Industry leaders hail India as a proving ground for next-generation agentic AI

Artificial intelligence has long played the role of a backstage helper, but the past two years have pushed Indian enterprises to expect far more. As operations grow more complex and customer volumes surge, companies are no longer asking how AI can support humans; they are testing how it can act on its own. Agentic AI is now slipping into core workflows, and India, with its scale and unpredictability, has become the perfect proving ground.

This shift set the tone for a closed-door roundtable hosted by YourStory in collaboration with Microsoft on Day 2 of TechSparks 2025 at Taj Yeshwantpur. Moderated by Ipsita Basu, Senior Director at YourStory, the session, themed ‘Building the future with Agentic AI’, brought together a uniquely diverse cohort of operators and technologists who sit at the frontlines of high-volume, high-stakes decision-making every day.

Around the table were Ajit Pendse, Senior Director - Technology at ElasticRun; Abhishek Kanthed, Lead - Data Science and Infrastructure at Aereo; Sunil Rai, Senior Director of Engineering at Magicpin; Kush Khurana, Head of AI and Machine Learning at Adda247; Jayprasad Hegde, Senior Vice President - Engineering, Research & Platform at TookiTaki; Hemant Mangla, Head of Engineering at Namma Yatri; Hariharan Ravishankar, Chief AI Scientist at Intangles; Akshat Mandloi, Co-founder of Smallest AI; Vipin Kumar, Chief of Staff to CTO (Director Engineering) at Swiggy; Aniruddh Jain, CPTO at Cars24; Rishabhdhwaj Singh, SVP Engineering at Urban Company; and Vishal Gupta, VP Engineering and Operations at Netradyne. Representing Microsoft were Kevin Scott, Corporate Vice President - Global Enterprise Sales, and Nitin Mittal, Digital Natives and Startup Leader for India.

Autonomy is arriving in fragments

Across sectors as varied as mobility, ecommerce, education, logistics, and compliance, attendees described a similar pattern where AI agents are no longer limited to answering questions or assisting workflows. They are now inspecting vehicles, running customer calls, qualifying leads, checking rider or partner behavior, auditing service quality, reviewing documents, and coordinating follow-ups, often at volumes that would be impossible with human teams alone.

For many of the companies present, these systems are already folded into day-to-day operations. Voice agents, in particular, are seeing rapid adoption, handling high-frequency interactions with increasing sophistication.

The practical roadblocks: Latency, cost, and Indian reality

Even as use cases expand, the challenges are hard to ignore. Latency remains a persistent obstacle, especially for voice-driven interactions, where even short delays cause drop-offs. Several leaders described instances where technically sound flows falter simply because the response speed isn’t fast enough for real users.

Cost was an equally present concern. High-quality speech synthesis, multi-dialect support, computer vision workloads, and real-time inference all create heavy load on infrastructure. While models are improving, the economics of fully autonomous workflows remain fragile for many companies.

Beyond these, participants repeatedly pointed to India’s operational environment in the form of messy data, unpredictable user behavior, infrastructural constraints as the ultimate stress test. The consensus was that autonomy must contend with conditions that rarely match lab assumptions.

A new kind of engineering talent

One of the clearest shifts discussed was the changing profile of talent needed to build AI-driven systems. Companies are now looking for engineers who can understand user context, design flows, refine prompts, run audits, and supervise agent behavior. Coding ability alone isn’t enough.

Some leaders noted that their teams have already created roles dedicated to overseeing agents, reviewing decisions, examining edge cases, and improving logic over time. This supervisory layer is becoming essential as agents take on more responsibility.

For companies working in regulated or safety-critical environments, reliability remains non-negotiable. Several attendees highlighted the risks posed by hallucinations or inconsistencies, especially when AI touches compliance, financial decisions, or road safety. Multi-layered verification, physics-grounded checks, and bounded autonomy were frequently cited as the safeguards needed before scaling further.

Microsoft’s call to redefine how AI should be viewed

Microsoft’s Kevin Scott steered the conversation toward a broader reframing where instead of treating agentic AI as software, consider it a form of digital human capital. In this view, agents aren’t tools but contributors with defined responsibilities, learning curves, and limits. This shift, he noted, creates room for hybrid work models where digital agents and human teams operate alongside each other rather than in opposition.

Participants engaged strongly with this framing, given they already see agents generating analyses, improving scripts, running A/B tests, and uncovering patterns that human teams would struggle to identify consistently.

India’s testbed advantage

What became unmistakable through the conversation is that Indian enterprises are advancing agentic AI in a uniquely grounded way, driven by the pressure of day-to-day operations. High transaction volumes, diverse user behavior, and the scale of frontline interactions give Indian companies a distinct vantage point.

In doing so, they are laying out a blueprint for how agentic AI can function inside complex businesses where autonomy is a practical necessity, and where the measure of success is not novelty but reliability.