• In 2024, Kailash Nadh said AGI in 2–5 years was impossible.
  • Now in 2026, with Altman's predictions unmet, here's who actually got it right.

Zerodha's CTO Said AGI Timelines Were Hype. Two Years Later, Let's Score the Debate.

In October 2024, Kailash Nadh stood at a conference stage and said something unfashionable. The Zerodha CTO, speaking at Cypher 2024, pushed back firmly against the AGI optimism sweeping Silicon Valley: achieving Artificial General Intelligence within two to five years was "no way." He pointed to business motives behind the projections. He noted that AI had been "five years away for many, many years." He questioned who exactly was defining AGI and why Western lab CEOs had appointed themselves to do so.

Eighteen months later, it is worth asking: how is the scorecard looking?

What the Short-Timeline Camp Predicted — and What Happened

The optimists had specific claims. In January 2025, Sam Altman wrote in a blog post that OpenAI was "confident we know how to build AGI as we have traditionally understood it" and implied arrival that year. 2025 came and went without AGI. Musk predicted AGI by 2025, then moved the target to 2026 without acknowledgment. Dario Amodei of Anthropic forecast AI systems "broadly better than all humans at almost all things" by 2026 or 2027, and Anthropic's formal March 2025 submission to the White House Office of Science and Technology Policy stated it expected "powerful AI systems to emerge in late 2026 or early 2027."

The definitional problem Nadh identified has proven accurate. Altman has since called AGI "not a super useful term," which gives him rhetorical flexibility if specific benchmarks fail to resolve by his stated dates. When a model passes a capability threshold previously considered AGI-indicative, critics redefine AGI upward. The goalposts move. As one analysis of 2026 predictions put it: "No 1–3 year AGI prediction from any major figure has been verified. Several have been quietly moved forward without acknowledgment."

That said, progress has been real. 2025 saw the arrival of agents capable of real cognitive work, and writing computer code changed substantially. Benchmark performance on coding, mathematics, and reasoning tasks improved dramatically. The question Nadh raised — whether impressive-looking AI behaviour constitutes AGI — remains exactly the right question.

The Skeptics' Structural Argument

Nadh was not alone in his scepticism, and his peers have made the argument with increasing technical precision.

Yann LeCun, Chief AI Scientist at Meta, argues that large language models cannot achieve genuine understanding because they lack grounded world models — internal representations of how the physical world works, learned through interaction rather than text prediction. In late 2025, he stated: "There is no such thing as general intelligence. This concept makes absolutely no sense."

Andrej Karpathy, who co-founded OpenAI and ran Tesla Autopilot for five years, estimates his own timeline is "five to ten times more pessimistic" than Silicon Valley predictions, and has stated that useful autonomous agents are a decade out. Karpathy is emphatic that the current moment is real and consequential — he publicly said in late 2025 that he had essentially stopped writing code by hand. But he will not call it AGI. His Waymo example is instructive: perfect self-driving demos in 2014, widespread economically viable deployment still not achieved in 2025. That gap — between a compelling demo and reliable real-world deployment — is the "march of nines" problem that AGI timelines consistently underestimate.

Demis Hassabis at Google DeepMind has described what he calls "jagged intelligence" — the observation that current systems show gold-medal mathematics performance alongside failures that a twelve-year-old could handle. That jagged profile is what makes short-timeline claims difficult to defend: an AGI cannot be better than all humans at some things while failing elementary reasoning tests in adjacent domains.

Where do prediction markets sit? An analysis of over 9,800 AGI predictions shows that the aggregate of forecasters, prediction markets, and industry insiders points to a 25% probability of AGI by 2029 and a 50% probability by 2033. That is materially later than what Altman, Musk, and Amodei have suggested, and broadly more consistent with the skeptical camp Nadh represents.

The Open Source Argument That Mattered More

The AGI timeline debate attracts the most attention, but Nadh's comments on open source deserve equal weight — and have proven equally prescient.

Nadh said at Cypher 2024 that many open-source models outperform proprietary alternatives, and that without open source there would be "no Zerodha, no Indian startup ecosystem." Three months later, DeepSeek released R1 — a Chinese open-source model that surpassed OpenAI's o1 on multiple reasoning benchmarks and briefly topped the Apple App Store. Meta's Llama series has continued closing the gap with proprietary models. Hugging Face launched Open-R1 to replicate and open-source DeepSeek's training process entirely.

For Indian founders specifically, this trajectory is not abstract. Proprietary API costs from OpenAI or Anthropic — denominated in dollars, billed at Western consumption volumes — are prohibitive for startups building for Indian market price points. Open-source models that can be fine-tuned and run locally change the build economics entirely. Zerodha's own AI deployments, which Nadh said replaced work previously done by roughly 100 people, are built substantially on open-source infrastructure.

The Zerodha commitment Nadh announced — an open-source fund to support global projects — reflects an understanding that the infrastructure Indian startups depend on is a public good that requires active investment, not passive consumption.

Bottom Line

The most reliable forecasters in the AGI debate think in decades, define their terms precisely, and correct themselves when the evidence demands it. By that standard, Nadh's Cypher 2024 remarks have aged well. The two-to-five-year AGI timeline crowd has not delivered AGI; they have delivered better models, more capable agents, and a definitional vocabulary flexible enough to claim progress regardless of outcome.

The open-source ecosystem Nadh championed has, in the intervening period, produced some of the most significant AI breakthroughs — often from outside the US labs making the loudest timeline predictions. For Indian founders building AI products today, his practical conclusion remains the correct one: build on open-source, ship useful things, and ignore the hype cycle.

Edited by Nabarun