沟通就是工作：Meta CTO 从早期 Facebook 学到的领导力

Bosworth says he joined as one of roughly six engineers who started the same day, after nine others, so he is somewhere between the 10th and 16th engineer. The period was full of camaraderie, but forged in intense conditions: almost everyone lived within a mile of the office and ate together because they were working. There was no expert to ask for help, and if servers ran out of capacity you went to Fry's Electronics, bought components and built one yourself.

He reminds listeners that companies today get payroll, finance, IT and HR handled by professionals, which was not the case in the early 2000s. He worked 120 hours a week, gained weight and had no hobbies. For about two years, after building an anti-scraping defense that had no 24/7 ops support, he never slept more than four hours at a time so he could check the attack report and fight back.

Nobody asked him to take on that problem; he simply saw it and chose that solution. He loves startup culture but wants people to know that the romantic version comes mostly from the winners. He also cautions founders to treat his advice with a grain of salt, since he joined almost two years after Mark Zuckerberg started the company and never handled fundraising or the board.

The advice Bosworth says he gives more often than any other is to leverage your leaders more directly. People want to do it themselves and fear that asking for help means admitting defeat, but he says the job is to get the thing done well, and the tools for that often sit with your manager, partner or mentor. He describes managers waiting months to find that someone misunderstood the brief, or that a task took a year because of obstacles a quick email could have cleared.

He quotes Patrick Stewart's point that no one wants you to succeed more than the person you are auditioning for, and applies it to managers who are rooting for you. He also says managers have a job to refuse to rule on things when someone has the right approach. The goal is light touches, not outsourcing the work.

For timid people, he suggests a frame that makes engagement easy: a five-to-ten sentence email with progress and blockers, and "no response required" if all is well. When truly blocked, make it cheap for the manager to help by drafting the note to another team or by framing yes/no questions. At his scale of 10,000 to 15,000 employees, he says his job is spinning plates, and a heartbeat ping tells him which one is losing momentum.

Facebook managers used a format called HPM, for highlights, people and me, sent upward every week. Bosworth adds that every manager is different, so ask early how they like to be informed. He also notes that a drafted message to another leader carries tone and history, not just content.

Bosworth says leaks feel like a violation of team trust, comparing them to a player calling the play to the other team. He defends Meta's transparency on the grounds that talented people are the company's most valuable resource, and withholding information blocks their time and talent and pushes them toward leaving. Great work, he says, has come less from top-down mandates than from people with the information to act on them.

He calls the top-down versus bottoms-up debate partly a myth. The ideas pursued are first of all Zuckerberg's, though Zuckerberg is open to others and leaves room for teams to bring back several versions of an idea. Anyone can ask him, Bosworth, Chris Cox or Javier any question, but the bottom-up energy operates inside a framework with strong opinions at the top.

Openness has a price: employees must build a system to triage incoming information, choose which groups to follow and which to take as push notifications. Bosworth even coaches senior hires from outside on finding signal amid noise. He adds that the most important thing is clear guidelines about where people have room to decide and where they do not.

Bosworth recalls that on News Feed Zuckerberg whiteboarded every pixel of the front end while telling him only to make the backend rank, and that Zuckerberg later got deeply involved in modernizing ranking systems. He himself forced the issue on hand tracking and mixed reality in the headset, setting tough performance benchmarks even though teams told him it was impossible and few supported it; it did great.

Asked how deep leaders should go, he says it depends on the weeds: some matter hugely and some matter without being the hinge of success or failure. Many founders he talks to regret delegating too much, or delegating something that turned out to be critical. The hard part is judging what matters most.

Inside Meta they call Zuckerberg's focus the eye of Sauron: once he decides something is the most important thing, no detail is too small, and a single review can range from strategy to one wrong pixel. Bosworth says he tries to lead similarly. Having run ads from 2012 to 2017 while Zuckerberg focused on mobile, he says you suffer both in the intensity of the gaze and in the shadow of its absence.

