The agentic AI stack is getting crowded. Orchestration could become the real enterprise advantage

Over the last two years, the enterprise AI market has expanded rapidly. Foundation models have multiplied. Cloud providers, enterprise software companies, and startups are embedding agents into their platforms. Enterprises themselves are building agents for customer service, software development, finance, IT operations, sales, and knowledge management. BCG’s analysis of about 4,000 companies shows agentic AI funding growing more than 60% a year since 2023, with over half of it going into the platforms on which agents are built and run.

The result is that the enterprise technology stack is becoming more intelligent, but also more crowded. According to Gartner, enquiries on multi-agent systems rose 1,445% in roughly a year. IDC finds that more than 70% of enterprises expect to buy prebuilt agents, build their own, and use agents embedded in existing software, all at the same time.

The fact most of the enterprises today get wrong is that deploying more agents does not automatically create an agentic enterprise.

The AI aspirants

Nasscom’s study of over 500 global enterprises found that 90% are still AI aspirants, experimenting or running isolated use cases. Even among the leading 10% who have reported publicly, only 38% have moved a quarter or more of their agentic pilots into production. BCG’s survey shows the same thing from another side: nearly 60% of enterprises have not yet seen measurable cost improvement in deals that include agentic AI.

While there could be multiple reasons for this, the most common factor is coordination. An enterprise may have multiple agents interacting with customers, retrieving information, working inside its CRM and triggering an ERP workflow. Each may work well independently. The harder problem begins when a business process requires them to work together. Many CIOs I meet worry about agent sprawl the way they once worried about shadow IT.

Most large companies operate across multiple clouds, generations of applications, SaaS platforms, proprietary systems, and large estates of legacy technology. Their data sits across functions and geographies. Processes contain approvals, exceptions, regulatory controls, and decisions where human judgement remains essential. Agents will have to operate inside this environment rather than outside it. This is why orchestration should not be mistaken for simply linking a few agents together or building a more sophisticated workflow.

Operationalising AI at scale

The second is viewing orchestration primarily as an IT integration problem. It is increasingly a business architecture question. It determines how work gets distributed between agents and people, where human intervention is necessary, and how accountability is maintained. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. It is not about agentic capabilities. Today the bigger challenge for enterprises is whether they can operationalise agentic AI reliably and at scale.

And this is precisely where I see one of the biggest opportunities for India’s technology services industry lie.

For decades, technology services companies have operated at the intersection of new technology and complex enterprise environments. From client-server and ERP to cloud and mobility have added multiple new layers. Making those layers work together, securely and at scale, became the foundation of India’s IT services business. Agentic AI follows the same pattern, only faster and at a much greater level of complexity.

Mckinsey predicts Agentic AI to unlock $300 - $400 addressable TAM by 2030 for tech services. BCG also estimates agentic AI can open up to $200 billion in net new value for tech services over the next five years. And we are already seeing substantial movement in this space. TCS has partnered with GitLab to orchestrate agents across the software lifecycle, building industry-specific workflow templates. Cognizant has brought ServiceNow's agents into its open-source multi-agent accelerator, so clients can coordinate agents across platforms without custom connectors. IBM and Yotta have launched sovereign Agentic AI cloud in India that is designed specifically for Indian enterprises to manage data securely.

The next services opportunity is continuous

Perhaps the most interesting aspect of orchestration is that it does not end at deployment. We will continue to see agents and models evolve. Enterprise data will change. Regulations at the same time will also evolve. Their performance, cost and risk will need to be continuously monitored and optimised.

That creates the possibility of a new generation of managed AI services, where providers will not just maintain applications but continuously orchestrate an intelligent enterprise environment.

For Indian technology services companies, capturing this opportunity will require change as well. They will need stronger combinations of AI engineering, domain knowledge, architecture, security and process expertise. Reusable IP and orchestration platforms will matter. So will commercial models that increasingly link technology interventions to business outcomes rather than effort.

As agents proliferate, becomes cheaper and more capable, orchestrating them to work together, safely and for a real business result, will be the real challenge. That is where I firmly believe the real opportunity for India’s tech services industry will lie, that is orchestrating intelligence better than anyone else.

(The author is President, Nasscom. Views are personal.)