In 2025, Indian companies embraced generative AI with unusual speed, launching pilots across functions and industries. What followed was a more sobering phase: Moving those experiments into production exposed weaknesses in data foundations, cost models, governance, and operating structures.
According to Nisheeth Srivastava, Chief Technology and Innovation Officer for India at Capgemini, organisations learned that success at scale depends on discipline. Cloud costs became harder to ignore, fragmented data limited reliability, and trust — spanning security, compliance, and accountability — emerged as a defining constraint.
The divide between stalled initiatives and those that reached production, he says, came down to early architectural and organisational choices.As attention turns to agentic AI and cloud-to-edge architectures, enthusiasm is being tempered by realism. Indian enterprises are prioritising control, governance, and resilience over autonomy, rethinking cloud-first assumptions formed years ago.
In the Q&A that follows, Srivastava discusses what enterprises got wrong while scaling AI, where agentic ambitions need restraint, and which architectural assumptions CIOs must now unlearn.
Edited excerpts:
2025 saw widespread experimentation with GenAI across Indian enterprises. From your vantage point, what were the hardest realities organisations encountered when they tried to scale AI beyond pilots?
In 2025, Indian enterprises recognized that scaling GenAI beyond pilots required stronger foundations and far more intentional design. As GenAI workloads grew, organizations gained clearer visibility into cloud consumption patterns and strengthened cost-management approaches to support ROI. Many also focused on improving data quality and governance only to discover that fragmented data estates and integration gaps limited consistent performance across use cases.Trust, security, and compliance became priority areas, encouraging more robust safeguards for responsible AI-practices.
At the same time, companies learned that moving from experimentation to scale demanded process redesign, operating model evolution, and meaningful workforce readiness. Ultimately, success depended on balancing innovation with discipline and intentional design, and reshaping systems to reliably support AI at enterprise scale.
What organisational or architectural decisions made the biggest difference between AI initiatives that stalled and those that went into production?
AI initiatives advanced to production when organizations made early architectural and organizational decisions that enabled scale. Standardized, well-governed data created a dependable foundation, while shared, scalable AI platforms reduced integration debt and prevented the fragmentation that plagued pilot-only deployments. Clear ownership across business, technology, and risk accelerated decision-making and addressed concerns proactively.
Equally important, companies redesigned processes and operating models so AI could be embedded into daily workflows rather than remain experimental. In contrast, initiatives stalled when data was fragmented, architecture was not built for scale, accountability was diffused, and legacy ways of working remained untouched, making pilot success challenging to reproduce at enterprise depth.
Agentic AI is being positioned as the next leap. Where do you see real enterprise readiness today and where do you see over-enthusiasm?
Enterprise readiness for agentic AI is strongest in areas with well-defined, rules-based workflows such as software development support, IT operations, service management, and internal productivity. In these environments, agents operate with clear triggers, constraints and human oversight, enhancing execution and accelerating routine decisions rather than acting autonomously.
Over-enthusiasm tends to appear when agentic AI is assumed capable of full end-to-end autonomy across complex, cross-functional decisions, capabilities that remain at the early stage. Companies are instead progressing steadily, strengthening governance, clarifying operating models, and defining decision boundaries before expanding the scope and autonomy of AI agents.
What guardrails (technical or governance-led) become non-negotiable as AI systems begin to act with greater autonomy?
As AI systems gain more autonomy, strong guardrails become non-negotiable. Clear human oversight and intervention paths are essential so decisions can be reviewed, escalated, or corrected when required. Governance frameworks must define who is accountable, who makes decisions, and how much autonomy is permitted across different workflows.
Technical controls should set explicit boundaries on what AI systems can do, supported by continuous monitoring, drift detection, and auditability. Strong data governance, security practices, and traceability help track actions and outcomes.
Well-defined operating models and roles ensure responsibility remains unambiguous as autonomy increases, enabling safe, predictable and enterprise-grade deployment.
How is the cloud-to-edge conversation evolving specifically in India across cost, latency, data localisation, and sectoral needs?
In India, the cloud-to-edge conversation is becoming more pragmatic as enterprises balance performance, compliance and costs. Organizations are rethinking architecture to better manage cloud expenses created by variable, consumption-based pricing, while latency-sensitive workloads are increasingly moving closer to where data is generated, especially in operational and real-time settings. Data-localization norms are also reshaping decisions, with sensitive and regulated data increasingly processed or stored locally at the edge, while central cloud platforms continue to support analytics, governance, and orchestration. Manufacturing, telecom, retail and transportation are leading this shift, adopting cloud-to-edge models to enable real-time operations, regulatory compliance and cost efficiency.
In 2026, what architectural choices should CIOs re-evaluate if they are still operating with cloud assumptions formed five years ago?
Enterprise architecture must evolve beyond cloud-only assumptions to meet the demands of speed, resilience, and AI-driven operations. The future requires seamless cloud-to-edge architectures where locality, speed, and resilience matter as much as scale. Monolithic platforms must give way to composable, API-driven designs that accelerate integration, reduce vendor lock-in and support distributed inference.
As AI moves from generative to agentic, architecture must enable systems that act within defined boundaries, not just provide recommendations. At the same time, trust must be foundational, with security, interoperability, and transparency embedded by design. Ultimately, technology choices should be guided by business outcomes, operational resilience, and long-term value creation.
Sustainability is now part of most technology narratives. Where have you seen green technology investments deliver measurable business outcomes and where has the ROI been overstated?
Green technology investments have delivered measurable outcomes when sustainability is tied to operational efficiency. Cloud optimization, workload rationalization, and energy-efficient infrastructure have reduced costs while lowering emissions, and data-led ESG platforms have improved accuracy and regulatory compliance. In these cases, sustainability and business value have moved lockstep.
ROI has been overstated when green initiatives were treated as standalone upgrades rather than embedded in operations. Investments focused mainly on offsets, reporting tools without process integration, or pilots that never scale, often fail to show financial impact. Sustainability creates value when woven into core architecture, operations, and decision-making.
Looking ahead, what is one mindset or operating assumption CIOs must unlearn as they step into 2026?
In 2026, CIOs must unlearn the mindset that AI is experimental, cloud-only models are sufficient, and technology can operate without integrated trust and sustainability measures. Artificial intelligence is now central to enterprise transformation, demanding systems designed for continuous learning, autonomy with guardrails, and workflow-level redesign. The future will not be cloud-only but a seamless interplay between cloud, edge, and distributed data.
Customers and employees alike will expect fluid, hyper-personalized interactions from local-language AI to immersive augmented reality/virtual reality (AR/VR) environments and digital twins. As digital interactions multiply and data becomes more decentralized, embedding cybersecurity, transparency, and digital ownership into everything is essential.
