For much of the past decade, artificial intelligence (AI) in the enterprise has been largely assistive. It flagged risks, generated insights, and automated discrete tasks, but decision-making remained in human hands. That balance is beginning to change. A new class of systems, commonly described as agentic AI, is now being deployed across Indian enterprises with the ability to plan, act, and adapt autonomously within defined limits.
According to EY India’s C-suite GenAI survey of 200 enterprises, adoption is accelerating. About 47% of organisations now operate multiple GenAI use cases, 10% are scaling them across business functions, and nearly half report that more than 21% of their proofs of concept have moved into production. The shift, EY says, reflects a broader transition as AI evolves “from assistive to agentic”, redefining how work is organised and governed.
Enterprises are increasingly experimenting with hybrid pods of humans and AI agents, combining scale and precision without adding headcount. The appeal is clear with faster decisions, reduced operational friction, and the ability to run complex workflows continuously.
Yet the transition is uneven. EY’s survey shows that 64.5% of enterprises cite data governance and security as “very severe” challenges, while 78% struggle with system integration. Despite this, companies are pressing ahead, often accepting imperfection in exchange for speed.
India’s broader policy environment is also shaping adoption. The IndiaAI Mission, backed by Rs 10,000 crore and 40,000 GPUs, aims to build domestic AI infrastructure and reduce dependence on foreign models. Analysts estimate that this push towards sovereign AI could add as much as $1 trillion to the economy by 2035. Alongside this, enterprises are investing in small language models that require less computation, support Indian languages, and offer greater control over data, reinforcing the appeal of locally governed agentic systems.
This sense of acceleration is echoed by consulting firms working closely with large enterprises. Aditya Priyadarshan, managing director and lead for AI at Accenture in India, describes agentic AI as having reached a decisive inflection point. “Agentic AI has rapidly moved from isolated pilots to real-world, enterprise-scale deployment,” he says. According to Accenture’s research, 86% of C-suite leaders in India now view AI agents as catalysts for transformation and growth.
Priyadarshan argues that the shift is not merely technological but structural. “We’re seeing a fundamental change in how work gets done, with multi-agent systems operating like coordinated teams, observing, deciding, acting and learning across core business functions,” he says. In sectors ranging from banking and life sciences to retail and consumer goods, Accenture is seeing agents accelerate finance and HR cycles, power always-on customer engagement, and enable more precise cross-sell and upsell strategies.
According to Mercer’s latest Global Talent Trends survey, on average, the typical worker in an AI-enhanced workplace will be able to save 36 days in a year and 80% of today’s jobs are likely to be affected by Generative AI. “Our clients, especially in Global Capability Centers, and Professional Services, have seen productivity and efficiencies unlocked, of about 15-20% using AI. The two areas which have seen maximum improvement are customer service operations and coding. One major retailer for example has deployed bots to manage multi-language customer skills that has reduced the need and dependence on translators,” says Nilanjana Dutta, senior director, talent and transformation, Mercer India.
Control & constraint
Despite the growing attention, agentic AI is not a single technology. Instead, enterprises are assembling systems from multiple layers of models, orchestration tools, and governance frameworks. Ashvin Vellody, partner at Deloitte India, characterises the current landscape in three broad categories.
The first is multi-agent workflows, where multiple specialised agents collaborate in coordinated sequences. “Enterprises are increasingly adopting frameworks that enable a planner agent, an executor agent and a supervisor agent to work together,” Vellody says. This category spans several channels. Large enterprise software vendors such as SAP, Salesforce and Oracle are embedding agentic workflows directly into their platforms, tightly coupled with existing business processes. Start-ups and specialist firms, by contrast, focus on narrow use cases and faster deployment through SaaS-based models. Hyperscalers including AWS, Azure and Google Cloud offer integrated orchestration, governance and scalability, while open-source frameworks such as LangChain, LangGraph and CrewAI allow organisations to build customised architectures.
SAP’s approach illustrates how deeply agentic AI is being woven into core enterprise software. Rahul Lodhe, vice-president and head of SAP Artificial Intelligence Technology India, says the company is moving rapidly from pilots to real-world deployment through Joule Agents embedded across SAP’s applications. “These agents act like digital teammates, using SAP’s process knowledge and a business knowledge graph to plan, decide and execute multi-step workflows autonomously,” he says.
