OpenAI stands at a delicate crossroads: on one side, a sprint toward what could be a trillion-dollar valuation in a “super IPO”; on the other, frequent executive departures and internal restructuring that have bred employee frustration. Inside the world’s most sought-after AI company, an unassuming email address—friction@openai.com—has become an unlikely tool for management to combat bureaucracy and keep the machine running at full speed.
According to Fortune, a single email sent to this address at OpenAI can directly shift a team’s priorities. The word “friction” refers to resistance or drag. Employees can flag technical system failures, broken internal processes, or even insufficient office parking by sending a direct email. The leadership team triages these issues; once deemed worth addressing, they enter a fast-track resolution channel, and significant problems may even reach CEO Sam Altman or President Greg Brockman.
OpenAI barely uses email—internal communication runs almost entirely on Slack—but friction@ is the exception. The issues handled through this inbox range widely. To ease parking congestion, the company once launched a pilot program prioritizing spots for employees with longer commutes. Another employee raised the need to improve the API credit allocation process to ensure stable, large-scale distribution.
A former OpenAI employee believes the mechanism gives management direct access to frontline feedback, allowing problems to be addressed without climbing layers of reporting and before they snowball into bigger issues. “Anyone can surface the bullshit problems of a big company directly to Sam, and they actually take action,” the former employee said. “If you want to ruthlessly eliminate bureaucracy, you have to be willing to be ruthless.”
Efficiency Anxiety Amid Hypergrowth
As OpenAI expands rapidly, this channel has become increasingly vital. The company expects headcount to surpass 8,000 by the end of this year, with offices across the United States and beyond. The larger the organization, the more friction arises from approvals, collaboration, and resource allocation.
The friction@ inbox has existed at OpenAI for some time but initially saw little use. In the fall of 2025, Fidji Simo joined OpenAI as CEO of the applications business. After arriving, she spent three months on a “listening tour,” speaking with employees at all levels to understand internal pain points. One concern came up repeatedly: as the company grows, can it maintain its original speed of execution?
Simo believed bureaucracy rarely appears overnight. An unnecessary meeting, an extra approval layer, a minor frustration—each might make work just 1% or 2% harder. These issues are easy to overlook, but compounded over time, they can significantly slow an entire organization. She subsequently built a clearer processing workflow for the friction@ system, assigning Irina Kofman, VP of Strategic Planning and Operations, to monitor the inbox and track issue resolution. The company also began sharing monthly progress updates with all employees via Slack.
In early July, Simo stepped away from OpenAI for health reasons to focus on her healthcare startup, but the mechanism she refined remains in place.
Executive Exodus and IPO Uncertainty
The existence of the friction mechanism reflects the organizational tension OpenAI currently faces. According to the Financial Times, as Altman intensifies preparations for what could be a blockbuster IPO, the company is grappling with mounting internal turmoil, having undergone nearly six reorganizations this year.
This week, OpenAI Chief Revenue Officer Denise Dresser announced her departure. She had joined just last December to lead enterprise customer expansion. Before her, former CFO and COO Brad Lightcap and AI ethics lead Chloé Bakalar had already left—Bakalar after less than a year in the role.
Simo, after a period of medical leave, stepped down from her full-time position to an advisory role. But two people close to OpenAI say she remains active behind the scenes, communicating with employees almost daily. “She’s still pulling strings,” said one person familiar with the internal power dynamics, who also described an environment of intense “political gamesmanship” where multiple executives are vying for Altman’s attention.
Multiple sources said that late last month, OpenAI also disbanded its “preparedness” team, which was responsible for assessing whether models could pose severe or catastrophic risks and researching mitigation measures. Senior staff were subsequently reassigned within existing teams to handle preparedness across domains such as biological and cyber threats.
OpenAI has framed the recent executive departures as part of pre-IPO organizational streamlining. Altman had previously asked employees to cut back on side projects and focus on ChatGPT and competing with Anthropic for enterprise customers.
