Inside the growing pain of Openai after launching Chatgpt: “Empire of Ai”

This is an extract of “Empire of AI: Dreams and Nightmares in Sam Altman’s Openai“By Karen Hao.

The book is based on interviews with around 260 people and a large correspondence and documents. All the e-mails, documents or Slacks Cités come from copies or screenshots of these documents and correspondence or are exactly as they appear in prosecution.

The author has contacted all key figures and companies described in this book to request interviews and comments. Openai and Sam Altman have chosen not to cooperate.

In November 2022, rumors began to spread within Openai that his anthropic rival was testing – and will soon be released – a new chatbot. If he did not go first, Openai risked losing his main position, which could offer a big morale for employees who had worked for a long time and difficult to keep this domination.

Anthropic had in fact not planned imminent versions. But for the managers of Openai, rumors were sufficient to trigger a decision: the company would not wait to prepare the GPT-4 in a chatbot; He would publish the GPT-3.5 compatible model of John Schulman with the brand new cat interface of the Superassic team in two weeks, just after Thanksgiving.

No one really imagined the societal phase discrepancy they were going to unleash. They expected the chatbot to be a flash in the pan. The day before the exit, they placed betting on the number of users could try the tool by the end of the weekend. Some people guessed a few thousand. Others have guessed tens of thousands. To be safe, the infrastructure team has provided sufficient server capacity for 100,000 users.

On Wednesday, November 30, most employees did not even realize that the launch had occurred. But the next day, the number of users began to increase.

The instant success of Chatgpt was beyond what anyone in Openai had dreamed of. This would leave the engineers and researchers of the company completely despised even years later. The GPT-3.5 had not been a great improvement in capacity compared to GPT-3, which had already been absent for two years. And GPT-3.5 was already available for developers.

The CEO of Openai, Sam Altman, later said that he thought that Chatgpt would be popular but by something like “an order of less size”. “It was shocking that people loved him,” recalls a former employee. “For all of us, they had demoted the thing we used internally and launched it.”

In five days, the co -founder of Openai, Greg Brockman, tweeted that Chatgpt had crossed a million users. In less than two months, he had reached 100 million, becoming what was the fastest growth consumption application in history. Chatgpt has catapulted Openai from a warm startup well known in the technology industry in a familiar name overnight.

At the same time, it is this very successful success that would make extraordinary pressure on the company. During a year, he would polarize his factions more and finish stress and tension within the organization at an explosive level.

Until then, the company had only 300 employees. With each team stretched dangerously, the managers begged Altman for more counting of the head. There was no shortage of candidates. After Chatgpt, the number of candidates who claim to join the rocket had multiplied quickly. But Altman worried about what would happen to the culture of the company and the alignment of the mission if the company has increased its staff too quickly. He firmly believed in maintaining a small staff and a high density of talent. “We are now in a position where it is tempting to let the organization develop extremely large,” he wrote in his vision note in 2020, in reference to Microsoft’s investment. “We should try very much to resist this – what has worked for us so far is small, concentrated, with high confidence, with low stool and intense. The general costs of too many people and too much bureaucracy can easily kill great ideas or cause sclerosis.”

Openai is one of the best places I have ever worked, but also probably one of the worst.

He now repeated this to the leaders of the end of 2022, emphasizing discussions on the number of heads the need to keep the business lean and the talent bar and not adding more than 100 hires. Other executives fell. To the pace that their teams were burned, many have seen the need for something closer to around 500 or even new people.

In several weeks, the management team finally compromised a number somewhere in the middle, between 250 and 300. The ceiling did not hold. In the summer, there were up to 30, or even 50, people joining Openai every week, including more recruiters to change the hiring even faster. In the fall, the company had blew well beyond its own self-imposed quota.

Sudden growth thrust has indeed changed corporate culture. A recruiter wrote a manifesto on how the pressure to hire so quickly forced his team to lower the quality bar for talents. “If you want to build Meta, you are doing a great job,” he said in a sharp blow on Altman, alluding to the very fears that the CEO had warned.

Fast expansion also led to an increase in shots. During his integration, a manager was invited to document quickly and to report any underperforming member of his team, to be released himself a little later. The endings have rarely been communicated to the rest of the company. People have regularly discovered that colleagues had only been dismissed by noting when a grayed -fashioned account to deactivate. They started calling it “getting disappear”.

For new hires, entirely adhered to the idea that they joined a rapid startup and earning money, the tumult looked like a particularly chaotic, sometimes brutal event, of standard business problems: mismanagement, confusing priorities, the cold cream of a capitalist company willing to treat its employees as available. “There was a huge lack of psychological security,” said a former employee who joined this time. Many people on board simply stood for life expensive up to their one -year brand to have access to the first share of their equity. An important advantage: they have always estimated that their colleagues were among the highest caliber in the technological industry, which, combined with apparently unlimited resources and unequaled world impact, could arouse a feeling of magic difficult to find in the rest of the industry when things were really aligned. “Openai is one of the best places I have ever worked, but also probably one of the worst,” said the former employee.

Sometimes there is not as much plan as there is just chaos.

For some employees who remembered the first disjointed days of Openai as a united lucrative non -profit organization focused on the mission, its dramatic transformation into a large faceless business was much more shocking and emotional. No more organization as they had known; In his place was something unrecognizable. “OPENAI is Burning Man”, explains Rob Mallery, a former recruiter, referring to the way in which the art festival desert has evolved to the point that he has lost contact with his original spirit. “I know that it meant much more for people who were there at first than for everyone now.”

During these first years, the team had created a Slack channel called #explainlikeimfive which allowed employees to submit anonymous questions on technical subjects. With the company that pushes 600 people, the channel has also turned into a place to broadcast anonymous grievances.

In mid-2023, an employee displayed that the company hired too many people who are not aligned on the mission or passionate about the construction of AG.

Another person replied: they knew that Optai was going down once they started hiring people who could look you in the eye.

As Openai was quickly professional and won more exposure and control, the top inconsistency became more and more substantial. The company was no longer just applied and research divisions. Now there were several public -oriented departments: in addition to the communications team, a legal team wrote legal opinions and treated an increasing number of prosecution. The political team extended to the continents. More and more, Openai had to communicate with a story and a voice to his constituents, and he needed to determine his positions to articulate them. But on numerous occasions, the lack of strategic clarity led to confused public messages.

At the end of 2023, the New York Times would continue Openai and Microsoft for copyright violation for training on millions of its articles. Openai’s response in early January, written by the legal team, delivered an unusually fiery return, accusing time to “intentionally manipulate our models” to generate evidence of his argument. This same week, the Openai political team presented a submission to the UK House of Lord Communications and Digital Select committee, claiming that it would be “impossible” for OpenAi to train its cutting -edge models without documents protected by copyright. Once the media has focused on the word impossible, Openai has hastily moved out of the language.

“There is so much confusion all the time,” said an employee in a public -oriented department. Although some of this reflect the typical startup pain, the profile and the range of Openai have exceeded the relatively early stage of the company, adds the employee. “I don’t know if there is a strategic priority in the following C. I honestly think that people make their own decisions. And then suddenly, it is starting to look like a strategic decision, but it is actually just an accident. Sometimes there is not as much plan as there is only chaos.”

Karen Hao is a award -winning journalist covering the impacts of artificial intelligence on society. She is the author of “Empire of Ai”.

Adapted from “Empire of AI: Dreams and Nightmares in Sam Altman’s Openai“By Karen Hao, published by Penguin Press, a Peublishing Group’s imprint, A Division of Penguin Random House, LLC. Copyright © 2025 by Karen Hao.