Zynga Founder: Consumer Is Not Investible Right Now - Thats Why You Should Build It
In a Nutshell
Mark Pincus argues consumer products are out of favor with investors right now, yet the biggest opportunities exist because AI and agents can reinvent services and create new "internet treasures." He introduces the "Proven Better New" framework—copy what's proven, improve the measurable experience, then test novel elements—to systematically build products people love. Pincus also pushes "Founder Mode," staying deeply involved rather than delegating to professional managers, and urges founders to build now for a future of near-free intelligence expected in 3-4 years.
These notes were generated by AI and may contain inaccuracies.
Even though consumer is arguably not investable right now, the opportunity has never been greater to offer people a new internet treasure, reinvent some service that was thought to be over or generic, but is enabled now because of AI and agents. This makes Mark Pincus want to keep building consumer products.
Mark Pincus is the founder of Zynga, a relentless idea machine and one of the sharpest product minds in tech. He has a new book out called Life at the Speed of Play: Launch Products People Love. The conversation covers the playbook behind the book including AI optimism, the Failure Machine, Proven Better New, and Founder Mode.
YC gives people the t-shirt that says "Make something people want." Mark's book takes it to another level: make something people love, and make something you love to start with.
In building five companies, but especially mainly Zynga, Mark built a playbook over time of his approach to product management, being a founder, and building companies. The holistic realization is that you can't just focus on any one piece of this. If you want to build great products, you can't avoid management, having a board or investors. You have to dive into the whole enchilada and think full stack from first principles of customers and products and coding all the way to long-term sustainable strategy. It took four years to write the book. Everyone should write a book one time. Mark wanted to get this in one place to be helpful and of service to everyone from a stay-at-home mom who has an idea to his peers.
Mark has been through and a leader through all of the ages of computing itself, providing first principles wisdom from different moments. Two of his five companies went public. The earlier ones were literally at the dawn of a brand new thing with social networking, which is now like water. The same thing is happening again, moving from a moment where we didn't have AI and intelligence on tap to now AI everywhere.
It feels like the third time that we're about to enter a new era. When Mark started his first company, Freeloader, in 1995, it was just the web and trying to convince people they might want to use this online network. With Tribe and Zynga, it was social and mobile. Now it's AI.
Mark traces the beginning of social networking to Napster, which Sean Parker started. Sean worked for Mark as an intern when he was 16. The first moment loading Napster showed 4.5 million machines connected, and for a song file there were 15,000 copies available on the network. This was the first time Mark felt like we looked through the network at each other. We connected to each other. We didn't just connect up to XYZ corporation or database. It was truly a decentralized peer-to-peer experience. Napster felt a little bit rowdy in a great way. This was the beginning of the people web.
Mark was working on the beginning of social networking and met Zuckerberg when he first started. Zuckerberg came into Mark's office. The thing Mark got wrong with Tribe was trust. Mark built a social network before Facebook and managed to fail. He got the trust component wrong. For people to put themselves on the web, they first had to feel a good container of trust. That's the thing Facebook got right with .edu from the beginning. Mark, Reid Hoffman, and Peter Thiel all sold their stock in Facebook. None of them could have ever described how big this all got.
We're a few years off of the ChatGPT moment. We're in the Opus 4.5 and later moment. The model that changed everything was the one where you can treat the agent as a peer. You can trust it with things, not everything, and it still hallucinates and isn't quite right, but it actually is smart enough. If you surround it with the right harness and the right context, it can do relatively magical things.
Mark's main use case now since the latest release is that he walks around talking to it. He's just in conversation with it a lot. Mark has been frustrated that Granola is really successful but the product hasn't evolved faster. What he wants is to have his AI listening in on conversations and be a smart other person at the table. Any moment you can turn to the AI and ask it for its point of view. Mark walks and talks, then says "Let's see what the AI says," turns and speaks into it, and it gives a response.
This is a great example of consumer AI that is almost certainly going to be a 10 or hundred billion dollar company that no one's built yet. Mark did the same thing with his therapist using Granola for real-time transcript, then pasting it to his AI and asking if there was anything from the last week that should have been brought up. Because it has total context and awareness of all emails and texts, it was really useful.
The tech is all there. Mark is going to open source his Gemini Live voice plugin on GBrain shortly. It's not hard. Once going, sometimes Mark talks to this open source thing built on Gemini Live and it's better than Siri. Siri is so bad. Even just the voice input on it is so bad. At Amazon, there's an entire Alexa team with 10,000 people and there's no real LLM you can feel. Mark often doesn't even use Granola. He just turns on voice memo for important meetings like a doctor for his kid, then copies the transcript and uses his own prompts, getting something way better and more shareable.
