Matt Wolfe built three working AI tools in hours with minimal prompting: a thumbnail generator that mixes headshots and logos, a personal website built in 45 minutes, and a video titler that ranks titles by predicted performance. He shows how current language models let you make functional things almost immediately.
I've been using these tools, specifically GPT 5.3, Codex and Claude Opus 4. Six for the past week, like non stop. And they definitely feel different. I built a tool that generates thumbnails for me. I gave it a whole bunch of headshots and facial expressions for it to pull in a handful of logos that I often use and some of my past thumbnails of examples of what I'm looking for. And I just walked through this step by step process and it generates thumbnails for me now. I built this in a couple hours. I built this mattwolf.com site with basically one prom prompt. It's very simple, but it automatically pulls in the subscribers from my YouTube channel. Even this down here where it asks you to subscribe to the newsletter. This number right here is actually pulled in automatically from my newsletter subscribers. And this whole site was built in 45 minutes. I built a tool called Video Titler. When I'm done recording this video, I'm going to take the video file, drag and drop it right here. And it's going to suggest a whole bunch of titles for this video and then rank it on which one it thinks is going to perform the best. And then I'm probably going to use that title for this video. This was probably done with one prompt. Like, we definitely crossed a threshold with these models.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
