AI isn’t just evolving, it’s accelerating at a pace that threatens to overrun India’s service-based economy. New data shared by Zerodha co-founder Nithin Kamath paints a stark picture: artificial intelligence is now capable of performing longer, more complex tasks than ever before, and it’s improving so fast that many human jobs may not stand a chance.
One chart reveals that the length of tasks AI can execute with 50% success is doubling every seven months. Just a few years ago, models like GPT-2 could handle seconds of basic queries. Today, GPT-4o can hold its own for nearly an hour.
And the most advanced models—like Sonnet 3.7—are already handling multi-hour jobs such as training image classifiers. These are no longer experiments. They’re replacements.
A second chart is more unsettling. It maps AI’s success across task difficulty and duration. On clean, short tasks, top models are now achieving over 90% accuracy. Even messier, ambiguous problems—once considered beyond AI’s reach—are seeing sharp gains. Only the hardest category, long and messy tasks, remains below 30% success.
But even that barrier is eroding.
Faced with this data, Kamath posed a sobering question on X: “How long do we have left?”
He’s not alone in raising the alarm. Atomberg founder Arindam Paul estimates that 40–50% of white-collar jobs in India—especially in IT and BPO—could vanish.
A recent IIM Ahmedabad study shows 68% of white-collar professionals believe AI will automate their jobs, at least partially, within five years. Jefferies forecasts that half of all entry-level white-collar roles—from marketing to sales to software—may be wiped out in just 1–5 years.
This isn’t theoretical. It’s happening. And India, with its heavy reliance on services and limited manufacturing capacity, is uniquely vulnerable.