War rooms, group chats, and video games: Inside Elon Musk's AI startup

• XAI merged with SpaceX and lost two cofounders — all in the same month.
• Current and former employees say Elon Musk has increased his focus on the startup over the past six months.
• Workers detailed Musk's 300-person group chat, war rooms, and their concerns about safety.

At xAI's Palo Alto headquarters, engineers cheered.

It was February 2, and they'd just received a memo from Elon Musk, informing them the AI startup would be acquired by his rocket company, SpaceX. In a Slack channel for the team that trains the company's chatbot Grok, memes flowed, and a few changed their profile pictures to depict themselves as astronauts.

The excitement was quickly overshadowed by the pressure building inside the startup racing to catch up with rivals like OpenAI and Anthropic, while preparing for an expected IPO.

Over the past six months, Musk has become deeply involved in day-to-day operations. He's run a massive group chat that's active at all hours, directed product changes, reassigned engineers, cut staff on key teams, and launched intensive "war rooms" to accelerate development. Several current and former employees said that the level of involvement has altered how the company functions. Leadership roles have narrowed, projects have shifted quickly, and teams have been pushed into what one described as a constant "fire drill."

Musk said that the changes are necessary as the company scales and the team's lean structure will be the secret to its success. Discussing xAI's restructure following the SpaceX acquisition, he posted on X:

"As a company grows, especially as quickly as xAI, the structure must evolve just like any living organism."

A lean Musk machine

Musk's memo about the SpaceX acquisition talked of building a "sentient sun" and data centers in space. Separately, the company reassured staff that little would change in their day-to-day work as the company geared up for an initial public offering that could value it at $1.5 trillion.

Within a week, however, the tone in the office had shifted. Two of the company's cofounders, Jimmy Ba and Tony Wu, announced they'd resigned as Musk restructured the company and narrowed their responsibilities. The leadership departures increase risk at "a sensitive stage" ahead of the company's IPO, said Dimitri Zabelin, a senior AI analyst at PitchBook.

One employee described co-founder Ba's departure as "incredibly disheartening," saying Ba, who studied under "Godfather of AI" Geoffrey Hinton, was one of the most respected researchers at the company.

In the following days, nearly a dozen employees took to Musk's social media site to say they had also left the company. Some had departed weeks earlier. A few left of their own volition. Others were affected by a restructuring following the SpaceX merger, in which Musk cut some members of the teams working on Grok Imagine, the chatbot's image and video generation feature, and the Macrohard team, which was developed last fall to help automate white-collar work, four people with knowledge of the issue told Business Insider.

Employees say the structural changes — coupled with an increasingly intense work culture since Musk wound down his involvement in President Donald Trump's Department of Government Efficiency and began focusing more on the AI startup — have significantly increased the pace and pressure inside the company.

Musk is a common sight at the office. Though he's heavily involved in the day-to-day, Musk does not use the company's Slack workspace, several people said. Instead, he communicates frequently through X, including in a direct message group of more than 300 engineers. Often, researchers will drop ongoing work to address the concerns that Musk flags.

In that chat, according to people familiar with it, Musk shares screenshots of his conversations with other tech executives and points out criticisms of Grok's performance he wants addressed. Late last year, Musk told staff that in conversations with his friends, he had been embarrassed by Grok Imagine's performance. As a result, some workers on the project were called to task, two people said.

Internally, Musk has expressed frustration with the pace of Grok's development. XAI released Grok 4.2 this week. Separately, releases of at least two other products were pushed back by several weeks, according to people familiar with the timelines.

In one instance last year, a model release was delayed for several days because Musk was dissatisfied with how the chatbot answered detailed questions about the video game "Baldur's Gate," according to people familiar with the matter. High-level engineers were pulled from other projects to improve the responses before launch, they said.

