Copy Proven Mechanics Before Testing New Product Ideas

Mark Pincus wrote Life at the Speed of Play partly as a product-management book and partly as a broader argument for approaching work with more play, speed, and experimentation. He says the book was originally called Proven, Better, New, because that framework felt like its core value. He later came to see the framework as serving a larger idea: in an AI era, more people can try, build, and test ideas without quitting a job, mortgaging a house, or raising venture capital first.

Pincus frames product building around a distinction he says founders often blur: instincts may be right, but ideas are usually wrong. The instinct is the felt recognition that something is broken, missing, or newly possible. The specific product idea that follows is much less reliable.

His example is a dating product such as Raya. Many people may have had the instinct that dating apps were broken or that a better version should exist. That does not mean they had the idea for Raya. The practical consequence, in Pincus’s framework, is that a team should not treat its first expression of an instinct as sacred. It should “prosecute” the instinct scientifically, in his words, rather than defend the idea emotionally.

That is the job of “proven, better, new,” the framework that Pincus says came out of the failure of Tribe and the subsequent commitment to winning at Zynga. It is a way to “isolate and box in your innovation in one place”: decide what is already proven and should be copied, what is genuinely better in a way users will immediately recognize, and what is new enough to attract attention but likely to fail.

The first category is the one Pincus believes founders often resist most. “Proven” means the mechanics that already work for users on an existing platform or in an existing product, where the goal is not to improve them. If those mechanics are legally copyable, he says, copy them. Reid Hoffman adds the working rule: stay as close as possible without doing anything illegal.

Pincus uses Threads as a contemporary example: Mark Zuckerberg’s fast follow of a successful product was, in Pincus’s grading, an “A minus” application of proven, better, new. The art is not merely copying. It is copying so well that the end user does not experience the product as a blatant clone.

Hoffman reinforces the point with an older example from 2004, when he went to Shenzhen and found three Chinese copies of LinkedIn so exact that he could fill out a profile without speaking Chinese. For Hoffman, that was “proven” in its most literal form: the game mechanic, product mechanic, user demand, usage signal, and value proposition were visible enough to replicate.

Pincus argues that junior or amateur product builders often do the opposite. They change too much because they feel a need to make the product their own. That can produce the wrong kind of failure. If the proven part is not executed well, the team may get a false negative: the product appears not to work, but it failed because the team broke something that was already known to work.

“Better,” in this framework, has a stricter meaning than founders tend to give it. What the team believes is better is, to Pincus, actually “new.” Something is better only if existing users of the product would immediately agree. His phrasing is intentionally blunt: 10 out of 10 existing users should say yes. Better is usually basic: it is free now; there is no download; a detail users already care about is polished “one inch” more. Hoffman describes it as improvement inside a familiar consumer experience, where the user can instantly say, “yes, this is better.”

“New” is the risky part. It is the “back of the box,” the new-and-improved hook that may be the reason people try the product and also the reason it fails. Pincus’s bad example is Risk, the board game. A version he disliked replaced the wooden pieces he loved with plastic ones and Roman numerals. For him, that was not better: he wanted to show his force, not interpret Roman numerals. The same box also advertised Castle Risk, a variant he never learned. His label for that kind of change was “proven worse and bad new.”

The lesson is not to avoid novelty. It is to assume that the first new idea is probably wrong, and plan accordingly. Pincus says teams should ask what the four other possible new ideas are, and where they can look in adjacent products or versions to learn faster. The operating posture is detachment: give the team permission to change everything in the risky zone because the instinct matters more than the first idea.

Reid Hoffman presses the framework against AI, where one obvious problem is that there is still relatively little “proven.” ChatGPT is proven in some ways, Hoffman says, and Claude Code is proven in some ways. Pincus accepts that, but calls them “pretty expensive proven.” His “10 billion” versus “100 billion” line lands as joking shorthand rather than a concrete market estimate: the point is that some AI examples may be proven without being cheaply copyable, especially once Hoffman adds the need for “the compute cluster.”

The harder question is whether AI is already a consumer platform. Hoffman separates two meanings of platform. AI is certainly a platform technology for consumer applications: underlying models can support many different agents. But there is also the “front door” sense of platform, the portal people habitually enter through. Google became synonymous with search because it served that function.