SAP has rolled out specialised agents across finance, supply chain, HR, manufacturing, retail and utilities. In finance, agents support cash application, forecasting and dispute handling, helping automate invoice matching and predict late payments. In supply chain and procurement, Joule’s sourcing agents manage supplier risk and adapt to geopolitical disruptions. HR agents embedded in SAP SuccessFactors support performance management and succession planning, with SAP reporting double-digit improvements in engagement and retention among early adopters.
The scale of ambition is significant. SAP plans to deliver around 400 embedded Business AI use cases and roughly 40 specialised Joule Agents, effectively turning its ERP and HCM products into an AI-native “system of intelligence”. Internally, SAP has deployed Joule across HR, finance, sales and travel workflows, accelerating contract booking and automating parts of the go-to-cash cycle. “We have moved beyond simple automation to a proactive system that manages complex corporate tasks at scale,” Lodhe says.
The second category identified by Deloitte is coding agents, which have seen some of the fastest uptake. “Developer productivity is easy to make sense of,” Vellody notes. Enterprise-approved tools such as GitHub Copilot dominate regulated environments, while more autonomous agents such as Replit and Lovable are popular among start-ups and digital-native firms. A newer class of agents focused on legacy modernisation, capable of reverse-engineering COBOL (common business oriented language) databases or monolithic applications, is also gaining traction as companies attempt to modernise systems.
The third category consists of general-purpose agents embedded in widely used platforms such as Microsoft Copilot, Gemini, Perplexity and ChatGPT. These tools increasingly allow users to move from conversation to action, executing lightweight workflows without custom development.
Findability Sciences chief executive Anand Mahurkar argues that the distinction between these tools matters less than how they are deployed. “Agentic AI is best understood as an architecture, not a single model or tool,” he says. In an enterprise context, agentic AI refers to systems that can autonomously plan, decide and execute actions within predefined business and regulatory guardrails. “Predictive systems generate insights; agentic systems act on them,” he adds.
Adoption has been strongest in operations-heavy environments such as manufacturing, agriculture and supply chains, where decisions are frequent and economically consequential. In these settings, agentic AI shifts organisations “from decision support to decision execution”, delivering gains in productivity, cost and resilience. Humans remain essential, but their role changes. “People define objectives, constraints and escalation rules,” Mahurkar says, while agents handle routine execution.
Most enterprises, however, remain cautious about autonomy. Ebix Technologies, which operates across insurance, financial services and healthcare, describes its approach as “bounded autonomy”. Its agents can ingest and classify documents, validate rules, flag risks and initiate workflows independently, but binding decisions and externally visible changes still require human approval. “Work moves from people pushing tasks forward to agents continuously managing flows in the background,” says Gagan Sethi, the company’s chief executive.
This emphasis on control is echoed across sectors. Lenovo, for example, is enabling agentic AI across task-level, workflow-level and multi-agent systems as part of its hybrid AI strategy. In India, deployments are concentrated in IT operations, cybersecurity, finance and engineering, many still in pilot phases. “Agentic AI only works at enterprise scale when autonomy is bounded,” says SK Venkataraghavan, director, solutions and services group at Lenovo India. “Our framing is that agents can act independently or alongside employees, but inside security guardrails and governance. Low-risk tasks can be automated, while high-impact actions involving regulated, financial, or legal decisions require human approval.”
Microsoft’s Dahnesh Dilkhush, chief technology officer for India and South Asia, frames the shift more broadly. “India’s AI adoption is shifting from assistive tools to agentic systems, intelligence that works across workflows and decisions,” he says. He points to what he calls “frontier firms”: organisations that redesign themselves around AI rather than merely adding tools. Examples include Apollo Hospitals’ Clinician Copilot, Mankind Pharma’s SuperAI, and IndiGo’s agentic engines for customer interactions. Enterprise adoption is also scaling rapidly, with more than 200,000 Copilot licences deployed through partnerships with large Indian IT firms.
