Generative AI Founders Aren’t Thinking Big Enough
This is the first of a three-part series that I’m calling Future Generative. All views are my own and do not represent those of DCM.
Attention founders: Generative AI is a feature, not a product.
Like every VC on the planet, I’ve been talking to generative AI companies over the past few months. Honestly, many have been uninspiring. I mean, the first idea that some founders had after being given the most powerful predictive models ever was… a profile picture generator? I don’t deny that the pictures are cool, but for this to be a billion-dollar business, you’d have to believe that nearly everyone will start changing their profile pictures on a weekly basis and pay much more than the $29.99 per year that Lensa is charging. This is why I think it is no more than a tool:
Why are so few people building products? I have two guesses. The first is generative AI is currently in the “any sufficiently advanced technology is indistinguishable from magic” stage of its development. The sheer power of the underlying models makes it easy to build something that looks like magic — something that looks like a product — but it only looks like magic relative to the status quo. As the tech commoditizes and matures, this magic will disappear and it will become apparent that these startups are, in fact, features. Features must support products at the end of the day, which is why I think products like TikTok, Canva, and PicsArt¹ will be the biggest beneficiaries of these first-order generative AI applications.
The second guess is based on my observation that many startups in this space are personal explorations or hackathon projects. People are trying stuff and seeing how far they can push the tech just because they can. I love that spirit, but it usually results in curiosities, not businesses. That’s why these startups are characterized by rapid user growth followed by equally rapid user loss. Once the novelty effect wears off, the project dies.
Either way, generative AI founders need to think bigger. What “product” means is anyone’s guess, but I am bullish on verticalized platforms that use generative AI in an opinionated way to serve a function in which personalization matters. A generative AI product must be verticalized to comprehensively meet all the needs of a function. It needs to use generative AI in a way that not only adds value, but fits natively into how its intended users think and work. Ideally, the founders are former would-be users with specific ideas about UX, systems and flows, defaults vs. options, etc. And personalization must matter to the function because that is the only moat I currently see, at least for the vast majority of startups that rely on someone else’s model.
For example, I’ve seen dozens of generative AI companies working on ad copy and image generators. Personalization obviously matters here, so this is a good space to build in. However, none of them are thinking about fully managing campaigns.² If you can generate assets, why couldn’t you also
- Integrate with Google and Facebook Ads
- Automatically generate assets for each target customer profile
- Predict asset performance³ and automatically submit bids
- Dynamically adjust bids by integrating with traffic and conversion data
- Continuously optimize or improve the campaign by introducing newly-generated assets in response to what has performed well
The point is that generating ad copy and images is not an end state, but the start of something that wasn’t previously possible. Founders recognize that generative AI reduces the cost and time of creative to near-zero, but they strangely don’t realize the potential this unlocks.
The barrier to innovation in generative AI seems to be vision, not tech. This is such a rare position for founders to be in, and I hope more of them take advantage of it. If you’re unsatisfied with simply having these superpowers and are using them to build something bigger, please send me a note at jiwang[at]dcm[dot]com. Let’s show the world what generative AI can really do. ∎
- Disclosure: PicsArt is a DCM portfolio company.
- Of course, brands will want to manually intervene. But the point about vision still stands.
- There are various predictive methods, but my favorite is the approach described in this paper.