BTG Advaya
AI has made tremendous progress in the last decade, reshaping industries and influencing every aspect of modern life. From advanced language models like GPT to intelligent recommendation systems, AI has demonstrated capabilities once confined to science fiction. Yet, the long-standing dream of AGI, i.e., a system capable of performing any intellectual task that a human can, remains elusive. What we have today are highly specialised systems, excelling in narrow domains but falling short of true human-like thinking and autonomy.
In this evolving landscape, a new avatar of AI is gaining attention – agentic AI. Unlike traditional AI models that require explicit prompts and instructions, agentic AI are systems that can act autonomously, making decisions, planning workflows, and achieving goals with minimal human supervision. Does this mean that we finally there? Have we built machines that can think, decide, and act without our involvement?
Vision of AGI
AGI is the holy grail of AI research, a system that can comprehend, learn, and apply knowledge across an extensive range of tasks. It has the flexibility and adaptability of the human mind. Current AI systems, while impressive, remain far from this goal. They are narrow, designed for specific areas such as language processing, image recognition, or strategy games. Even the most advanced models, like GPT-5, cannot continuously learn or reason like humans. As OpenAI’s CEO Sam Altman recently admitted, GPT-5 still “falls short” of AGI because it lacks persistent memory, true understanding, and self-directed learning.
Agentic AI, however, promises to move us closer to that vision. These systems are designed to perform multi-step workflows, integrate with tools, and make autonomous decisions within a defined scope. They can book a restaurant, schedule meetings, compose emails, or even write code, without constant micromanagement. But despite these capabilities, they are bound by their programming and training data. True AGI, capable of adapting to unfamiliar situations with human-level reasoning, is still years or even decades away.
Rise of Agentic AI
While AGI remains a distant goal, agentic AI is a practical and near-term reality. In mid-2025, OpenAI launched a ChatGPT-based “agent” that can browse the web, interact with files, and perform tasks like making reservations or drafting business proposals. Google and Anthropic have introduced similar autonomous assistants. These systems combine powerful language models with tools, like browsers, calendars, spreadsheets, to execute multi-step activities.
Corporations worldwide are racing to integrate agentic AI. Oracle and Salesforce are promoting “autonomous agents” to improve productivity. Indian IT giants are actively piloting these tools with surveys showing over 80% of Indian firms exploring agentic AI for automating routine tasks. From customer support to software development, agentic AI promises significant efficiency gains.
However, industry analysts caution against excessive optimism. Gartner predicts that 40% of current agentic AI initiatives will be abandoned by 2027 due to high costs and disappointing returns. Today’s agents, while impressive, are still fragile. They require frequent human intervention and struggle with unstructured problems. They also lack genuine understanding of real-world context.
India’s AI Landscape
India has rapidly emerged as a global hub for AI adoption and innovation. According to the Stanford Artificial Intelligence Index Report 2025, India ranks second worldwide in AI skill penetration and is among the top 10 countries in private AI investment. The government’s ambitious IndiaAI Mission, backed by $1.25 billion, aims to build world-class AI research infrastructure.
Indian companies, particularly in IT services, banking, and telecom, are deploying generative and agentic AI in a range of applications. Customer support bots, automated coding assistants, and AI-powered analytics platforms are becoming increasingly common. Some firms are experimenting with AI-driven recruitment systems that shortlist candidates, schedule interviews, and even draft performance reviews.
Yet, this enthusiasm comes with challenges. Experts warn of potential job displacement, especially in white-collar sectors, and the risk of algorithmic bias. For instance, recruitment tools may inadvertently penalise candidates who lack who lack native-level English proficiency. These issues highlight the urgent need for legal and ethical frameworks to govern AI deployment in India.
Global Regulatory Landscape
While technology races ahead, regulation struggles to keep pace. Around the world, governments are grappling with how to manage the risks of AI, including agentic systems. The European Union has taken the most comprehensive approach with its AI Act, which is being enforced in phases. The Act adopts a risk-based framework, imposing stringent requirements on high-risk applications, such as those in healthcare, finance, education, or biometric identification. Providers of high-risk AI must conduct risk assessments, maintain detailed records, and ensure effective human oversight.
The EU has also updated its liability framework. Under the revised Product Liability Directive, AI systems are treated as products, making developers strictly liable for harm caused by defects. This means that if an autonomous agent causes injury, physical or financial, the manufacturer can be held accountable.
In contrast, the United States has opted for a lighter regulatory touch. There is no dedicated federal AI law. Instead, regulators rely on existing statutes and issue non-binding guidelines. The Federal Trade Commission has warned against unfair or deceptive AI practices, but enforcement remains fragmented. While this approach encourages innovation, it leaves gaps in accountability, particularly when autonomous systems cause harm.
