Gaps are emerging between AI’s promise and delivery: Is there an opportunity in this for India’s IT firms?

Over the past year, large technology firms have cut thousands of jobs while increasing investment in AI. These have been framed at least partly as a pivot to an AI-driven future. The storyline is neat: machines will do more, humans will do less and productivity will rise accordingly. Markets seem to favour the narrative. The problem is that the technology has not quite caught up with it.

Across companies, AI is improving productivity, but not in the clean substitutional way headlines suggest. Engineers are writing code faster, but also spending more time reviewing and debugging it. Designers are generating more options, but not necessarily better ones. Analysts are processing more data, but still need to verify what the system confidently presents as fact.

This creates a peculiar situation: firms can feel both more productive and more constrained at the same time. One might call it efficiency with supervision.

For Indian enterprises, this is especially relevant. Historically, Indian businesses have been quick to adopt global technology trends, often using them as signals of modernization as much as tools for efficiency. Cloud, digital transformation, and analytics all followed this pattern. It is now AI’s turn.

But the operational reality in India is often complex. Enterprises deal with fragmented data, legacy systems, regulatory nuance and multilingual environments. AI systems, which thrive on clean data and consistent context, do not therefore perform predictably. Pilot projects may look promising; scaling them across the organization is harder.

That is why claims of AI-led efficiency should be treated with caution. In many cases, layoffs attributed to AI appear to coincide with broader business corrections. Over-hiring during the pandemic years, slowing demand and margin pressure are all part of the story. AI conveniently offers a forward-looking explanation for what is often a backward-looking adjustment.

The phrase ‘AI-led efficiency’ can quickly become a catch-all justification for restructuring. The risk is not just semantic. If organizations reduce capacity based on expected rather than realized gains, they may render themselves under-resourced. Markets may applaud at first, but execution eventually reveals the difference between cutting excess and cutting capability.

That distinction is especially sharp in the Indian IT services industry. For decades, its model has been straightforward: clients outsource work, firms deploy large teams and revenue scales with effort. AI introduces friction into every part of that equation.

With code being generated faster, testing partially automated and documentation synthesized instantly, clients are asking why they should continue paying for hundreds of engineers when AI-assisted workflows appear to promise similar outputs with fewer people.

Yet, the answer is not as simple. AI-generated work still needs validation, integration and contextual understanding. Errors, especially subtle ones, can be costly. Productivity gains are real, but not yet fully bankable, measurable or repeatable.

This creates a margin puzzle. Do firms pass efficiency gains on to clients and risk revenue compression? Or do they retain pricing and justify value through outcomes rather than effort? The likely direction is a gradual shift towards outcome-based models, but such transitions are rarely smooth.

Global capability centres (GCCs) claim to work under a different calculus. Their touted goal is not immediate cost arbitrage, but capability building. They have better access to proprietary data, tighter integration with business processes and a longer view—meaning more freedom to experiment without the pressure of external billing models.

If that is truly so, GCCs are likely to lead in developing domain-specific AI applications that work at scale. They can iterate, fail and refine in ways that services firms, constrained by client expectations and contractual obligations, may struggle to match. Over time, this could widen the gap between the two models, with GCCs becoming centres of innovation and services firms repositioning themselves as integrators and orchestrators.

Through all this, one fact is constant: we are still in the experimental phase of AI adoption. The technology is powerful but uneven. The use cases are expanding but not always stable. The economic implications are significant, but still not fully understood.

The most important shift is not technological but managerial. Leaders are making decisions today based on projections of what AI might achieve tomorrow. Sometimes those projections will prove accurate. At other times, early signals will be misread as structural change. Watch out!

For Indian enterprises and the IT services industry, the opportunity is substantial, but so is the responsibility. Adopting AI thoughtfully requires investment not just in tools but in training, governance and organizational design. It requires recognizing that productivity gains are not the same as capability gains, and that scaling impact is far harder than demonstrating potential. And occasionally, it requires admitting that the machine is impressive but needs an adult in the room.

If you choose to invest, only trust IT services managers who you feel have the best judgement. Most just have herd instinct.

The author is co-founder of Siana Capital, a venture fund manager.