Google Cloud Next 2026 in Las Vegas was, on the surface, another product launch event. Polished demonstrations of the Gemini family of models, partner announcements, and the usual choreography.

The substance, however, was different. Enterprise AI today mostly sits in chat windows and embedded copilots: an employee types a question, the model answers, the employee acts on the answer. Useful, but the human is still doing the workflow.

Google will stop selling such AI assistants — software that answers a question or completes a single task — and has started selling the Gemini Enterprise Agent Platform, a stack of tools that lets a company create, deploy, supervise and audit autonomous "agents".

These act on their own, instead of waiting to be prompted task by task, across the company's systems for hours or days at a stretch. Microsoft, Amazon, Salesforce, and other cloud providers are racing to assemble similar stacks.

It will not matter whose model is smartest, but who can give enterprises the best tools to deploy, supervise, and control armies of AI agents reliably and safely.

Very recently, Nasscom President Rajesh Nambiar told The Core that India’s GCC’s have “truly moved up the innovation curb”.

“The new GCCs have been positioned as AI-first GCCs, which means that they've been able to do a lot more interesting work, own a broader problem for their enterprise, represent much more than probably some of the service providers,” Nambiar said.

However, Google’s shift from AI copilots to autonomous agent operating systems could upend the economic logic of India’s GCC sector by replacing labour-arbitrage with cost-per-task automation.

As global parents and consulting firms seize control of the agent architecture, governance and data layers, most Indian GCCs risk being reduced from strategic hubs to execution back offices unless they quickly move up the value chain into AI oversight, enterprise data and decision-making roles.

GCCs Face Agentic Reckoning

Five points stood out. First, Google is building a workforce operating system, not a chatbot platform. Second, the control layer — these rules decide which agents can act, on whose behalf, with what audit trail — is now more valuable than the underlying models, which are turning into commodities.

Third, the next bottleneck is not AI capability but enterprise context: messy data, weak descriptive labels (technically called metadata), and no shared way for agents to fully understand what the data means.

Fourth, AI pricing will shift from cost-per-prompt to cost-per-completed-task, breaking every pricing model anchored on human time. Fifth, Google has created a $750 million partner fund and put its own engineers alongside Accenture, Deloitte, PricewaterhouseCoopers, KPMG, Capgemini, Cognizant, Tata Consultancy Services, and HCL Tech to roll these AI agents inside their clients’ workflows.

Each shift hits directly India's Global Capability Centre (GCC) sector — the more than 2,000 offshore captive units employing close to two million people that multinationals run from India to handle finance, technology, customer operations and, increasingly, research.

When The GCC Becomes Optional

The first implication is that the parent enterprise no longer needs an offshore headcount to scale its work. It needs a fleet of agents, a governance layer, and a small number of human supervisors.

The capabilities are now generally available on every major cloud provider.

Take, for example, a typical 4,000-person banking captive running global operations for a US or European parent. The largest single work is finance and accounting, followed by customer operations, technology support, and compliance and Know-Your-Customer (KYC) checks. The rest is spread across analytics, project management and back-office support.

A parent that buys the agent platform is buying the option to compress captive headcount in the first four of those five functions by 30 to 60 per cent over two to three years.

The Controls Layer Is The Prize

Identity, audit, permissions and monitoring are the plumbing that lets a company watch what an agent is doing and stop it when needed. These, plus human review checkpoints, all need deep institutional knowledge and the trust to act on regulated work. A GCC ought to be the natural owner.

In practice, almost no Indian GCC owns this layer. Decisions on the agent platform are being made at the parent’s headquarters, in partnership with the consulting firms named in Google’s $750 million fund. Audit and monitoring tools are bought globally.

Identity and permissions sit with the parent’s Chief Information Security Officer. The Indian GCC might run the resulting system, but does not design it.

The exceptions are revealing. A handful of GCCs got themselves into the controls conversation early. The India operations of many US investment banks, a few global pharmaceutical R&D centres, and some semiconductor design houses now host model testing, regulatory checks, and agent rollout for the parent globally. These might come out of the next two years larger and more central. But they are a small minority of the 2,000.

