The unseen side of building stuff in AI

Hi folks,

Last year I pushed some AI-related products with running now around 30K monthly users, just sharing some things I learned on the go.

✅ A Figma plugin to rename layers with on-device AI
Monthly users: 20K+
What I learned: Open-sourcing your project is trading short terms revenues for long-term marketing value. You have to plan that 80% of the time will be selling not building.

Connecting two fields that you know very well is the key.

✅ A text to 3D material generator (Acquired)
Monthly users: 2k+
What I learned: There is no edge in either AI or technology. The only optimization to find is in the market, not in fancy AI or tech.

If you know very well that niche AND you can program there is a chance you're part of the 5/10 guys in the world in a competition to ship a great product.

✅ An image-to-pixel art converter (Acquired)
Monthly users: 13k+
What I learned: Time flies, and building a UI wrapper for APIs call is not an opportunity anymore.

Build stuff for people that actually do AI stuff, like pre/post-processing tools, storage services, MLops tools, infra... But don't build AI-powered X stuff.

Always ask yourself "Is the more AI in the world the happier I am?" If the answer is no, you currently build AI-powered X and you probably have 0 edge because the next week Adobe will launch the same thing but they have the distribution.

Naive examples: If you build an AI writing assistant, the more time passes the less happy you are, because Notion has big teeth and strong distribution. But if you build a boring, non-AI product, image storage service, the more time passes the happier you are.

So tldr:

1: Think like if EVERYONE holding the last fancy tech and can ship the product you have in mind tomorrow. The tech should not be your edge.

2: You should have the energy to sell it, "AI" in the name is now a con, not a pro. Way too much noise.

3: Don't build AI-based products. When a gold rush kicks, you know what is good to sell ⛏️

4: If you don't know how to program, you should. No, ChatGPT will not program your idea, it might help to code but that's it.

When I say "you" it's not imperative, these are just extracts from my own notes, so basically it's for myself.

Cheers