On News Feed, the first algorithmically ranked feed on a social network, Bosworth says the lesson is to have conviction in what you are building. You choose your customers as much as they choose you, and early users can pull a startup in a direction orthogonal to the larger market. He adds that News Feed was an easier case than people think: everyone was outraged yet immediately doubled their usage, a gap between stated and revealed preferences.

He rejects the lesson that you should not listen to customers. The team did screw up details, he says, such as taking wall posts and broadcasting them at volume, like a private comment suddenly heard by an entire party. The art is telling when a small thing is wrong from when the thing itself is wrong, which requires a clear vision of what you expected and a diagnosis of the gap.

He and Chris Cox built the engineering side, and he notes it was probably the first consumer AI on a website of that kind. Lenny jokes that Boz, News Feed and ads created a trillion dollars of value, to which Bosworth replies that it is not nothing. He also says the experience showed that passion makes people better and an order of magnitude more productive.

Bosworth says he has learned he is someone who can get passionate about many things. He resisted working on ads, then on hardware, because he thought of himself as an AI and infrastructure software person, and then loved both. His lesson is to give yourself room to find out whether you can get excited about something, while recognizing that if after six months the passion never comes, it is time to move.

His career moved roughly every six months for years, through integrity work, News Feed, site speed, bootcamp, messaging and groups, and he compares it to the Karate Kid painting fences. Peers on a single track were promoted faster at first, but when he took the ads job for five years his career went vertical. He advises optimizing for learning, like compound interest that only looks impressive after ten years.

On where to work, he names two good places: carrying a lot of water on something important the company is ignoring, or being on the most important thing and taking a piece you can crush. He recalls asking Zuckerberg to keep putting him on the biggest fire every six months after the first mobile ad product in 2012, and Zuckerberg replying that this is not a real job. He then stayed in ads for four and a half years.

Asked about giving Zuckerberg critical feedback, Bosworth calls him the strongest-willed person he has met, but one who is voracious for information and points of view. He will usually tell you why you are wrong, yet over the next week or two you see things shift. Bosworth jokes that information accumulates and Zuckerberg recompiles the world overnight.

This applies to product and to his own presence in meetings. Zuckerberg will ask others in following meetings what they think of the feedback, without attribution, to triangulate points of view. Bosworth finds him satisfying to give feedback to, as long as you take the long view and accept that he usually has reasons for what he did.

Bosworth explains his Veritas tattoo, Latin for truth and also Harvard's motto, and says getting it changed how he thought about bodies, time and mortality. Truth matters to him, but he says he once saw honesty as a get-out-of-jail-free card and that his biggest professional regrets are times he was not kind. Being kind, he says, is not being nice or telling half-truths but delivering feedback productively so it helps.

On his essay *Communication Is the Job*, he says that having an impact on the world happens only through artifacts or verbalizations that affect other humans. Saying you wrote something up a year ago means little if you did not break through, and the fault is not the audience's. When someone returns with the wrong work, he "takes the L" because he did not communicate expectations clearly, a link to the idea of extreme ownership.

He argues that org charts are communication devices, that silence is communication, and that you cannot not communicate. The core skill is understanding the mental model of your audience, where they are and where you want them, which he says takes empathy rather than a fabricated persona. He even explains his flickering camera on the call, so the guest is not thought incompetent.

His practical tactics include repetition in several modes, such as giving an all-hands and then writing a post. Another is addressing people's fears up front, since they will not trust your conclusions if they think you do not understand the problem. He adds that coaching at Meta, with mentors like Sheryl Sandberg and Zuckerberg, and reading parenting literature, made him a better partner and manager.

Bosworth says that when the Vision Pro came out, he and Zuckerberg felt relief because nothing inside was a fundamental breakthrough apart from Apple silicon. After trying it together, they concluded they have the best headset overall, though the Vision Pro shines for stationary movie watching, with sharp resolution and nice UI and eye tracking. He prefers Quest's hand tracking, wider field of view, brighter displays and less motion blur in mixed reality.

He acknowledges he is biased and urges people to try Quest 3 before judging, since most haven't. He asks whether you would rather have seven Quest 3s for you and six friends or one Vision Pro, and notes Quest can stream a desktop with several monitors. He is also puzzled that Apple's much higher price drew less outrage than Meta's $1,299 headset, though he says he is a fan of Apple.