Enterprise Context Is The Bottleneck

Here, the GCC has the strongest theoretical case and the weakest actual delivery. A GCC sits inside the enterprise. It has the messy data, the process exceptions, the institutional memory of why a particular invoice format breaks the standard pipeline. An outside system integrator does not. The GCC should be the natural owner of the agent-readiness work every enterprise is now commissioning.

In practice, the work is going elsewhere. The data function — labels, lineage tracking, the shared definitions and master data records that make AI agent workflows possible — has typically been understaffed in Indian GCCs because it was treated as overhead.

When the parent now wants the agent-ready data estate built, the GCC finds it lacks the depth in data engineering, the seniority in data product management, or the architectural decision rights to lead the work. It is invited to staff the project’s implementation. But the consulting firm leads it.

The painful illustration is the period-end finance close. Every Chief Financial Officer in the Fortune 500 wants the process of finalising the company's books each month or quarter compressed from the industry median of around 8.5 days to 3-4 days. Top performers use agentic reconciliation, automated journal entries, and intelligent variance investigation.

This work requires deep knowledge of the parent's chart of accounts, intercompany flows and consolidation logic — the institutional memory a finance GCC has built over a decade. Yet in most live programmes today, the data engineering and agent design are led by an external firm with the GCC only providing the bench.

The Old Labour-Pricing Model Won’t Work

The GCC value pitch rests on one number: the total cost of an Indian full-time employee versus the same role at the parent’s headquarters. Cost-per-task makes that number irrelevant.

If a task can be completed end-to-end by an agent fleet for a few cents in computing cost and a small share of a supervisor's time, the parent does not care whether the human alternative would have cost ten dollars an hour in Bengaluru or eighty in New Jersey.

The relevant comparison is no longer human-to-human across geographies. It is human-to-agent within the same task.

A GCC contact centre supporting a global retailer might run at twelve to fifteen dollars per resolved customer query in India, against thirty-five to forty at the parent. The cost gap looks healthy.

However, once an agent fleet handles routine queries at well under a dollar per resolved case, the GCC's twelve-dollar number is no longer competitive against the parent's own internal alternative. This is structurally different from every previous shift the Indian services sector has absorbed. Each prior transition still required armies of people. The agentic transition does not.

GCCs Aren’t In The Room

Decisions over architecture, platform selection, governance design, and operating model are all being made by Accenture, Deloitte, PricewaterhouseCoopers, KPMG, McKinsey and Boston Consulting Group in conversations with the parent's C-suite. The GCC is rarely in the room.

The Indian IT services majors that delivered the previous offshoring wave — TCS, HCLTech, Cognizant, Capgemini — are on Google's named partner list and might get a share of the build phase. The GCCs are not on that list. They are being asked to run someone else's design without the client relationship or procurement contract that an IT services firm at least retains.

What Next?

The sector might keep growing in revenue as parent enterprises themselves are growing, the highest-value work is migrating, and the most strategic GCCs are becoming more central. But the headcount story and the revenue story will decouple sharply.

A typical 4,000-person banking GCC will likely be a 2,400-to-2,800-person operation by 2028. Routine finance work, basic technology support, basic compliance and KYC will shrink hard. But three areas will grow: data engineering and AI infrastructure, agent governance and observability, and a small but rising population of senior subject-matter experts who supervise the agents and own the judgement calls that remain.

The GCCs that get this right will own outcomes, hold decision rights in India, and feed people into the parent's C-suite. The ones that do not will quietly shrink, function by function.

Zinnov and Indiaspora's March 2026 analysis found 55% of India's GCC work in the bottom two tiers of their four-tier classification — Commodities and Procedures — directly exposed to AI displacement. The question now is how fast these can move to the other 45%, which the report calls Grey Hair and Rocket Science.

The Cloud Next signal is less about Google than the moment. The architecture of enterprise AI is being packaged into something that can be bought off the shelf. The GCC sector has to decide soon whether it wants to shape how that architecture is rolled out inside its parent enterprises, or merely operate someone else's design. The Indian IT services industry made that choice a decade late. The GCCs cannot afford the same mistake.