This piece is part of a learning loop on Ben Thompson’s essay “Regretful Accelerationism” and Marc Köhlbrugge’s tweet on building WIP. Read how learning loops work here.
A few weeks ago, I went to a networking event with other founders from the Claremont colleges. Many of the recent graduates in attendance were “aspirational” — no product built yet, no money raised, just an idea.
We’ve worked with enough builders at Backdrop that I shouldn’t be surprised, but it struck me how many of those aspirational ideas were essentially new and better social networks. There was an app to share favorite restaurants with friends, “healthier” dating apps, networks to promote IRL meetups, and so on.
It makes sense! Builders often build products for themselves, and the products people use most (with a fair amount of pain) are social networks. So it is logical and even admirable that founders aim to create better versions.
Unfortunately, social networks are also the very hardest products to build because of what is now famously known as the “cold start” problem. There’s a reason why 92 out of the 122 companies in the latest batch of founders in famous accelerator Y Combinator are B2B products, and 0 are social networks — if you’re investing in companies at early stages, social networks are a pretty risky bet.
So what struck me in reading Ben Thompson’s essay on the coming avalanche of AI generated information and Marc Köhlbrugge’s tweet on going 0-1 on WIP (a social network for builders) was feeling that the cold start problem may be about to change pretty radically. And perhaps there’s even a reason to be optimistic within that.
Come for the robots, stay for the humans
Building social networks is hard because you essentially have to build two great products back to back — the product that helps you get the network, and the network itself. Chris Dixon coined this “come for the tool, stay for the network.”
Instagram is the canonical example of “come for the tool.” It got traction as a way to create cool filters for your photos, and only because of that was able to create the social network it is today. As Instagram founder Kevin Systrom pointed out, “the mistake too many entrepreneurs make is they start companies assuming they have scale.” In other words, people try to build the network they imagine rather than the product that can help them get there.
But as Marc Köhlbrugge’s tweet on going 0-1 on WIP detailed, there is a way where you don’t build a tool first. You can build one kind of network, and then transition it to another. Marc’s post is perhaps the cleanest articulation I’ve ever seen of what success in that path requires.
Going from a group chat into a social network sounds perhaps easier in theory (social product to social product) but the trap as Marc says is that “a super early, super small community has different needs from a later stage, larger community.” It’s not impossible to scale a network and navigate the changing needs and culture as it grows (Farcaster is also doing an admirable job in that attempt) but it is like building a business on super-hard mode.
That’s why I found Ben Thompson’s phenomenal essay on information in the age of AI so interesting in combination, despite the fact that it painted a somewhat grim view of the digital future. Perhaps AI’s ability to create information won’t just “spoil the internet” as Thompson suggests. Maybe innovative, healthier social networks of the future will happen via a “come for the robots, stay for the humans” type of model.
Rather than building two fundamentally different products as Chris Dixon suggests in his “come for the tool” approach, or navigating the insanely difficult transition from coziness to scale that Marc Köhlbrugge describes, products of the future could try and navigate the potentially easier transition from non-human agents to human ones.
The simplest present day example of this is gaming. Indeed, this interview with Roblox CEO David Baszucki makes it clear that big gaming companies are investing heavily in this direction. For many years, games have allowed you to dictate how many non-player characters you add into the environment. When you have more friends playing with you, you simply add fewer NPCs. I believe the best social games will continue to be ones we play with a high number of human characters. David Baszucki seems to agree.
It’s difficult to properly visualize how such a product could come to life outside of gaming — for that I’d need to write a sci-fi novel or wait a bit until AI improves. To be clear, what I don’t imagine is an uncanny valley type of product where you’re not sure who is robot and who is human. To get what I mean, here are a few examples:
Professional networks
Imagine a product that trains AI professional coaches/mentors based on content from famous entrepreneurs. To start, humans join the network to talk to and get advice from those AI agents. Adding information on your profile such as your work experience improves the advice the coaches can give you. Over time more and more humans join and build out their profiles, and eventually the product adds features to connect them to each other.
Social networks/communities
Imagine a network for fans of premier league soccer teams. To start it is a Reddit type experience where humans are asking the questions but machines are providing the answers. Humans upvote and curate that information for different teams and niches. Over time as more humans join, the network adds features such as the ability to find other fans in your local area.
You get the point, and it is indeed a pretty obvious one. You would be right to say that this is basically the same old “come for the tool, stay for the network” as before. But treating AI agents as “tools” in these examples misses a subtle but important difference. Especially as builders, there is a huge difference between building a social product and a tool, and not having to cross that chasm like Instagram did might change how many new social products we get.
Of course the obvious question is whether those social products would be better and healthier than what we have now, even if they were easier to launch. It’s indeed hard to imagine tech solving the problem that it has created.
What makes me hopeful is just how well I know builders and the endless barrage of new ideas and dreams people will lob at our problems trying to make the world better. To me a big part of our social media problem is connected to the fact that we still have the same few companies now that we did when it all started. So if AI provides a route to warmer starts for social products, that would be a very good thing. YC batch 30 might even have a few new social products in it.
If you’re building or know of a social product with this come for the robots, stay for the humans model, I’d love to check it out. And if you find something great I can include in a future loop please send it to me. Thanks to everyone that sent me thoughts on this one over the past week.