Hoffman thinks people will have an agent, or perhaps a couple, that becomes their general front door. He says ChatGPT is closest to that position now. Pincus agrees that ChatGPT is closest, but argues that AI is not yet a consumer platform. By comparison, early Google was already a platform enabler and magnifier for web distribution because SEO and SEM existed around it. AI has not yet reached that kind of consumer distribution structure, even if it is huge.

Pincus’s answer for builders is to look for “fuzzy proven.” If there is not yet a settled consumer AI playbook, a founder should search for experiences proven somewhere else and ask whether they can be applied in the new context. That is still a better vein than trying to invent something “the world’s never seen before.” Behavior, Pincus says, has a way of staying the same even as the surface changes.

The historical analogy they both reach for is the consumer internet after the early crash. Hoffman recalls a period when the internet was considered “concluded” by incumbents such as Yahoo and Amazon, and Silicon Valley moved heavily toward enterprise software or cleantech. He and Pincus thought the consumer internet had only just started. Hoffman sees the same pattern in current AI: enterprise is where many investors and operators see the money, but he believes consumer AI may be much larger than people expect.

Pincus compares the moment to the “internet nuclear winter” of 2002 to 2003, when Hoffman was starting LinkedIn and Pincus was starting Tribe. The market had not figured out the next consumer pattern, but both believed it would. They backed Friendster not because they believed that specific company would necessarily win, Pincus says, but because the experiment needed to happen for the consumer internet. What surprised him later was the sheer demand: Friendster did not understand virality or growth hacking, he says; “the wave picked up the boat and a tsunami moved it along.”

Pincus expects AI to produce many more product makers, not just more products. He describes his partner Hillary using Claude Code to build a small mobile app for their family and nanny. Based on location, it recommended activities for each of their five children by age. He was struck by the design aesthetic and by the fact that she had moved ahead of him in “vibe coding.” His daughter Georgia, he says, is making a poker app. The conclusion he draws is that when the stack becomes automated and promptable, many more people become full-stack product makers, and the differentiator shifts toward taste.

Hoffman says Zynga helped write the modern book on game dynamics: product design, interaction, mechanics, and the way games such as FarmVille and Words With Friends became household names. Mark Pincus says games are a particularly clean canvas for learning proven, better, new because they can be broken into mechanics. But he insists the framework applies to consumer products generally.

For AI-native companies, Hoffman suggests some obvious Zynga-derived lessons: launch early, learn through engagement, and perhaps build only levels one through ten to see whether the core works. Pincus agrees, but sharpens the advice. His line is not “build it right.” It is “build a culture of building it wrong.”

His objection to “minimum viable product” is the word “viable.” Pincus says he admires Eric Ries and shares the same spirit, but believes “viable” is harmful because it smuggles in the hope of viability. Viability sounds weak to him. The stronger thing to seek is conviction.

What he wants founders to copy from Zynga is a testing cadence: test more ideas in a week than the industry tests in a year. In consumer products, he says, the testing and failure machine should usually operate at the top of the funnel. The first question may be as basic as whether anyone will click. If that is the question, do not build the product. Build the landing page, or the ad link, and nothing else.

Pincus sees a paradox in AI. Because AI can help teams get to something that looks like a minimum viable product in three months instead of three years, they are tempted to do exactly that. But then they become attached too soon. They have “a lot in this,” so they keep iterating rather than killing the idea. The efficiency of building can make the emotional cost of killing harder, not easier.

His preferred test is not whether a launch is polished. It is whether the product has been turned on to the level needed to learn. He and Hoffman both invoke the familiar idea that if you are not embarrassed, you are not launching early enough, though Pincus says “launch” may itself be the wrong word. The point is exposure calibrated to learning, not a theatrical release.

This connects to his definition of ambition. Founders often say they are a 10 out of 10 on ambition. Pincus argues that real ambition includes the willingness to be wrong for a long time. A founder can be wrong for three years and then become “really right,” or be “a little bit right every day.” Most people, he says, choose the latter.

He remains, like Hoffman, an extreme AI optimist. His optimism is not limited to productivity gains. He expects a proliferation of new kinds of services “for humans, by humans,” and says that, as with the internet, web, and mobile eras, many future industries are not yet imaginable.

Reid Hoffman raises the question from the perspective of a parent speaking to someone who helped build an industry often accused of addictiveness: what should parents and the industry do about phones, games, apps, and AI potentially being used against users’ attention?

