Coforge’s AI Reality Check for Indian IT to ‘Strip Away the …

Coforge is witnessing an upward move in its growth trajectory as its big bets on generative AI and agentic AI provide a positive momentum. Just this quarter, the company became the seventh largest Indian IT firm posting ₹3,689 crore in revenue for Q1 FY26, a staggering 56.5% jump year-on-year.

Leveraging AI and adapting to the new tech landscape is the main factor driving this growth, according to CEO Sudhir Singh. At the helm of these AI advancements over the last one year has been Vikrant Karnik as the EVP for digital, data, cloud, and AI at Coforge. 

Karnik avoids abstract promises or lofty slogans as he has seen the technology’s hype cycles over decades starting his career from EY to Capgemini to Genpact, and now at Coforge. “I think AI was always seen as something with a lot of potential but that would never come to realisation,” he said in an interaction with AIM.

Karnik is blunt about the influence AI is having on Indian IT. “AI is stripping away layers of ‘fat’ that once masked underperformance, exposing who is actually adding value.”

The Reality Check

AI hype for Karnik is about speed, decisiveness, and delivering measurable outcomes. In his view, some leaders will inevitably be replaced in this shift. The winners will be those that quickly double down on what works. “I expect some experiments to fail — perhaps four out of ten — but the winners will be those that double down quickly on the six that work,” he said.

According to him, the Transformer architectures and GPT models convinced enterprises that AI could be both valuable and practical, but India presents unique challenges.

With low labour costs, automation doesn’t create the same shareholder value argument as in high-wage economies. “It’s a little bit like asking a hotel to list electricity as a revenue line,” he said. “We’re here to deliver shareholder value for our customers. If AI helps us do that, we’ll use it.”

Despite that, he points to bright spots — such as a large Indian bank that has become one of Coforge’s strongest AI case studies. In Southeast Asia, some of the most complex AI deals involve custom infrastructure and distilled models, which then become examples for Western clients.

From Ghost Pilots to AgentSphere

Coforge has worked with machine learning for over seven years and with generative AI and agentic architectures for more than three. One recurring problem was the “ghost pipeline” — pilots that never make it into production. 

Research with a partner agency showed that pilots with clear business outcomes were three to four times more likely to be deployed.

That principle led to the creation of AgentSphere, a catalog of over 100 AI agents. It wasn’t a product born out of marketing ambition, but from client needs. “We are an engineering company at heart,” said Karnik. “We specialise at the intersection of domain and engineering.”

One client, a company that supplies construction materials, struggled to help seasonal sales associates navigate a 57,000-item catalog. Even with generative AI and RAG, results weren’t accurate enough. 

Coforge built an agentic architecture instead, which revealed the need for a centralised “agent catalog” to manage agents across Salesforce, service systems, and in-house tools. 

The company also launched Forge-X, an AI-powered delivery platform designed to accelerate application development and modernisation across client portfolios.

When it comes to AI, Karnik categorises customers into three groups. First are specific-need clients who know exactly what they want built. Second are the explorers who suspect value in a cumbersome workflow and want to reimagine it.

And the last group are the transformers with executive mandates for large-scale change, such as improving EBITDA by hundreds of basis points. For this, trust is key. Karnik recalls a case where Coforge wasn’t the client’s largest tech provider — Capgemini and Cognizant were involved — but was still chosen for cost-transformation work.

Vibe Coding and Quasar

Like many IT firms, Coforge is experimenting with AI-driven coding. Cognizant, for example, aims to generate 50% of its code with AI within a year, and has already trained 35,000 developers on GitHub Copilot. 

Coforge has tested tools like Cursor and Windsurf, building a secure setup for its internal TechCon conference where engineers could describe applications and have them generated by on-premise LLMs.

“We didn’t just connect to public LLM APIs,” Karnik explained. “The minute you use an external model, your code is going outside. We wanted something customers could run in their own environment.”

This approach led to Quasar, launched in May last year. Powered by 23 LLMs — including OpenAI’s GPT models, Google’s Gemini, and open-source alternatives — it enforces 90 coding rules to keep AI-generated code maintainable. Coforge often starts with Mistral’s CodeStral for performance reasons but tests new entrants, including OpenAI’s recently open-sourced models.

Clients, such as a large specialty insurer, have also used Coforge’s environment to run secure AI hackathons. “The point isn’t to sell products — it’s to show what a real, working framework looks like,” Karnik said.

With AI and automation, IT firms are debating whether to stick to headcount-based billing or move to outcome-based models. For Karnik, the answer is simple: “If we deliver more, we will grow. If we deliver less, we will shrink.”

The post Coforge’s AI Reality Check for Indian IT to ‘Strip Away the Fat’ appeared first on Analytics India Magazine.

Generated by RSStT. The copyright belongs to the original author.

Source