软件不再是护城河：Snap CEO 谈分发、扁平设计团队与眼镜

Asked why so few consumer social products have lasted, Spiegel points to the two recent exceptions Lenny names: TikTok and Threads. In his view both succeeded because they solved distribution. Snapchat launched when the App Store was young and people were eager to download new apps, which is no longer true.

TikTok, he says, used money, spending billions subsidizing both sides of its video marketplace by acquiring viewers and paying creators. Threads leaned on Meta's existing distribution. Spiegel thinks founders obsess over whether they have product-market fit and do not spend nearly enough time on how people will find the product.

Snapchat's own answer was a different view of network effects. Instead of assuming the biggest network always wins, it bet that most of the value lies in connecting you to your best friend, partner or spouse. Growth came from close friends, not from having the most friends. Lenny adds that AI may make distribution the ultimate moat, since it can help with building but not with attention, and Spiegel says new platforms such as glasses will open fresh opportunities.

Spiegel jokes that Instagram Plus is the latest copy, arriving right after Snapchat Plus passed 25 million subscribers and a billion-dollar revenue run rate. Being copied, he says, beats making things nobody wants to copy, but it forces strategic thinking. The core lesson, learned about 15 years ago, is that software is not a moat, something everyone is discovering again with AI.

Snap's response was to build things that are hard to replicate. One was ecosystems: the relationships between creators and Snapchatters, and an augmented reality platform on which developers have built millions of lenses. The other was hardware, where a fully vertically integrated AR stack is very hard to clone. Network effects remain an important piece, he adds, but patents and network size did not protect the ideas from being iterated on.

Spiegel says his motivation for hardware comes from loving computers while noticing that they pull people out of social life, from the childhood computer lab to teenagers looking at their phones together. People spend seven or eight hours a day on screens, and AR on a phone is like seeing the world through a keyhole. He thinks it is time for a new kind of computer that keeps people grounded in the real world and frees their hands.

He sketches the roadmap: the first Spectacles in 2014 simply moved the camera off the phone, then came a second camera for depth, then a display to overlay digital objects, and in 2024 an operating system for developers to build full software. Specs, the consumer product, comes out this year. Lenny, who previewed them, praises the multiplayer features.

On the risk of being hooked to a screen all the time, Spiegel says social norms will matter but that Specs anchor content in the world rather than putting a notification screen in the corner of your eye. He does not think phone notifications on your face are a compelling product, and says the real value will be new ways to hang out with friends.

Spiegel recommends *Loonshots* by Safi Bahcall as the best summary of how Snap thinks about innovation. Large organizations need hierarchy and operational rigor to deliver at scale, but hierarchy makes people promotion-focused and risk-averse. Innovation needs very flat structures, and successful companies hold both, with leaders responsible for a healthy relationship between them.

At Snap this means running a large public company serving almost a billion people, which must be reliable, alongside a design team that oscillates between nine and 12 people with no fancy titles. Tension arises when each side mocks the other, so he works to build mutual respect. He ties it to his own early partnership with Bobby Murphy, a computer science and stats person who loves design, and himself, a design person who understood computer science.

The most important thing for the design team, he says, is velocity. He meets designers for a couple of hours every week to look at new work, hundreds of ideas a week by his estimate. This mixes the empathy-driven process he learned in Stanford's product design program with the grinding output and brutal critiques of art school. Their favorite saying is that to have a good idea you must have lots of ideas.

Lenny cites a guest who says consumer founders should not talk to customers. Spiegel disagrees: share ideas early and often, and listen. He does not mean surveys, but long conversations of an hour or two about how people use technology and how it fits their lives. You need not take their advice, but customers are an endless source of inspiration.

His example is Stories. Users asked for a send-all button, but they also said social media felt permanent, judgmental and pressured by likes and comments, and found reverse-chronological feeds strange. Stories answered those insights without the literal request: easy sharing with all friends without spamming, no public metrics, disappearance after 24 hours, and chronological order. It evolved from early ideas about status updates, along with gestures like tapping forward and back and full-screen vertical content.

