Gijs Verheijke

Back to articles

Generative AI opportunities

8 May 2023

I've been trying to start distilling some kind of framework to assess the opportunities, and felt confident enough to share.

Frontend: "We are an AI Startup", Backend: API call to OpenAI

Let's talk about Lensa AI. They were one of the successful first-mover money grabs, at one point making $2 million IN A DAY on the App store (after fees). Lensa build a simple frontend to generate avatars through Stable Diffusion, one of the big image gen models.

The party for Lensa is over, and bad news, you're already too late to partake in this get-rich-quick scheme.

So, what is to come for a company like Jasper? Essentially a frontend for OpenAI to write marketing copy, but with sizeable first-mover advantage and great execution on their go-to-market.

More importantly, if you had to start something now, or are an investor, what do you look for? Here are my thoughts:

  1. Proprietary data: Have some kind of data 'moat'. Internal company data, a primary or secondary usecase for your product that captures unique data, private data from consumers. Probably the most inevitable decacorn is Quickbooks but completely AI driven. So think smaller models, trained on internal company data.
  2. Play in an inaccessible niche, or one that is big but not big enough for OpenAI: High-end preventative medicine, nutrition, fashion design. Places where the data might not be proprietary, but for different reasons is unlikely to be part of the training data for OpenAI's models. You can also think of training an opensource LLM on niche data (like medical journals for example) ONLY. Again, think smaller models trained on datasets curated for a specific need.
  3. Tightly integrate with the workflow: I like because the product of the AI part (automated written meeting notes, with summary and action points) is provided with tools to continue collaborating like comments. It is a cognified internal information platform. These startups will not be AI startups. They will just be normal startups with AI. This will also be done by existing companies who do a good job in adding AI to make their existing owned workflows better.
  4. Killer GTM: Barriers to entry to make products are drastically being lowered by Gen AI. By how much, remains to be seen. But with screen time peaking and the amount of content/startups exploding 1000 fold, distribution will become 1000x more challenging than it already is.