, confidence is represented and the system can know what it knows, when it was true and on what authority. Where human beings remain in the loop not as clerks correcting outputs, but as stewards of meaning.
Many of the components already exist in open models, knowledge graphs, retrieval systems, verification layers and human-in-the-loop workflows. The challenge is architectural, not just computational. India does not need to win the GPU war to lead here. It needs to build the layer the frontier labs have underweighted while they scale the reasoning engine.
They are racing to build a better navigator. India can build the library.
That is the more interesting sovereign AI project. Not a model that merely speaks in Indian languages, but a knowledge substrate that understands Indian context. Not just an assistant that answers questions, but an intelligence system that preserves institutional memory across education, enterprise and government.
Where AI goes from here will not be defined only by who can generate the most convincing sentence. We already have machines that can produce language with extraordinary fluency. The next frontier is whether they can be grounded in truth, memory and intent.
For India, the choice is not whether to participate in the global AI race. It must. The question is which race it chooses to run.
If we chase only bigger models, we may remain dependent on other people’s infrastructure, capital cycles and policy decisions. If we build the trusted knowledge layer underneath AI, we may create something more durable: sovereign memory, grounded intelligence and an educational paradigm fit for a world where execution is abundant but intent is scarce.
We often ask whether AI will replace humans. I think history may ask a different question.
Did we build machines that merely generated language, or did we build systems that preserved and amplified human knowledge?
One creates better chatbots.
The other becomes civilisation’s memory.