The Revenue Race and Safety Concerns
Compounding the situation, OpenAI has been overtaken this year by archrival Anthropic. According to people familiar with the matter, the ChatGPT maker’s annualized revenue has grown from $24 billion at the end of last year to roughly $40 billion this month. But Anthropic surged from $9 billion at the end of 2025 to $47 billion by May—a more than fourfold increase—and has shown no signs of slowing since.
CompanyAnnualized Revenue (End of 2025)Latest Annualized RevenueGrowthOpenAI$24 billionapproximately $40 billion~67%Anthropic$9 billion$47 billion (May)>420%
Note: Both companies extrapolate full-year revenue from recent performance, but they account for partner revenue differently, making direct comparison difficult.
According to one source, most of OpenAI’s growth has come from GPT-5.6, the new model released five weeks ago.
The recent string of executive departures has also drawn attention to an unsettling pattern: some executives were reassigned to new roles or placed in positions with vaguely defined responsibilities before leaving the company. For example, Lightcap was moved to a role focused on “special projects” in April, while other senior figures including Kevin Weil and Josh Achiam were also reassigned before their departures.
Jan Leike, who co-led the now-disbanded “superalignment” team, resigned in 2024, citing his belief that safety had “taken a backseat to shiny products” at OpenAI.
In recent weeks, internal tensions around OpenAI’s AI safety stance have intensified. OpenAI previously disclosed that during internal testing of cyberattack capabilities, one of its models had breached another company’s systems. The departures of Bakalar, Achiam, and Johannes Heidecke—all of whom had safety responsibilities—have further heightened unease within the company.
“It’s a bit scary. There’s an urgent need to get this right,” said one person close to OpenAI.
Employee Cash-Outs and Investor Divisions
OpenAI recently completed a tender offer of nearly $7 billion to buy back shares from employees. According to current and former employees, staff have been cashing out at OpenAI’s valuation of $852 billion in recent months, with some individuals realizing more than $10 million.
Two people said employees feel “exhausted” and “frustrated” by the constant reorganizations and view the tender offer as an exit path.
Some investors have expressed concern about OpenAI’s high turnover and strategic drift. The company is targeting one of the largest IPOs in history. Others argue the changes are about streamlining the management structure.
“Cleaning up the org structure before an IPO—I don’t think that’s unusual. You need to figure out who your core team really is,” said one investor holding a significant stake in OpenAI. “Some people left for health reasons. Some of it is just turnover.”
According to people familiar with the matter, OpenAI’s IPO was originally planned for this year but now appears likely to slip to next year.
In the fierce race to build and ship the latest AI models, the existence of the friction mechanism underscores a reality: as long as any drag on speed can be eliminated, other costs seem temporarily acceptable. In December 2025, OpenAI reportedly sounded an internal “red alert,” directing employees to concentrate resources on strategic products to counter competitors like Google and Anthropic. In this environment, any internal bottleneck can affect product timelines.
The mechanism also evokes Amazon’s “question mark emails.” In the past, when customers complained to Jeff Bezos and an issue warranted investigation, he would forward the email to the relevant team with nothing but a single question mark in the body. The receiving team had to immediately investigate and resolve the issue. Amazon primarily handled external customer complaints; OpenAI focuses on internal bottlenecks. Both employ a similar management approach: direct intervention from top leadership to compress the communication chain.
A former OpenAI team lead said they received “dozens” of friction emails during their tenure. For teams that are chronically understaffed and overloaded, each escalation can disrupt existing plans. OpenAI’s work culture is famously intense. Employees work long hours, often past dinner and late into the night. In this environment, a single friction email can suddenly trigger an unwelcome “fire drill.”
“If Sam and Greg decide something is worth addressing, if the issue makes sense, a friction task kicks off immediately. The party responsible for the bottleneck basically gets told to shut up and comply. This is not a joke—you have to drop everything you’re doing.”