Why the Indian IT Bench is No Longer a Nursery

In the traditional Indian IT pyramid workforce model, large volumes of campus-hired freshers fuelled bench growth and served as a cushion against demand fluctuations. That model is now shifting.

Companies are now focusing on hiring developers with specialised skills and doubling down on resource utilisation. This structural reset is not merely a response to market cooling; it reflects a deeper shift in how the industry values human resources in the age of agentic AI.

As software giants move away from volume-driven hiring, the era of the “safe-harbour” bench is giving way to a high-velocity, precision-led model, where code is increasingly generated by machines and validated, rather than written, by humans.

Part of this change is partly driven by Indian IT companies onboarding agentic developer tools such as Cursor IDE and Cognition’s autonomous coding agent Devin. These are not conventional vendor relationships aimed at marginal productivity gains, but experiments in embedding AI agents directly into the software delivery stack and rethinking how work is split between humans and machines.

One of the clearest signals of this shift has come from Cognizant, which has partnered with Cognition to explore the use of Devin across enterprise software development. The aim is to accelerate delivery, particularly by automating routine engineering tasks.

Other large technology firms are also testing similar tools. Last month, Infosys partnered with agentic coding platforms Cognition and Cursor as part of its broader AI-led delivery initiatives.

Outside Indian IT, ServiceNow has partnered with Anthropic to integrate Claude and Claude Code across its global workforce, citing faster engineering cycles and reduced preparation time in sales and delivery functions.

The implications of such engagements extend beyond productivity claims, impacting hiring models, training pathways and the economics of the bench itself.

In a conversation with AIM, Jaspreet Bindra, Co-founder and CEO of AI literacy initiative AI&Beyond, says Cognition-style engagements represent a structural shift rather than a narrow efficiency play. Autonomous agents such as Devin, he notes, compress large parts of L0 and L1 work—routine coding, basic debugging, and documentation—without eliminating junior roles outright. Instead, they raise the bar for what qualifies as entry-level work.

“As foundational execution increasingly moves to machines, skill formation migrates upstream, happening before hiring rather than safely on the IT bench,” Bindra adds. The bench then ceases to function as a training buffer and instead becomes a short-term holding pool for already deployable talent.

That philosophical shift is now visible in operational policy. Tata Consultancy Services has reduced the time employees can remain on the bench to about 35 days, sharply curtailing a system that once allowed freshers to spend months in training or awaiting redeployment. Across the sector, utilisation thresholds have tightened as firms seek to align human capacity more closely with real-time demand, aided by AI tools that absorb foundational tasks.

At the same time, the nature of engineering work itself is changing. Karthik Vaidhyanathan, Assistant Professor at the Software Engineering Research Center at IIIT-Hyderabad, notes the early “zero-to-70” phase—establishing basic structure and functionality—is increasingly automated, allowing teams to reach a usable baseline much more quickly.

“The real expertise now lies in taking that 70 to 100,” Vaidhyanathan notes. That final stretch requires a deep understanding of organisational context, business constraints, and system integration—areas where human judgement remains critical. As a result, junior engineers are being pushed to think less like coders and more like architects, reviewing and contextualising AI-generated output.

Entry-level roles, he adds, are not disappearing; rather, they are being redefined. Where syntax proficiency once sufficed, abstraction, reasoning and critical evaluation are becoming essential. Fundamentals matter more, not less, because they enable engineers to judge whether AI-generated code is truly production-ready.

From an organisational perspective, this shift also challenges long-standing cost structures. The traditional logic of labour arbitrage—replacing one expensive engineer with several cheaper ones—weakens when AI delivers significant productivity gains. While AI tools are not free, they are often cheaper than adding incremental headcount, resulting in more selective hiring and faster upskilling expectations.

Yet the bench's retreat as a training layer carries risks. Kamal Karanth, Co-founder of staffing firm Xpheno, recalls that bench rationalisation began well before agentic AI entered mainstream delivery. Post-pandemic overhiring between 2021 and mid-2022 pushed bench levels as high as 16–18%, he explains, before margin pressures forced a reset to leaner levels of around 10–13% or lower.

“Agentic AI accelerates this trend by absorbing entry-level and low-cognitive tasks,” Karanth adds. Over time, he expects the bench to become more conditional and specialised, focused on niche skills and just-in-time deployment, supported by staffing partners rather than large internal pools.

However, shifting skill formation outside the enterprise through lateral hiring or contractors only defers costs. Without sustained fresher hiring and internal build engines, Karanth warns the industry risks a future supply crunch at mid and senior levels, driving cost inflation and weakening long-term engineering depth.

“We are already seeing early signals of this dynamic in niche AI and data roles, where limited availability has driven sharp cost inflation,” he notes, adding that without a balanced approach where AI adoption is coupled with deliberate investment in building human capability, the industry risks hollowing out its future expertise base.

“Sustaining long-term competitiveness will require balancing AI-led efficiency with a renewed build strategy, ensuring that productivity gains today do not undermine engineering depth tomorrow,” he signs off.