Cybage says enterprise AI needs redesign, not another layer

Enterprise technology has traditionally evolved in layers, with new capabilities added without disturbing existing systems. AI, however, has not followed that pattern. At the recent NASSCOM event, ‘Scaling Enterprise AI – From Vision to Value,’ Aneesh Nathani, VP, Data & AI Service Line, Cybage Software, spoke about how enterprise systems are being redesigned. Leading Cybage’s Data and AI service line, he represents a generation of technology leaders focused on practical enterprise AI adoption. During his session, ‘Masterclass: Enabling the Agentic Enterprise’, he captured the current industry sentiment in simple terms: “Some days it feels like an opportunity. On others, it feels uncertain.” That tension continues to shape enterprise adoption. Over the past decade, organizations have invested heavily in systems built to capture data and generate insights through reporting. These systems continue to function well, but their limitations are becoming clearer in environments where data is continuous and decisions need to happen faster. Part of the uncertainty also comes from changing market dynamics, with enterprises moving beyond being application providers while SaaS companies increasingly embed AI into their products. Aneesh’s central argument was that the issue is not the absence of intelligence in systems, but the way systems themselves are structured. Most enterprise architectures were designed to support execution while decision-making sat outside them. As response cycles accelerate, that separation is increasingly becoming a constraint. Designing Systems for Continuous Context The approach he outlined focuses on systems where intelligence is embedded directly into workflows rather than added as a separate layer. This means enabling agent-led execution where context is continuously carried forward. He pointed out that many current AI initiatives remain stuck in experimentation. He described several of these as “Dead-on-Arrival”: AI sandboxes that never reach production, bespoke RAG projects quickly absorbed into foundational models, and use-case libraries lacking real engineering depth. “Every pitch now has the same AI slide,” he noted, adding that differentiation is disappearing in quarters, not years. Industry data reflects this challenge. McKinsey’s 2026 State of AI Trust report suggests that only around 30% of organizations have achieved meaningful maturity in AI strategy, governance, and agentic AI controls, despite accelerating adoption. The gap is not in ambition or investment, but in the architecture and governance required for AI to operate reliably at scale. The focus, therefore, is shifting from adding AI features to redesigning systems where decision-making becomes part of operations. Enterprise data remains distributed across systems, making coordination and integration critical. The session also explored possible scenarios for enterprise AI evolution. In one scenario, models become commoditized, shifting value toward engineering capability and proprietary data. Another, considered more likely, sees platforms and agent frameworks converging, with differentiation driven by integration and workflow design. Operating Within Defined Boundaries The agentic enterprise, as Aneesh described it, is less a technology category and more a way of structuring systems so they can operate autonomously within defined limits. In this model, systems handle specific tasks within set parameters, while people remain responsible for oversight and governance. The objective is not to remove human involvement but to shift human focus toward control and decision supervision. This model is already visible in operational environments where systems can respond to inputs and execute actions without waiting for manual intervention. Addressing one of the biggest barriers to enterprise AI adoption, risk, Aneesh emphasized that governance must be built into systems rather than added later. Security and compliance can no longer function as administrative layers introduced after deployment. Instead, systems need to be designed so actions are tied to permissions and business rules from the outset, making accountability and traceability inherent to the architecture itself. This shift also changes how technology delivery is measured. Traditionally, delivery has been evaluated through effort and timelines. As outcomes become increasingly dependent on system design, organizations are beginning to seek clearer alignment between work and business impact, moving toward outcome-based delivery models. Rather than measuring how many employees “use” AI, Aneesh suggested that enterprises should focus on measurable efficiency in outcomes. He referenced engineering agents for pull requests and unit testing, as well as operational agents for claims filing and legal benchmarking, as examples of how businesses can automate end-to-end processes. Cybage is already applying this philosophy internally through SmartPal, its platform designed to democ