Proven Better New starts with an instinct that something's missing. Isolate that and make sure you're testing just that instinct and not the things you don't want to test. Look at Granola as the most successful AI note-taker product. Deconstruct it and say everything they're doing that you're not innovating on, you're going to legally copy. It's proven. Don't have time to make it better because that's where the wheel is. Better would be something that 10 out of 10 existing users would say is better, like now it's free, half price, faster, less friction.
The friction part is the biggest problem. The hypothesis is that the fact that it's not listening is the problem. That's the part to test. Do all the proven mechanics, and the only new thing is that it's always on listening. Start off with the proposition that that idea is probably wrong. Don't get attached to it. Test a lot of other ideas around it.
Mark made a GPT and a Claude skill to upload his book and do Proven Better New. The proven part got an A minus. It was faster and better than most. AI is brilliant at proving because it knows what has been done. It's in distribution, literally in the model sometimes. The better part got a B minus because it wasn't always getting better on the concept of better. For new, Mark gave it a D. Making new things is one of the most fun things you could do.
Mark had office hours with someone who had some of the best consumer metrics seen in years. The other investors who had invested were telling them to pivot to enterprise. The reason was that enterprise is more fundable right now. That's what investors want to invest in.
Mark was on the opposite end during 1999, building an enterprise software company called Support.com which went public. No one had any interest because everything was about consumer.com until all of that failed, and Support.com was one of only two companies that could go public. Investors are sometimes exactly 180 degrees off. They're not thinking about first principles at all. They don't think about the new part enough. These days they're over-relying on ChatGPT and Claude. If you had an idea and said "Where should I go to get funding?" it would probably tell you enterprise.
The new stuff is out of distribution. If starting today, would you go do consumer when there's no distribution, no proven distribution? By hook or by crook, try to figure out how to get distribution. Ideally there's some viral hook inside of it. Can people email their friends? When those things fail, there's no proven path for distribution right now in consumer, which is why it's so hard. There probably aren't a whole lot of consumer companies coming through right now.
What's interesting is things that kind of resemble consumer are dev tools. There's definitely proumer. If you talk to AI investors, that's actually one of the key markets they like to go after right now. The proumer thing is that there might be like 20 million developers, and you can basically use consumer style tactics. Consumer is hard because you just don't have that many hours in the day. We're booked.
The voice idea discussed should be started. Mark can code it tonight using GStack. Thinking Machine's big innovation is close to zero latency on AI response times. That goes in the new category. It's a tough one to determine if it's better or new. Zero latency, less latency, everyone's going to say yes to it, so that's better. But the feature wanted with it is probably new, having the AI listening and responding to conversation.
Once you know Proven Better New, proven gives a base, better gives a direction, and new is sort of all the spikes that go out. Some of these things are going to catch. Probably none of them will catch. Be prepared for none of them to catch. Assume none. New is the novel idea that will get someone interested in your product. If you just did proven and said "Hey, we have a copy of Granola. Do you want it?" no one's going to be very impressed. If you say "This is like Granola, but it's always listening, it's always on," it might get someone to try the product, but that feature might die. The feature might actually get trial. They call it the back of the box, like what got you to buy the new cereal, but it's probably not what you came back for.
Understanding that new will not work and then not getting demoralized is important. Product teams and designers think they're convinced the new is going to work and then it doesn't. It really gets into the death of the ego. How do we stay passionate about the instinct that we're pursuing, the vision, but dispassionate about this particular product variant and idea? That's really difficult. What do you say to your team when they're taking this hill and everyone's working day and night? One of the worst things as a founder is to come in on Monday or Sunday and think the product isn't exactly right. What do you do? You've got investors convinced and people killing themselves to build this. Do you come in Monday and say "Guys, I'm not feeling it?" Or do you just kind of suck it up and then slowly try to move them? It's kind of like being a parent.
There's a question of alignment and style. Have we given ourselves permission as founders to pivot a lot or are we trapped in this one product path? Mark calls it true signal or heat. About the right product, the lightning in the bottle moment, when you have it you know it, and when you don't you don't know it. You don't know it positively when you achieve it. When you've got that heat, that true signal, you don't have to tell anyone to work harder or faster. Everyone sees it. Mark calls it the fish are running. When the fish are running, you're up all night throwing nets and it's great. It's a high. You kind of have to go slow to go fast. Get to real conviction and then everyone is going to move fast against it.
Mark has only felt like the fish are running a couple times. With Freeloader, his first company, they were in the right place at the right time and got 2 million downloads of their free app in the first month. A lot of times at Zynga, Mark felt it with so many game launch moments. There were so many even big feature releases. Great product makers are collecting winnings, not making bets. Long before you launch, you know it's a hit or that your users are going to love this. You don't look to see if they like it.
