As Artificial Intelligence evolves, so should the architecture behind the systems. It is not enough to bolt on to AI; it must be integrated into data, infrastructure, applications, and user experience.  

Traditional architecture is no longer fit for the pace and demands of AI evolution, said Bala Prasad Peddigari, chief innovation officer (Technology Software and Services Business Group) at TCS, as he cut to the heart of a pressing industry dilemma, during his address at the NIIT StackRoute Digital Architect Conclave 2025.

In today’s world, he said, traditional architectures are showing cracks. “We’re facing performance bottlenecks, scalability limitations, and integration complexities.” Many pilots in AI remain unscalable because they lack foundational AI-ready architectures.

Peddigari questioned the audience: “Are architects drafting blueprints, or are they navigating AI battlefields?” The answer, he believes, is clearly the latter. 

Most architects are reacting to the speed of change, rather than leading it. He cited a CTO who admitted buying AI tools without preparing the underlying architecture to support them, something he called the “architect’s dilemma”.

Tracing the journey from rudimentary client-server models to today’s AI-native imperatives, Peddigari set the tone with an anecdote, an encounter with a manufacturing firm plagued by 100 system shutdowns a day. 

The issue wasn’t just technological. “Is it an architectural issue, a philosophy issue, or is it a code issue?” were the questions asked, according to him. The answer, as it turned out, lay in their mounting technical debt and an inability to adapt.

Peddigari and his team replatformed the legacy-heavy operation by integrating IoT sensors and AI, reducing shutdowns to zero. The assembly line ran uninterrupted for 90 days. The transformation, he said, was driven not merely by technical fixes, but by deep understanding of business domains, architecture, and the discipline of marrying tech trends to real-world problems. 

He compared this approach to Aadhaar’s simplification, from 20 fields to six, arguing that “removing the unnecessary” is what makes systems more intelligent, leaner, and functional. The crux of his talk was about architecting AI-native platforms, a term he views as both, a mindset and a capability. 

AI-native platforms need this

Peddigari presented a layered framework that breaks down the building blocks of AI-native platforms. The central thesis, according to him, is that intelligence must be embedded from the ground up. 

This means designing systems that are capable of data feedback loops, real-time monitoring, and “observability” — all vital to keeping AI systems aligned and operational. Data fabric, Peddigari stressed, is the core. Without intelligent pipelines, robust catalogs, annotation and privacy measures, models cannot thrive. 

From there, he pointed to MLOps as essential for governing the lifecycle of models, from experimentation to drift detection. He also emphasised the importance of the experience layer, where AI becomes meaningful to users, and the trust layer, where fairness, safety, and explainability must be designed in, not added later.

The need for autonomous agents, orchestration layers, and LLM-powered cognition also surfaced in his breakdown of a modern digital intelligence framework. He advocated for a platform mindset, solutions must be scalable, not siloed. This includes planning for extensibility, low-code/no-code participation, and vibe coding acceleration, while recognising their current limitations at enterprise scale.

The three levels of maturity

He mapped out three levels of organisational AI maturity: siloed experimentation, sandboxed development, and true enterprise AI platforms. The last, he said, is where the vision lies , not only democratising AI, but institutionalising it. He called for cross-functional teams to collaborate, breaking silos between decision-makers, developers, data scientists, and architects.

Peddigari touched on the importance of GenAI Ops, governance, observability, and evaluation mechanisms tailored for generative models. He also explained how prompt engineering, plug-ins, and hierarchical agents must work in concert with authentication and orchestration to drive real business applications.

In their LinkedIn posts, Mahmoud Abufadda, senior consultant (Enterprise architecture, digital transformation, AI) at Moro Hub, and John Chavner, senior principal at Strategic Technology Advisory Services, both warn that AI is compelling enterprises to confront long-ignored architectural flaws.

Abufadda stressed that business architecture must evolve in step with AI’s rapid advances, embedding it into strategy, processes, and IT assets to avoid fragmented, unscalable efforts.
“Organisations must therefore incorporate AI into their strategic planning and enterprise architecture roadmaps, ensuring they build the capabilities today that will be standard in their industry tomorrow,” he said.

Chavner pointed out that legacy systems, designed for stability and monolithic operation, are fundamentally misaligned with AI’s requirements for seamless data access, real-time processing, and modular integration. Both caution that without foundational modernisation, AI initiatives risk stalling and eroding competitiveness.

“The cost of modernisation is substantial, but the cost of inaction is existential. Companies that delay AI implementation while competitors advance their capabilities face a future where their market positions become increasingly untenable,” said Chavner. 

Peddigari’s concluding advice to architects was to start small, build iteratively, and embed AI into every workflow with intention. The goal, he said, is to move from interfaces to conversational interactions, from blueprints to real-time navigation. 

Peddigari ended on a note of certainty. AI-native architecture is not optional. “From silicon to software systems, massive investments are being made. No industry will be left out,” he said, pointing to trends like GenAI and Quantum AI as the next frontiers.

Source link