IN THE SUMMER OF 2023, the conversation about artificial intelligence in enterprise settings was dominated by a single word: copilot. AI was the tireless assistant that summarised emails, drafted reports, suggested code completions, and answered questions in natural language. Impressive, certainly. Transformative, in a limited sense. But fundamentally reactive — the AI waited for a human to ask, provided an output, and stopped. That paradigm is now obsolete. The defining shift in enterprise AI in 2025 and 2026 is not a new model or a new algorithm; it is a categorical change in how AI systems relate to action. Agentic AI — systems that perceive their environment, reason over goals, plan multi-step workflows, execute those workflows across connected systems, and adapt based on outcomes — is moving from pilot projects to operational infrastructure at a pace that is reshaping industries faster than regulation, governance frameworks, or organisational culture can comfortably absorb.
The conceptual distinction is deceptively simple but operationally profound. A traditional AI system, including the most capable large language models, is a tool: it responds to inputs and produces outputs. An agentic AI system is an agent: it maintains persistent goals, monitors its environment for conditions that require action, selects and deploys the appropriate tools from its available repertoire, executes multi-step tasks across APIs, databases, and software systems, evaluates outcomes, and corrects its trajectory when results diverge from objectives. The difference between a doctor who answers questions and a doctor who monitors your vitals continuously, detects anomalies, orders appropriate tests, and adjusts your medication regimen without waiting to be asked — that is the difference between generative AI and agentic AI.
The Numbers Behind the Shift
The scale of enterprise adoption is significant and accelerating. Gartner projects that by 2026, 40 percent of enterprise applications will embed task-specific AI agents, a dramatic increase from low single-digit penetration just two years ago. By 2028, the research firm estimates that 33 percent of enterprise software applications will include agentic AI, enabling 15 percent of day-to-day work decisions to be made autonomously — up from zero in 2024. A spring 2025 survey by MIT Sloan Management Review and Boston Consulting Group found that 35 percent of respondents had already adopted AI agents. Among IT leaders, 93 percent report intentions to introduce autonomous agents within the next two years, with nearly half already having implemented some form of agentic system (MuleSoft and Deloitte Digital, 2025). Investment follows intent: year-on-year AI spending is expected to grow 31.9 percent between 2025 and 2029, potentially reaching $1.3 trillion by 2029, according to IDC.
Boston Consulting Group’s analysis of deployed agentic systems reports that AI-powered workflows can reduce low-value human work time by 25 to 40 percent and accelerate business processes by 30 to 50 percent in areas ranging from finance and procurement to customer operations. An SAP-integrated supply chain agent, for example, can detect that logistics costs are trending upward, autonomously trigger a reassessment of procurement forecasts in the connected finance system, compare alternative supplier options, flag the optimal choice, and initiate the contracting workflow — all without a human generating a report, convening a meeting, or issuing an instruction.
Healthcare: The Most Consequential Deployment
Among the industries deploying agentic AI, healthcare stands out for the magnitude of both the opportunity and the stakes. The healthcare sector already reports one of the highest AI agent usage rates — 68 percent in a 2025 survey — and Accenture projects AI applications could generate up to $150 billion in annual savings for the global healthcare industry by 2026. More concretely, 40 percent of healthcare executives already use AI agents for inpatient monitoring and early warning systems, with full implementation of agentic clinical AI expected within three years (IBM, 2025). Frost and Sullivan projects that AI-powered imaging solutions could prevent up to 2.5 million diagnostic errors annually.
AtlantiCare Health System in New Jersey deployed an agentic AI clinical assistant in 2025 that handles ambient documentation — listening to patient-physician interactions and generating clinical notes in real time, reducing physician administrative burden by an estimated 30 percent per consultation. A 2025 research paper from MIT, examining an AI agent deployed to detect adverse events in cancer patients from clinical notes, found that the most demanding aspect of implementation was not model training but data engineering, stakeholder alignment, and workflow integration — consuming 80 percent of the total project effort. This finding, sobering as it is, actually reveals the maturation of the field: the algorithmic challenges are largely solved; the organisational and infrastructural challenges are the real frontier.
The Governance Problem
Autonomy introduces accountability. When an AI agent executes a consequential action — approving a loan, adjusting medication, rerouting a supply chain — the question of who bears responsibility for errors cannot be answered by pointing to a model weight. Deloitte’s 2025 Emerging Technology Trends study reveals a concerning implementation gap: while 68 percent of organisations are exploring or piloting agentic AI, only 11 percent have systems in active production, and 35 percent have no formal agentic strategy at all. Among those with deployments, governance — establishing clear human accountability, audit trails, and escalation procedures — is cited by enterprise leaders as the most critical unresolved challenge.
Gartner has warned that over 40 percent of agentic AI projects will fail by 2027 because legacy enterprise systems cannot support the real-time data access and API connectivity that agentic architectures require. The transition from a reactive AI tool to an autonomous agent demands what Deloitte describes as a paradigm shift in data infrastructure — from conventional pipelines to enterprise-wide search and indexing systems, contextualised through knowledge graphs, that allow agents to find and act on relevant information without laborious ETL processes.
India’s Agentic Moment
For Indian enterprises, the agentic shift arrives at a moment of significant readiness. India’s digital infrastructure — UPI, Aadhaar, the Account Aggregator framework, and deepening cloud penetration — creates a data environment potentially suited to agentic deployment at scale. Indian IT services companies including TCS, Infosys, Wipro, and HCL have all articulated agentic AI strategies and are actively building agent orchestration platforms for enterprise clients globally. The question for Indian industry is not whether to engage with agentic AI but how to govern it — particularly in regulated sectors like banking, insurance, and healthcare, where autonomous decision execution requires regulatory frameworks that Indian policymakers have not yet fully developed. The 2025 amendments to India’s Digital Personal Data Protection Act begin to address some of these concerns, but sector-specific agentic AI governance guidelines remain an urgent gap.
-Rashmi kumari