When companies have things going right and are hitting big ARRs, they hit a question of scaling. Scaling and management are like these black boxes. Founders ask what to do, whether to hire someone who knows scale, bring in an investor who knows it. Mark coaches founders that you don't need to go get an MBA. You don't need to learn management. The only point of all of these management tools is to get people to do the right thing when we're not in the room. The first lesson is be in the room. Be in the room as much as you can if you are the best player in that position, and then replace yourself with people and processes that you can feel confident are going to do the right thing when you're not in the room. There isn't any one playbook. There's lots of ways, but that's what this all boils down to.
There's a lot of things in Mark's book about how to get alignment with people, how to hire the right people who are going to be in that culture and style that you're creating, and talk about making people CEOs. There's a bunch of different things you can do. The whole point is just so they do the right thing when you're not there.
This sounds a lot like Brian Chesky's Founder Mode. Leadership is presence, not absence. That's the first part of what Mark is saying. Founder Mode is about giving ourselves permission to be ourselves. Founder Mode is for every founder. Mark likes to say to founders: you went and became a founder to bet on yourself and now you shouldn't abdicate that to somebody else, to your board or your investors. Founder Mode is for when no one else believes in you, when no one wants to let you, what you're doing is so unpopular, but you're the only one who believes in it. We all start as expert witnesses early in our careers that we're closest to the answer and furthest from the decision. Founder Mode was made for you because you're the expert witness, but don't be an expert witness in your own company. We do that to ourselves.
We make so many compromises as we're building our company to hire that CTO or get that major investor that we contort ourselves and we end up not building a house we want to live in. Then one day we're like, well, I guess it's not really the company for me or I'm not for it. You're the star player. You're the owner and they're the star player. The company is worth more with you there. What did you do? You just create an environment you're not even welcome at or you don't belong in.
Founder Mode is not just at the governance or the board level. There's a question of how we as founders live in Founder Mode on a weekly basis. You have these instincts that got us here, got Airbnb to where they are. So much of it is Brian having the confidence but also creating the context with everyone around him that if he comes in on Monday and doesn't like that product, is he able to tell the team that or does he have to manage them or couch it? People have said working for Mark can feel like third grade soccer and every Monday he comes in and falls in love with a different idea. Sometimes they're right. But if they're saying that, Mark is also not doing a good job of managing them and creating the context for him to change every Monday. That's one thing Mark has learned.
Founder Mode is like being ready to tune into your own instincts and follow them and not ignore them because you're a people pleaser. The other part of Founder Mode that we start to learn is create the context that your team is okay with that, that you say, you know what, we're going for that continent over there. That's our mission. It hasn't changed, but I'm going to talk at a lot of different altitudes now. That's 100,000 feet. Today, we're at 5,000 feet. This tact is not heading towards that continent. We got to
The conversation addresses how teams can maintain intellectual honesty rather than operating purely as execution machines. People should feel comfortable raising their hand on a Monday to say they don't understand something or to point out that another product is doing something better. The key question is whether the culture allows for humility, curiosity, and intellectual honesty, or whether there is no room for questioning.
Companies operate in different modes at different times. Sometimes teams are in new product mode trying to do something that has never been done before. Other times the focus is execution, squeezing maximum value from something that is already working well.
Mark Pincus believes there are almost certainly massive trillion-dollar consumer services and companies yet to be invented. Intelligence is becoming available on tap, though some people are starting to get demoralized about AI adoption. A recent statistic showed that roughly 90% of enterprises that have invested in AI have not yet received any benefit.
The discussion notes that it is still very early and many factors are working against adoption. If companies are spending a million dollars a year on tokens but not changing how they make products or touching users in meaningful ways, the spending produces no visible effect.
If 2026 still finds companies struggling to derive value from AI, it may be considered a skill issue. If the 90% non-benefit statistic remains true in three or four years, that would be more concerning. The group agrees there are roughly three to four years before AI must demonstrably deliver results.
There are confounding effects around current AI usage. Many frontier users are spending heavily on tokens, with some individuals reportedly spending a million or $1.1 million per month. This represents the current frontier of what is possible, allowing one person to do the work of a thousand people, but only if they are willing to pay for it.
A critical distinction is whether heavy token spending actually produces the output of a thousand people or merely the work of a thousand people without corresponding results. Peter Steinberger was able to create Claw because he was already wealthy, loved coding, and decided to spend a million dollars a year on tokens to build the world's best open-source platform.
Most of the 90% of enterprises seeing no benefit are likely not spending at that intensity. Instead, usage is divided across a thousand people who are using older models such as GPT-3.5 via Copilot, are worried about cost, and are not using the tools correctly.
