That assumption is becoming harder to defend. Inside offices, call centres, software teams, and government departments, work is moving differently. Decisions are taken faster. Drafts are completed sooner. Errors are caught earlier. Coordination takes fewer rounds. None of this shows up clearly in gross domestic product (GDP) or labour productivity data. And yet, it is impacting everyday work.
India’s headline productivity numbers barely register this change. There is no sudden jump in GDP. Wages have not surged. Labour productivity statistics show little sign of an “AI dividend.” To many observers, this seems to confirm a familiar suspicion — that artificial intelligence (AI) is being oversold. That conclusion may be too quick. The problem may not be AI. The problem may be how productivity is measured.
Most productivity metrics were designed for an industrial economy in which intelligence was scarce, slow, and closely tied to human labour. Output rose when firms hired more workers, installed more machines, or extended working hours. AI does not fit neatly into that model. It delivers cognitive assistance without adding labour or time in the usual way. As a result, many of its effects pass through organisations without appearing clearly in the statistics.
This puzzle is not unique to India. In the United States, labour productivity growth has averaged around 1–1.5 per cent a year for more than a decade, with no obvious acceleration even after the widespread release of generative AI tools in late 2022. This is striking because adoption has been unusually fast. By 2024, nearly a third of US workers were already using generative AI at work.
High adoption, modest productivity growth — that tension now sits at the centre of the global AI debate. Look more closely, however, and the picture becomes less puzzling. At the level of individual tasks, AI’s impact is often substantial. Field experiments in customer support show average productivity gains of around 14 per cent, rising to over 30 per cent for newer or less-experienced workers, while response quality improves, error rates fall, and employee attrition declines. Studies of management consultants report faster task completion and smoother coordination.
These are meaningful gains, but they rarely translate into higher measured output. Firms use them to shorten turnaround times, reduce training costs, standardise responses, and smooth internal workflows. The number of calls handled or reports billed often stays the same. Output appears flat. Work changes underneath.
India is already seeing this pattern. Large IT services firms report 20-30 per cent reduction in coding and testing time in selected workflows using generative AI. Yet billed effort and overall revenue growth remain subdued. Faster delivery is absorbed into fixed-price contracts, margin protection, and quality control rather than expanded output.
From the perspective of national accounts, little seems to happen. The reason is fairly simple. AI mostly saves time, not labour. It speeds up drafting, searching, debugging, summarising, and decision-making. Productivity statistics, however, are built to capture changes in output per hour worked, not changes in how quickly uncertainty is resolved or coordination improves.
There is another blind spot. Productivity analysis focuses on averages. AI’s most consistent effect is on variation. Lower-performing workers tend to improve the most. Error rates fall. Outcomes become more predictable. For firms, reliability and consistency often matter as much as higher average output. Official statistics barely shows these takes.
Economic history offers a reminder. In the early decades of electrification, productivity growth remained weak because factories initially used electric motors as direct replacements for steam engines. Only when production was reorganised around electricity did productivity surge, decades later. In 1987, economist Robert Solow observed that computers were visible everywhere except in the productivity data. The computer-driven productivity boom arrived only in the late 1990s, once organisations adapted.
AI appears to be following a similar path. It is altering work faster than it is boosting measurable output. For India, this matters. The country’s growth strategy increasingly depends on digital public infrastructure, services exports, and the productivity of its large working-age population. AI is already present across these domains. But its early gains will mostly appear as faster workflows, fewer errors, and tighter coordination, not as immediate jumps in output per worker.
AI is unlikely to announce itself through a dramatic spike in GDP. It will arrive more softly, through small changes that accumulate over time.
The real risk is not that AI will fail to transform India’s economy. The risk is that it already is — while our measurement systems continue to look the other way.
The author is a theoretical physicist at the University of North Carolina at Chapel Hill, US. His forthcoming book is called Last Equation Before Silence
 "Artificial Intelligence, AI Technology, IT Sector")