Across his companies, Musk has long been known for his intensity. He's said he slept on Tesla's factory floor during the "production hell" for the Tesla Model 3 in 2017. The company is now the world's most valuable automaker. Now, xAI is getting the same treatment. Several workers told Business Insider that 12- and sometimes 16-hour workdays are common. Two said managers told them they are expected to respond to Slack messages within 30 minutes, regardless of the time of day.

"Because the company is so small, everything is a fire drill," one former worker said.

At any given time, multiple "war rooms" operate out of conference rooms at the company's headquarters in Palo Alto, employees said. When researchers are "war-rooming," teams temporarily relocate to a shared space to work side by side on specific problems — sometimes for months. At the end of 2025, at least five war rooms were running simultaneously, according to three people. One, they said, was dedicated to teaching Grok how to play one of Musk's favorite video games, "League of Legends."

Shifting priorities around Ani and safety

Last year, Musk made it clear within the company that xAI would prioritize improving Ani, a hyper-sexualized, anime-inspired AI companion, people with knowledge of the issue said. He characterized Ani, a virtual companion who is capable of sexual role-play, as a way to set xAI apart from other AI companies.

Some employees told Business Insider they were unsettled by the company's focus on the product. Today, the character is displayed prominently at the company's headquarters. At a company holiday party, paid actors dressed as Ani and another character hosted a robot fight club-style event, according to videos shared on social media.

Several employees said they had joined the company hoping to work on systems that would push the boundaries of science and were discouraged to find significant resources devoted to Ani.

Grok's behavior on social media has also been a source of tension inside the company. Musk has characterized xAI and its chatbot as the "anti-woke" counterpart to ChatGPT. Public backlash against the chatbot, including for Grok's series of antisemitic rants and instances in which it digitally undressed people on social media without their consent, has put strain on employees, who have expressed concern about how their place of employment is viewed in the AI space.

Until last year, the company did not have a team of researchers specifically geared toward working on safety concerns associated with its large language model. It hired its first dedicated safety researcher in February of last year, around the same time that the Human Data team — which trains the chatbot — began reporting issues with reviewing large amounts of X user-generated requests for child sexual abuse material. One person with knowledge of the team said it was known within the company that a large portion of Grok usage was adult role-play, and the use case was discussed in several meetings with researchers.

The safety team grew to roughly half a dozen employees before three people left in December, shortly before users on X began reporting instances of the chatbot creating non-consensual sexual images of people, including some minors.

According to people familiar with the team, xAI's safety teams lacked the authority to formally block product launches and were focused primarily on adjusting model outputs after training rather than conducting broad pre-launch risk reviews.

Musk said on X last week that "everyone's job is safety" at xAI.

"It is not some fake department with no power to assuage the concerns of outsiders," he said.

XAI still has a handful of workers in safety roles and was hiring for additional safety staff in January, according to posts viewed by Business Insider.

Some former workers say the backlash and pace of work prompted their exit. Last summer, xAI shortened its vesting period from the industry standard of one year to six months, making it easier for workers to leave without forfeiting large portions of their equity.

Employee retention challenges are not unique to xAI; OpenAI and Anthropic have faced waves of departures in recent months. Across the industry, AI companies are competing fiercely for a limited pool of top researchers, some of whom have begun voicing concern about the narrowing space for exploratory work as companies increasingly orient themselves toward product over research in the competition to build ever larger models — a dynamic that poses a particular problem for xAI, which is younger and smaller than its main competitors.

Still, some analysts say the race is far from settled. Andrew Rocco, a stock strategist at Zacks Investment Research, compared the current AI landscape to the early days of the internet.

"I don't think Grok is that far behind, and it's still early in this race," Rocco told Business Insider.

Musk has publicly signaled confidence. He told staff last week during the all-hands, referencing xAI's moment of transition: "There's some people who are better suited for the early stages of a company and less suited for the later stages."

Do you work for xAI or have a tip? Contact this reporter via email at gkay@businessinsider.com or Signal at 248-894-6012. Use a personal email address, a nonwork device, and nonwork WiFi; here's our guide to sharing information securely.