The first text rendering technology we developed for our meme generator ended up being reused in Figma, showcasing how early experiments can inform future products.

AI app developers often treat system prompts like traditional code, maintaining a divide between users and developers that limits customization and personalization, leading to AI applications functioning as lowest common denominator software with generic outputs that lack individual expression.

Creating an app that allows youngsters to express themselves in a coded language, which adults cannot understand, could be a successful venture.

Anthropic's 3.5 Sonnet model was the critical breakthrough that made AI-generated code production-ready, unlocking the entire text-to-app market.

Vibe coding products like Cursor and Bolt are emerging, allowing users to create applications from text prompts, although they have yet to reach mainstream adoption.

Imagen 4 excels at generating text and typography, allowing for creative choices in design, such as using dinosaur bones in fonts for a music festival poster.

Plant identification apps can generate significant revenue from a small user base, as they cater to a specific audience willing to pay for their utility.

I experienced a moment of feeling AGI when the AI agent demonstrated good intuition in app design, suggesting features like a slider bar with emojis without being prompted.

My favorite feature of Codex CLI is its multimodal reasoning, allowing it to generate code from sketches or notes.