Another favorite is screenshot detection. Skeptics said disappearing photos were pointless since you could take a screenshot. Snap found that holding a finger on a snap while screenshotting caused a touch event the app could detect, and notified the sender. Spiegel says users did not mind being saved, they just wanted to know, and the feature helped early traction.

On why Snap waited until about 200 employees to hire its first product manager, Spiegel says his view was driven by designers. In many tech organizations, designers produce visuals in response to PM direction. He did not want that, so his early line to designers was: if you need PM support, do it yourself. At today's scale, with legal and trust and safety needs and data science, he says PMs play an important coordination role.

Asked about the three-way standoff between PMs, designers and engineers over AI, Spiegel calls it dysfunctional but welcomes people rethinking how they work. He says designers feel somewhat vindicated, and many at Snap are now shipping code. Tools that remove friction between idea and impact at nearly a billion users level the playing field for creative people.

Design at Snap is deliberately a bottleneck: things need design approval to ship, which can annoy people and slow shipping. He argues that this produces a cohesive customer experience, unlike apps built by teams each owning a page. He also stays deep in the details himself because he loves it, and says every leader should stay close to customers, product and team.

Snap hires designers on portfolio alone, and most join straight out of school. Spiegel looks for wide range, which he says separates design from art: a distinctive personal style is art, while design means empathizing with an audience and building things that look very different for different needs. He interviews every designer and asks them to pick any work they feel strongly about and tell the story of why they made it and what they learned.

To develop young talent, new designers present work on their first day, and the pace of making lets the ego melt away since most of a thousand ideas will not be great. Designers also rotate across products so they do not get stuck on one area for years. Spiegel says there is no gate to the weekly meeting: any idea can go on the list, because over-reliance on critique filters out great ideas.

Shipping code is not a requirement for designers, Spiegel says, but they are curious and used to figuring things out. The real question when anyone can submit a PR is how not to break things at nearly a billion users. Snap has built AI tools and guardrails, including automated code review that has caught close to 10,000 bugs.

Internally, employees can shake the internal app to report a problem and agents now debug what went wrong and suggest a fix, with implementing the fix expected soon. To bring order to the excitement, Snap listed the jobs to be done for its community, such as downloading the app and adding close friends, and for advertisers, such as configuring a campaign. That shows where to use agents, build cross-functional teams and track business outcomes, while still letting a thousand flowers bloom.

Reflecting on 15 years, Spiegel says the biggest surprise is how much the job changes: from designing the product, answering support emails and sorting legal filings to developing people, culture and strategy, and steering through the pandemic, an ad platform rebuild and now AI. He says he had to learn to communicate, recalling advice that a president is the "explainer in chief". A board member pushed him to do all-hands and Q&As, and he now enjoys them.

He calls the year a crucible moment. Snap is nearing a billion monthly users and the Fortune 500 and about to launch Specs after 12 years of investment, but it is still not net income profitable. This year it must prove it can be a strong, profitable business, with growth in engagement from things like topic chats, Spotlight and games played by 200 million people a month, because winning in Specs needs a solid foundation.

He describes Snap as the middle child of the market: much bigger than Pinterest or Reddit and far smaller than Meta and Google, with the challenge of defining itself. Getting Specs into consumers' hands, he says, is how people will finally understand Snap's next chapter.

In the AI corner, Spiegel says he built an agent in Glean that combs through company dashboards, documents and data, plus weekly three-things updates from leaders, and flags what he should focus on. He says it sometimes catches something he missed and supports the flat, fast leadership structure he wants. Snap is also pushing whole workflows into agents: a go-to-market agent takes a product idea through spec, sign-offs, legal and trust and safety risk analysis and launch materials, and the team uses Claude for much of this work.

His contrarian view is that humanity is far more important than technology because humanity dictates how technology gets adopted. He thinks the industry underestimates how much human comfort will determine AI deployment, and expects a lot of societal pushback. The industry should put humanity first and make sure tools advance humanity's goals as well as business ones.

On his own family, he describes a range of screen policies across his four boys, from none for the two-year-old to full tech engagement for the 15-year-old. He closes optimistically that computing can become more human, and says more will be shared on Specs later this year.