The Company That Told People Not to Come Won the AI Talent War

Deep thinking, discussing AI and aspirations —— ByteDance Deep Thought Circle

When Meta launched its sky-high poaching offensive, two leading AI companies responded in two different ways. OpenAI followed market convention: counter offers for targeted employees, retention bonuses, eliminating the vesting cliff for new hires to let equity vest faster. Anthropic’s response was strange. Management told employees: you came here first and foremost for the mission, not to continually bid up your price through external competition. We won’t offer you ten times the salary of equally excellent colleagues sitting next to you just because Zuckerberg happened to target you. That’s not fair. If you want to leave, leave.

The results speak for themselves: OpenAI lost dozens of people, Anthropic lost only 2—and those two were former Meta employees with 6 and 11 years there anyway. That same year, Anthropic’s ARR grew from $9 billion to $45 billion, and its secondary market valuation touched $1 trillion, overtaking OpenAI.

The confidence to tell sky-high offers “leave if you want to leave” doesn’t come from money—it comes from an organizational design outsiders don’t quite understand. This design is worth unpacking, because it may be the most important organizational specimen in the AI industry over the past three years.

Recruiting Starts with One Thing: Screening Out the Wrong People

Anthropic’s recruiting logic runs counter to industry mainstream. The mainstream tries to bring in the strongest people possible; it tries to screen out the wrong people as early as possible.

Interviews include a dedicated culture round: one hour, 15 to 20 scenario questions. The most typical leaked question: If the company ultimately decides not to release a model because it cannot guarantee safety, would you accept your equity going to zero? This round screens for three things: Does mission truly come before profit? Are you kind, low-ego, able to admit ignorance and mistakes? Can you handle complexity and reason through how one decision affects other parts of the system?

This filter has costs. They’ve genuinely passed on engineers with more impressive backgrounds and stronger capabilities. Before Stripe’s former CTO Rahul Patil decided to join, he spoke at length with Anthropic’s then-CTO, who not only didn’t recruit him but spent two to three weeks repeatedly discussing why he shouldn’t join—unless you truly align on culture and mission, it’s not worth coming. The head of growth put this strategy plainly: We’re very good at filtering out people who come for money and status.

The Place Most Prone to Fiefdoms in the World Doesn’t Grow Them

Frontier AI labs are the most prone places for internal politics. Researchers are among the smartest and most ego-driven people, naturally wanting to propose different solutions and build their own empires, while compute and resources are extremely limited—conflict is almost inevitable. People who’ve jumped from elsewhere describe other model companies’ interiors as individual fiefdoms managing themselves and quietly competing.

Anthropic’s approach is to write values directly into governance structure. The company has 7 co-founders. Dario insisted all have equal equity. Everyone advised him this was a disaster—ambiguous control, misaligned incentives, the company would easily fracture from infighting. His thinking: the company revolves around the mission, not around any one founder, and equal equity is the most unforgeable proof of this principle. Seven people equals seven culture replication nodes; as the company scales rapidly, the original culture is less likely to be diluted.

Another move targets titles. Below executives, there’s no title differentiation—everyone is called MTS, Member of Technical Staff, deliberately erasing identity lines between researcher and engineer, senior and junior. The contrast is stark. OpenAI has a recognized internal hierarchy: researchers above research engineers, above software engineers. Product is often subordinated to research; research teams notify product teams of new results only at the last minute, then look for nails with their hammer.

Maintaining cultural density relies on unglamorous work. Dario himself says he spends a third to 40% of his time ensuring the culture is good—this is his highest-leverage work. Every two weeks, an all-hands meeting where he speaks for an hour on everything from company direction to industry changes, taking live questions. He records unfiltered in his channel what he’s worried about and how he sees issues people care about. Transparent to the point of being challengeable: someone disagrees with an all-hands statement, goes directly to his channel to publicly state disagreement, debate unfolds on the spot. This atmosphere produces: in complex, shifting circumstances, everyone knows how decisions are made, enabling relatively aligned distributed judgment.

Why This Matters Especially in AI

Culture isn’t just atmosphere—it directly translates to moat. The hardest moat in AI competition, at the deepest level, is data: the tasks themselves, execution trajectories, evals, validation systems. These are all unglamorous work—critically important when done well, yet unlike publishing a paper or launching a new product, they don’t become personal highlight moments. From feedback across companies’ researchers, one of OpenAI’s organizational challenges is difficulty getting hundreds of people with elite backgrounds and high aspirations to methodically work on data. They won on paradigm breakthroughs, so naturally they want to make their own next bet rather than clean up the last one’s mess.

Anthropic does the opposite. Co-founder Jared Kaplan personally leads teams through data daily, with extremely meticulous cleaning. A low-ego, mission-driven organization’s advantage is precisely amplified in this era: it can organize the smartest people around unglamorous work. Results correspond: OpenAI’s models remain strongest on competition-level programming challenges—that’s more a research problem. On everyday agentic tasks, Claude often leads—that’s more an engineering problem, testing data, systems, and execution details. Researcher Shunyu Yao puts it more directly: The era of individual heroism is over. The most important quality is reliability and attention to detail.

This also explains the opening poaching story. People who come for money and status are precisely those who can’t be organized to do unglamorous work. Filtering them out isn’t moral purity—it’s competitive strategy.

What You Can Copy, What You Can’t

It’s tempting to copy at this point. Distinguish three things first.

Mission can’t be copied. Anthropic has people who truly put mission ahead of company survival. An early employee said at all-hands: if the mission is ultimately achieved while the company itself fails, that’s still a good outcome. Most companies don’t have such a mission. Forcing it only yields team-building rhetoric; employees can tell real from fake better than anyone.

Structure can be copied. Reverse-screening scenario interviews, unified titles suppressing fiefdoms, founding team equality as proof of commitment, leadership high-frequency transparent writing, distributing tradeoffs to frontlines rather than escalating to executives for arbitration. These are mechanisms that work in any company, requiring no world-saving narrative first.

Origin can’t be copied, and is most easily overlooked. This culture is largely the counterforce to founder trauma. Dario saw fights over control tear teams apart at Baidu’s AI lab, experienced trust collapse among executives at OpenAI, and finally left with the core team to start anew. Having seen balancing acts exhaust trust, he emphasizes authentic transparency. Having seen political struggle dissolve organizations, he encourages bringing conflict forward openly. Values are half what you believe, half what you deeply loathe—the latter’s shaping force is often stronger, and loathing can’t be imitated.

Finally, a calibration to prevent this narrative from being pitched as a panacea. This is a three-year sample, not a permanent answer. Coding is already a known game; developers show signs of migrating from Claude Code toward Codex. Compute is becoming the new deciding factor, and OpenAI locked in far more early compute than Anthropic. Cult-like organizations turn slowly; they may not have advantage in the next paradigm leap. High retention may also mean people who should have left didn’t. Anthropic’s own people say culture is the most defensible thing. That’s true, but the defense is against talent wars, not technology wars.

One action item for founders: examine what your filter is screening for. If it’s screening for the most impressive resumes and highest aspirations, in an era competing on unglamorous work, the direction may be exactly reversed.