Asia presents a mixed picture. China has enacted stringent rules on algorithmic transparency and fairness, holding firms accountable for misuse. Japan relies on voluntary guidelines and model contracts to allocate responsibility in AI transactions. India, meanwhile, is still formulating its strategy. The Digital Personal Data Protection Act, 2023, governs data usage but was not designed for AI. A draft bill to create a National AI Technology Regulatory Authority is under discussion but remains unpassed.
Across jurisdictions, one principle is clear, AI systems do not have legal personhood. Responsibility lies entirely with the humans, i.e., the developers, operators, and businesses, behind the technology.
Why Agentic AI Isn’t Fully Here Yet
Despite the excitement, several technical barriers prevent agentic AI from reaching its full potential. Current systems can follow instructions and even break down complex tasks into steps, but their reasoning is occasionally weak, and their understanding of the real world is superficial.
Language models, the backbone of today’s agentic AI, generate responses based on patterns in data, not genuine comprehension. They lack common sense and struggle in unfamiliar contexts. Moreover, they cannot autonomously acquire new knowledge or continuously learn in dynamic environments, which is a key prerequisite for AGI.
Experts widely agree that human-level AGI is still a distant prospect. Gartner predicts that by 2028, 15% of routine work decisions will be made autonomously by AI agents, and about one-third of enterprise software will feature some agentic capabilities. While breakthroughs are possible, most researchers anticipate incremental progress rather than sudden leaps.
Legal Challenges – Liability, Data, and IP
As agentic AI becomes more prevalent, it raises complex legal questions, particularly in the areas of liability, data protection, and intellectual property.
a) Liability and Accountability
Who is responsible when an AI agent causes harm? Traditional legal systems assign fault to human actors, but autonomous agents blur these lines. No major jurisdiction has granted legal personhood to AI, so accountability typically falls on the developers or operators. India’s Consumer Protection Act and product liability provisions could apply if a defective AI system causes physical or financial harm. However, these frameworks are designed for tangible products and foreseeable defects, not for autonomous learning systems that make unexpected decisions. The IT Act offers safe harbour to passive platforms, but an AI agent that actively modifies content or takes independent actions would fall outside this protection. Companies deploying such agents face direct liability risks, emphasizing the need for clearer rules or mandatory insurance models.
b)Data Protection and Privacy
Agentic AI’s reliance on vast datasets creates compliance challenges under privacy laws like the EU’s GDPR and India’s DPDP Act. Both frameworks emphasise informed consent, purpose limitation, and accountability, these principles are difficult to reconcile with autonomous systems that adapt and repurpose data dynamically. This obliges companies to ensure human oversight for critical uses such as credit scoring or contract terminations. In practice, “human-in-the-loop” mechanisms will remain essential for legal compliance.
c) Intellectual Property
Perhaps the most unsettled question concerns ownership of AI-generated works. Indian copyright and patent laws require a human author or inventor. If an AI system independently creates a poem, design, or algorithm, that work arguably falls into the public domain. This legal gap discourages innovation and complicates commercial use. Legal scholars propose various fixes, such as treating the human user as the author or creating new work-for-hire rules for AI-generated content. Until the law evolves, companies will need to ensure meaningful human contribution to claim IP rights over AI-assisted creations.
Preparing for the Future – What India Should Do
India stands at a critical juncture. Its rapid AI adoption offers economic opportunity, but it also increases legal, ethical, and social risks. A balanced regulatory framework is the need of the hour.
The DPDP Act should be supplemented with explicit provisions for automated decision-making and dynamic consent. A risk-based approach, similar to the EU AI Act, could classify AI applications by potential harm, imposing stricter obligations on high-risk systems. Intellectual property laws should be revised to address AI-generated works, ensuring incentives for innovation while safeguarding public interest. India’s proposed Digital India Act could integrate these reforms, creating a legal foundation for the AI era.
Industry, too, has an important role. Businesses deploying agentic AI must establish ethics boards, conduct algorithmic audits, and draft contracts that clearly allocate risk. Human oversight should remain central, especially in sensitive domains like finance, healthcare, and law.
Conclusion – A Wave, Not a Tsunami
So, are we at AGI? The answer is no—not yet. Today’s agentic AI systems are impressive, but they are far from human-like intelligence. However, the wave is building. Over the next decade, agentic AI will become more capable and more integrated into everyday life. This progress demands proactive governance. India has a chance to lead by crafting forward-looking regulations that balance innovation with accountability, privacy, and fairness. If we get it right, autonomous agents will enhance productivity, fuel creativity, and eventually transform our economy, without leaving us in chaos. Agentic AI may not yet be the “thinking machine” of our dreams, but it should not be ignored.
