Indian Enterprises Turn to ‘Agentic Watch Guards’ to Police AI & Human Behaviour

Indian enterprises are increasingly turning to a new layer of artificial intelligence built specifically to monitor other AI systems, and the humans who work alongside them. 

According to EY’s report, The AIdea of India: Outlook 2026, nearly 65% of Indian companies identify data governance and security as severe challenges in AI adoption, with adoption often occurring outside formal technology or risk functions. 

The same report notes that 58% of global capability centres in India have already invested in agentic AI, with a further 29% planning to scale. Gartner estimates that 40% of enterprise applications will carry task-specific AI agents by the end of 2026.

This scale of deployment has produced what industry observers now informally call “agentic watch guards”, supervisory systems sometimes themselves AI agents, tasked with monitoring prompts, tool calls, data access and decisions made by both autonomous software and staff. 

Unlike earlier compliance tools that reviewed outputs after the fact, these systems are designed to intervene during execution, before an action such as a payment, a data transfer or a client communication is completed.

The trend has gathered pace as India’s largest IT services firms scale generative and agentic AI across their own workforces. Microsoft confirmed in June that TCS, Infosys and Wipro had each expanded Microsoft 365 Copilot deployments beyond one lakh employees, taking their combined total past three lakh seats. 

TCS reportedly employs over 200 full-time staff solely to monitor Copilot usage and tune governance tools, while Infosys is said to be building a blockchain-based audit trail for agentic activity and Wipro requires manual approval for external-facing AI actions.

Enterprises Reframe Oversight as a Shared Discipline

Industry leaders describe the shift as treating AI agents much like new employees who have not yet earned full trust, requiring close supervision until their behaviour is proven reliable and consistent with company policy.

Supratik Shankar, Co-Founder at Dview, says TCS, has publicly talked about embedding agentic AI into daily operations. He said TCS has enhanced over one lakh associates through the Copilot rollout, framing the effort as central to building an AI-first culture and a Human-plus-AI operating model for the future. 

He has also argued that embedding agentic AI into daily workflows is reshaping how work gets done across the company, with employees increasingly collaborating with monitored, policy-bound agents rather than unsupervised tools.

Pari Natarajan, Chief Executive and Co-Founder of Zinnov, has framed the same shift in terms of enterprise value rather than headcount reduction. 

“AI is a transformative force to reimagine enterprise value, Infosys is extending that ambition across delivery, engineering and corporate functions as adoption scales past one lakh employees,” Natarajan says.

He has also indicated that oversight mechanisms are being built directly into agentic workflows, ensuring that expanding autonomy does not undermine accountability or client trust in regulated engagements such as banking and telecoms.

Banking, where regulatory scrutiny is highest, is emerging as an early adopter of layered oversight combining human and machine review. 

Ratan Kumar Kesh, Chief Operating Officer at Bandhan Bank, has pointed to the Reserve Bank of India’s Framework for Responsible and Ethical Enablement of AI, known as FREE-AI, as a natural starting point for deployment. 

He has said the bank is being extra careful not to take undue risks around customer protection, treating the RBI framework as a baseline for responsible rollout. 

Kesh has also suggested that continuous, auditable monitoring of both agents and employees allows the bank to expand automation gradually, rather than granting full autonomy before governance and risk controls are proven.

A Structural Change

EY’s analysis argues that agentic AI represents a fundamental change in what governance must achieve. Rather than validating whether a single output was correct, organisations must now continually assure that an autonomous system behaves as intended over time, across shifting workflows and dependencies. 

The report recommends treating AI agents as privileged users, subject to least-privilege access, time-bound permissions and explicit allow-lists for the tools and systems they may touch.

This creates an added layer of complexity for large, federated Indian organisations, EY notes, where ownership of AI initiatives is often distributed across business units rather than centralised within a single risk or technology function. 

“That fragmentation increases the likelihood of so-called shadow autonomous agents operating without formal sign-off, a risk that dedicated watch guard systems are partly designed to surface and close,” Kesh adds.

Sector-specific adoption is already visible beyond banking. Channel research firm CRN Asia reported that Wipro launched a dedicated AI-Native Business and Platforms Unit in April, aimed at enterprise-scale agentic deployments delivered on an outcome basis, while Infosys has established a centre of excellence with Anthropic, beginning with telecommunications and expanding to financial services and manufacturing. 

Regulated sectors, where governance and compliance requirements are strictest, are cited as the areas of strongest current demand for continuous oversight tools.

For now, most Indian enterprises appear to be layering watch guard systems onto existing human review structures rather than replacing supervisors outright. 

Analysts caution that as the number of AI agents grows into the thousands across a single organisation, purely manual oversight will not scale, making some form of automated, always-on monitoring of both AI and human activity increasingly unavoidable across the sector.