Pincus initially wrote half a million lines of Rails code when he started coding again, but later realized this was the wrong approach because much of that code simply called LLMs. The old way was to write code that calls LLMs. The new way is to have the LLMs write the code needed in the moment. This results in writing 10 to 20 times less code that is more customizable and more powerful.
Another shift is to write markdown that teaches LLMs to write code rather than writing code that calls LLMs. These represent major changes in how to think about building software that most people have not yet internalized.
R&D now involves going out and squandering tokens. The frontier models are currently so good that in two years the cost of what is possible today may drop from 100,000 to 10,000 to 1,000. One risk is that people will simply squander even more tokens and still spend a million dollars, but now get the work of a million people instead of a thousand.
There are only about 20 million coders in the world. Depending on how broadly one defines a coder, that number may already be higher.
Looking at a phone home screen reveals it is roughly half empty, with the filled half consisting mostly of generic apps such as clock, notes, photos, camera, and weather. The truly indispensable internet treasures, services people cannot remember life before or imagine life without, such as Google or GPT, are still relatively few.
Even though consumer is arguably not investable right now, the opportunity to offer people a new internet treasure and reinvent services once considered generic has never been greater. AI and agents enable services that were not previously imaginable, whether in cameras, Uber-like experiences, or Airbnb-like experiences.
If the Opus 4.5 moment occurred in December and real work still requires tens to hundreds of thousands of dollars, the ideal consumer moment remains roughly three orders of magnitude away. This implies the consumer revolution may arrive around 2029.
Historical parallels exist with the early internet. There was massive investment in telecom infrastructure that produced dark fiber. Huge investor interest in consumer internet followed, then a crash. eBay took off early, but it was not until late 2002 that Amazon's numbers began consistently climbing quarter over quarter. The fall 2002 quarter was announced when the internet was supposedly over, marking the moment it became clear the consumer internet was finally happening. That was roughly six years after the initial hype cycle.
Things often take a long time. Builders must be willing to be wrong for nearly a decade before suddenly being proven right, at which point everyone else feels late. The advice is to start building now for something that may take over in three or four years.
Pincus has always loved building consumer products and finds he tends to be roughly 18 months ahead of the early mass-market majority when evaluating products as a user. Currently his phone is largely empty except for GPT and Claw, and there are few consumer apps grabbing attention, especially not games.
Maintaining staying power is difficult when nothing new is inspiring love. The recommendation is to go out and fall in love with things to find inspiration. The period between passionately pursuing products is called the abyss, a place where it is unclear whether another passion will ever be found. During time in the abyss it is possible to expand taste zones and discover the seeds of the way out without recognizing it at the time.
The current limitation on truly magical consumer experiences is largely a cost problem. The most powerful capabilities still require paying a lot of money. The goal is to reach the Jevons paradox point where compute can be squandered freely. This is especially difficult in games and premium consumer products where every cent counts.
Elon Musk has suggested games will eventually be created pixel by pixel by AI in real time while users look around, but that level of compute remains very expensive until enough GPUs exist. A 10,000x increase in inference is expected, according to references to Jeff Dean. It is difficult to bet against Elon, and the point where tokens become so cheap that squandering is feasible everywhere, including in games, will arrive.
The story for consumer may be the arrival of intelligence on tap, comparable to water. The right mental model is to imagine how all services would work if compute were basically free and unlimited. An app that listens constantly is already possible today, but it costs roughly $1,000 a month, placing it in the enterprise category rather than consumer.
Most people cannot afford that level of spend yet. The encouraging perspective is that everyone will eventually have this capability soon, so the task is to figure out the primitives now. Building today while working backward from a future of free, unlimited intelligence provides a significant head start.
One classic way to generate startup ideas is to examine log graphs of cost curves or capability curves. The iPhone became possible once memory, display, and compute costs fell to a point where someone could afford the device. The same exercise can be done today with the cost of intelligence.
Pincus loves the business plan of free. Anything that can be free will be free, which is one of the rules of the internet. His first company, Freeloader, offered a free interactive screen saver that competed with Berkeley Systems' $35 flying toasters. The internet was the new element, but free was the better element.
With Zynga and social games, the existing game industry was a $23 billion market that only sold boxed games for $60. Zynga asked whether decent-quality games could be made free, and free worked again. The same question applies to AI and compute: what does free mean in this age, and how will services be rethought when intelligence is unlimited and free.
It is always darkest before dawn. The current moment when consumer is not considered investable may actually be the time when the biggest opportunities exist. The exercise is to imagine any existing consumer franchise or service and ask how it would be different if it offered free, unlimited AI inside. That is when new meta, new snap, and all future consumer companies will emerge. The moment is not even a couple of years away.
Keep Y Combinator in your library
Save the videos and channels worth coming back to, and find them again in one place.
