Agentic AI - Warm Prospects and Cold Realities

Large Language Models (LLMs) have slipped quietly but unmistakably into everyday life. They draft emails, summarise documents, generate lesson plans, translate languages, and even script parts of films and advertisements. Customer service agents, HR recruiters, students, coders, and designers now lean on conversational AI as routinely as they once did on search engines. The influx of these systems into daily routines is so rapid that their presence is more often felt than noticed. What was once science fiction has become mundane utility.

India is at the heart of this transformation. The country has emerged as one of the largest global bases of generative AI adoption, with OpenAI counting India as its second-largest user market worldwide. In fact, India has the world’s highest proportion of student users of ChatGPT - evidence that the next generation is already fluent in these tools. 

Reflecting this scale, OpenAI announced earlier this month that it will open its first office in New Delhi in 2025, a landmark moment for the Indian AI ecosystem.

Affordability has played a decisive role. The ChatGPT Go plan, launched at Rs 399 a month specifically for India, has expanded access by offering greater speed, capacity, and functionality at a price point far lower than in Western markets. 

This democratisation of AI has been complemented by a growing domestic ecosystem: startups like Sarvam AI are building multilingual models, the government has announced billion-dollar AI infrastructure plans, and regional innovators are creating tools for languages as diverse as Tulu, Bhojpuri, and Tamil. Together these developments position India not just as a consumer but as a potential innovator in global AI.

Also Read: The Last Station on the Train Line

This far it’s a tool based narrative. Productivity is the primary lever with a human in the loop and possessing agency. It is against this backdrop that the conversation around imminent “agentic AI” has gathered momentum. Agentic systems promise a step beyond conversational tools as they are designed not merely to respond but to act. In theory, such agents could initiate a process, plan its execution, call external tools, and deliver an outcome without human oversight. A sales agent could autonomously draft, negotiate, and close deals. A logistics agent could route shipments and resolve delays. A medical scheduling agent could optimise resources across a hospital chain.

The promise is compelling but the reality is far more limited. Current experiments show agents can indeed sequence tasks and interact with APIs. Yet they falter in unpredictable environments, often hallucinate, and frequently rely on hidden human interventions to succeed. In practice, they remain fragile prototypes.

For India, the challenges are sharper. Agentic AI requires not only powerful models but also clean, interoperable data systems. Most Indian enterprises, even in advanced sectors, struggle with silos, inconsistent taxonomies, and fractured governance. Without reliable data foundations, autonomous systems risk automating error rather than progress.

Institutional readiness is another brake. Indian enterprises are still designed around human accountability. If a machine autonomously approves a loan or alters a supply chain contract, who is responsible when things go wrong? Regulatory frameworks are not yet prepared for such dilemmas, and risk-averse sectors like healthcare and finance will take years of negotiation before delegating real authority to machines.

Also Read: Attention CXOs: Culture, not code, will decide your future in the AI era

Society may be even less adaptable. Indians, like citizens elsewhere, are comfortable with automation so long as a human remains nominally in control. Removing that presence triggers anxieties about fairness, ethics, and legitimacy. People may welcome an agent that books a train ticket but resist one that determines insurance payouts or employment eligibility. Trust in agentic systems will have to be earned slowly, and likely only in low-stakes contexts at first.

What does this mean for the future? The likely trajectory is not a sweeping revolution but gradual, uneven adoption. India will see early successes where data quality is strong and risk is low: logistics optimisation, customer service, back-office operations, and targeted marketing. More heavily regulated or sensitive domains will move slowly and cautiously.

This does not diminish the significance of the shift. The lesson for Indian leaders, whether in government, business, or education, is to prepare deliberately rather than reactively. That means investing in strong data governance, building institutional frameworks for accountability, and cultivating new workforce skills in AI oversight and integration. It also means embracing India’s linguistic and cultural diversity as an asset: vernacular AI may be the bridge that brings millions into the fold of the agentic future.

The story of AI in India is thus one of both promise and realism. Large language models are already reshaping daily life here more visibly than in many other countries. Agentic AI could, over time, extend that transformation—but only if data, institutions, and society evolve alongside the technology. 

The prudent course is cautious optimism. India is well positioned to benefit, but the path will be measured in decades, not quarters. Those who prepare soberly will ultimately be better placed than those who fall for hype.