Mark Pincus acknowledges the irony that many people who build apps shield their own children from them. He says he tried to keep his twin daughters, now 15, from having smartphones until they were 16. They eventually persuaded him to get them a flip phone, which he framed as making them the coolest kids because they would be the only ones with one. They did not agree, but the compromise bought him a couple of years.

At 14, they interned at Niantic on Pokémon Go and built a new first-time user experience. Pincus says he was proud of their work, and they convinced him they needed phones for it. He was not able to take the phones back, so he put them “permanently on probation”: if he saw them on phones at dinner or not paying attention to younger siblings, the phones would be gone forever.

His broader position is that families should look at what adults model. He says his daughters are on their phones much less than other kids, which he considers healthy, and he has noticed that long screen sessions can leave one of them more isolated, flat, or sad. In those cases, he takes the screen away and tells them to go outside.

Pincus does not primarily blame Facebook or games. He says he does not think the analogy to cigarettes or drugs is quite right. Screens are now part of culture and society, and parents are constantly on their own phones. Blaming companies while modeling the behavior ourselves strikes him as hypocritical, though he also acknowledges the real challenge: the pattern is “kind of giving all of us ADD” and making attention harder.

For his younger children, one year old and four years old, Pincus says his partner is vigilant about no screens, and many new-parent friends take the same approach. He does not claim to have a magic answer. His rule has been to keep children off screens as long as possible, then steer obsessions toward healthier forms. Georgia’s current obsession with chess and chess.com feels more acceptable to him than other kinds of screen use. Hoffman adds that obsession and commitment can contribute to success, provided the version is healthy rather than unhealthy.

The screen activity Pincus actively wants children to use is coding with AI tools such as Claude Code or Codex. His ideal is not general screen immersion but efficient learning, building, and outdoor life. He says watching high school now is painful because it feels like the end of a hundred-year cycle: students spend all day in school, then come home to large amounts of homework. His children homeschooled for a semester, missed the social environment, but told him they learned more and had far more time. In his ideal world, learning would be much more efficient; if children were on screens, they would be making things with AI coding tools; otherwise, they would be outside.

That same distinction shapes his view of games. He wants games and gamified services to become a better use of time: more playful, more social, and more connected to real-world skills and outcomes. His complaint is not that games are powerful. It is that, given the size of the industry, they should do more for the people playing them.

Mark Pincus says the metaverse is starting to happen and may become as big as Zuckerberg thought — just not inside an immersive goggle that is painful and hot to wear.

Pincus’s metaverse is less dependent on 3D embodiment. It might involve devices that listen, AR Ray-Bans, or simply the phone people already use. His word is “metaversian”: more products and services becoming playful, social, gamified, and connected to real-world activity.

For games, he sees the unlock in bringing more of the real world into them: real humans, real social interaction, and real-life value. Games, he says, should be a good use of time. Given the scale of the industry, he calls it a waste if games do not improve life skills. He imagines games doing something more like LinkedIn: helping a person’s real-world career through play, even to the point of getting a job by playing a game.

The same movement goes the other direction. Services such as Duolingo, and apps more broadly, will become more gamified. Winning services will become more fun and rewarding. For Pincus, that is part of the promise of the metaverse even if it does not look like an immersive virtual world.

Consumer agentic AI is where the idea becomes explicit. When people have avatars roaming the world while they are absent — networking, finding dates, finding jobs, and discovering ways to make money — Pincus says that is the metaverse. It does not need to be immersive 3D. It could arrive as a message: an AI version of Hoffman gave a talk in Bombay and “crushed it.” The value, in his framing, comes from agency, social presence, play, and real-world outcomes more than from wearing a headset.

Mark Pincus recalls a junior partner at Accel expressing concern, during his Zynga fundraising period, that he was not coachable. Pincus’s answer was that he had many coaches, but he had chosen them. He did not consider himself coachable by a venture capitalist who had never held the operating job or sat in the founder’s seat.

The implied advice, as he heard it, was to perform deference to the person with money or to the supposed adult in the room. He says he has generally refused that posture.

Reid Hoffman agrees and names a failure mode in venture capital: the belief that pattern recognition is enough. Patterns are useful, he says, but investors also need to know how to play the game. Pincus adds that if all you do is look at patterns, you may miss the “pattern interrupt.” Hoffman calls that interrupt the source of discontinuous investments.

For Pincus, this is the same strategic lesson in another form. Everyone else is looking at patterns and playing the same game. If you expect a different outcome, you need a different strategy in your back pocket.