Lenny shows off his Ray-Ban Meta glasses, and Bosworth says glass is the way to avoid a phone screen coming between you and the real thing. He has used the multimodal assistant since December, for example asking it to read a sign in a ski village and getting directions. He also hints at coming glasses with full AR capabilities, teasing them as almost confirmed.

On whether Zuckerberg's video comparing the headsets was a strategic move, Bosworth says no: it was simply what he thought. He compares it to other companies' rally-the-troops gestures and says the team understands Apple's design choices because they had the chance to make the same ones and chose differently. People expect four-dimensional chess, he says, but sometimes it is just saying what you believe out loud.

Asked about the swing from an $80-ish stock price to $487, Bosworth notes Meta had the largest single-day drop in history followed 18 months later by the largest gain. He cites Lou Holtz that you are never as good as they say when winning nor as bad as they say when losing. He tells his team they know more about their company than critics, analysts or podcasters do, but they should still read criticism carefully, watching for confirmation bias.

He explains Gell-Mann Amnesia, where you spot errors in a newspaper article on your own field and then trust the next article on a topic you do not know. The lesson is to read everything while integrating it as another point of view. He says Meta did not communicate well about its two ten-year investment areas, AI and Reality Labs, and notes that FAIR is the second most cited AI research lab behind Google.

When the tide went out, he says, investors wanted to cut future bets, but he told employees that a company that kills all future growth in tough times is committing itself to dying a little later. On the layoffs, he says COVID-era hiring assumed a lasting shift in e-commerce and remote work that did not pull forward, so the economics stopped making sense, and he calls the result awful. Many workers who joined after 2009 had never seen a downturn, he adds.

Asked about a perception that only intense people remain, he declines to generalize about those who left. He suspects people who were coasting were probably already working hard, and that what changed was a selection effect rather than how hard individuals work.

In his Failure Corner, Bosworth says he has built unused products and unscalable architectures but does not regret most failures, because he learned from them. The failures he takes seriously are personal ones in which he affected someone in a way he is not proud of. His example is a debate with a close friend and brilliant engineer over whether a new RPC system should use binary or plain-text encoding.

He wanted binary for efficiency because he stored the calls in logs to replay for News Feed ranking, and his friend wanted readable descriptors. In a room with about half the company's engineers, Bosworth was red and sweating and shouting, even though the simplest fix was to build two encoders. His friend turned out to be right, since memory bandwidth stopped mattering within a year.

The real problem was identity threat: his identity was being right, and when a core part of how you see yourself is questioned, you defend it with all your fiber. He was 22, looked unhinged, and recalls that people remember how you made them feel. The incident was one step in recontextualizing what it means to be open-minded and curious.

Coaching helped him recognize the feeling in his own body and repair afterward. The greatest lesson came from watching Ami Vora, who answered someone's crazy-sounding view with "Fascinating, tell me more" and meant it. Her genuine curiosity tore down walls, and she would change her mind when she was wrong. Bosworth says he now tries to respond to challenge with curiosity, and that this has changed his life.

In the lightning round, Bosworth recommends *The Dream Machine*, a history of modern computing around J.C.R. Licklider, which he links to today's human-in-the-loop AI. He also recommends Dr. Becky's *Good Inside*, calling it a parenting book with lessons for managing your own emotions. He picks *The Mandalorian* as a favorite show, enjoyed with his kids.

His favorite interview question asks candidates what people who have worked with them would say are their greatest strengths and weaknesses. He values it for how they respond to criticism, and he cares most about strengths, since people tend to be accurate about those and less so about weaknesses. He also names the Mercedes-Benz AMG EQS as the best augmented reality product you can buy, because of its head-up display.

His family motto is "trust yourself", inspired by a neon art piece in the hallway and a family crest. He ties it to his conviction at startups, and the host adds that trusting yourself includes taking risks. On photography, he says art should be made for yourself, citing Rick Rubin and Ansel Adams, and picks an unsharp photo of his son jumping in a puddle as a favorite.