OpenAI marks two years of ChatGPT with sights on AGI and agents

Sam Altman's 2025 New Year reflection reads like part founder therapy session, part corporate war memoir, and part warning shot to the entire AI ecosystem.

Writing on his personal blog, the OpenAI CEO shared a remarkably candid look back at the chaotic two years since ChatGPT turned the obscure research lab into a global juggernaut. Along the way, he detailed his dramatic November 2023 firing, gave rare inside details on ChatGPT's origins, and laid out where OpenAI is marching next: autonomous AI workers in 2025, followed by superintelligence.

A Friday in Vegas and the fog of war

A little over a year ago, Altman was sitting in a Las Vegas hotel room on a Friday morning. Out of the blue, the board fired him over a video call, hanging up and immediately publishing the news online.

Altman described the ordeal as a dream gone wrong and a bizarre fog of war where nobody could get straight answers. He openly termed the crisis a massive governance failure by well-meaning people, including himself. Looking back, he admitted he wished he had handled things differently, noting that good governance requires deep trust and a board with diverse perspectives.

He also gave credit where it was due. Altman singled out veteran investor Ron Conway and Airbnb CEO Brian Chesky as the two people who kept OpenAI from completely unravelling. In a footnote, he wrote that he was reasonably confident the company would have fallen apart without their round-the-clock intervention, strategic advice, and backstage maneuvering.

From 'Chat With GPT-3.5' to AGI

The post also shed light on how accidental OpenAI's breakout moment really was. Back in 2022, the team was working on an internal tool temporarily dubbed Chat With GPT-3.5—with Altman deadpanning that OpenAI has always been much better at research than at naming things.

After noticing developers loved chatting with models in their API playground, they put together a demo to show off future possibilities and gather safety feedback. They mercifully renamed it ChatGPT, dropped it on November 30, 2022, and inadvertently unleashed the fastest adoption curve in tech history. Over the past year alone, weekly active users jumped from roughly 100 million to more than 300 million.

Scaling at that pace, however, felt like cramming a decade into two years. Altman acknowledged rampant internal turnover, former colleagues leaving to launch rival firms, and endless friction. But as models transition into complex reasoning, Altman claims OpenAI is now confident it knows how to build AGI as traditionally defined. His immediate milestone? In 2025, the first AI agents will join the workforce and materially alter corporate output. Beyond that, the company is turning its sights directly toward superintelligence.

Why it matters

Altman is officially closing the book on the foundational era of LLMs and setting the table for 2025's agent rollout.

For workers and businesses, his prediction that agents will join the workforce this year is not hyperbole—it is OpenAI's product roadmap. If 2023 was about chatbots and 2024 was about chain-of-thought reasoning, 2025 will be about software taking concrete actions on your behalf. Companies will need to decide whether to integrate these autonomous agents into production workflows or risk falling behind competitors who do.

At the industry level, Altman's pivot to superintelligence is a clever way to keep the funding narrative alive. By asserting that OpenAI already knows how to reach traditional AGI, he is justifying the astronomical capital and compute requirements needed for the next phase. Yet, as his frank account of the Vegas firing shows, rapid scaling combined with ideological drama makes OpenAI's governance uniquely fragile. The race to superintelligence won't just test compute clusters—it will test whether OpenAI can hold itself together under escalating pressure.