Productivity gains, talent lacunae
The most visible impact of agentic AI so far has been productivity. In logistics, Delhivery is using agentic AI to re-architect transport operations. “With TransportOne, AI agents plan shipments, negotiate rates, run spot auctions, monitor execution, and reconcile invoices,” says Uday Anand, vice-president of product at Delhivery. “The system runs autonomously in the background, escalating only true exceptions to humans.”
In travel, EaseMyTrip is integrating task-specific and workflow-focused agents to automate routine management and personalise itineraries, while retaining human oversight for complex or sensitive decisions. “The goal is not experimentation for its own sake, but building a future-ready and seamless travel environment,” says chief executive Rikant Pittie.
Similarly, at Booking.com, the focus is on moving beyond recommendations to autonomous trip management. “We are already collaborating with leaders like OpenAI and Microsoft on agentic developments, such as the OpenAI Operator and the Microsoft’s Copilot tool, to envision a future where AI doesn’t just suggest a flight but proactively manages a disruption – automatically rebooking a hotel or suggesting a new itinerary during a delay without the traveller needing to intervene,” says Santosh Kumar, regional manager, South Asia at Booking.com.
During the recently concluded India AI Impact Summit, MakeMyTrip has also announced that it will collaborate with OpenAI to deepen AI-led travel discovery and capture high-intent travel queries. As part of this collaboration, the company uses OpenAI’s APIs to power new AI features in its app, enabling travellers to seamlessly move from conversational inspiration to booking within the MakeMyTrip’s Myra interface.
“Our collaboration with OpenAI ensures that when travellers start their journey through conversation, MakeMyTrip becomes a seamless extension of that discovery process. When AI is anchored in the proprietary travel data and deeply integrated into the marketplace, it moves beyond inspiration to deliver personalised, bookable outcomes at scale. This is about transforming curiosity into confident decisions,” says Rajesh Magow, its co-founder and group CEO.
Healthcare illustrates both the promise and the limits of agentic AI. In cancer care, most deployments in India remain semi-agentic pilots rather than fully autonomous systems. These include tools that flag high-risk radiology findings, triage patients, pre-fill clinical documentation and optimise chemotherapy scheduling. The greatest promise lies in care coordination and longitudinal patient management, not independent diagnosis. “Clinical decisions still require explicit clinician approval,” practitioners emphasise, noting that explainability and trust are critical in oncology.
Behind these deployments lies a growing demand for new skills. A Quess Corp report estimates that agentic AI and specialised GenAI roles will grow 35-40% annually, while the demand-supply gap remains above 50%. Roles that barely existed three years ago, such as AI orchestration engineers, agent safety specialists and agent lifecycle managers, are now in demand. Salary premiums are strongest in senior architecture and governance roles, reflecting the importance of oversight.
Manav Subodh, founder of skilling organisation 1M1B, argues that the real bottleneck is not technology but people. “You still need a human who’s managing agentic AI and AI solutions, not the other way around,” he says. Applied AI, he adds, is creating opportunities across functions traditionally seen as non-technical, from HR and procurement to finance. But he cautions that guardrails must come first. “You can’t build the 100th floor without taking care of your foundation,” he says, pointing to unresolved issues around data security and privacy.
Subodh also sees a longer-term opportunity for India to move beyond implementation and become a product nation for agentic AI, particularly in local and vernacular use cases. Examples include AI systems for state transport scheduling or local healthcare clinics, where indigenous data and context matter more than scale. “Agentic AI by nature is supposed to be local,” he says, arguing that open-source models and local developer ecosystems will be critical over the next decade.
Looking ahead, enterprises expect agentic AI to evolve in several directions, with stronger memory and context, deeper integration with physical systems such as robotics and manufacturing, and more rigorous governance frameworks. Lenovo expects agents to become more personalised and multimodal, operating across devices, edge and cloud environments. Microsoft sees “human-led, agent-operated enterprises” emerging as a new organisational model. Accenture, meanwhile, argues that agentic AI will increasingly serve as the engine of enterprise reinvention, provided organisations invest as much in governance and skills as they do in technology.