Liam's AI Lab
My AI experiments, in public. No hype, just the experiment, what happened, and what's worth borrowing.
Welcome to the lab. AI isn't coming for your job. Standing still is. So I'm running experiments in public, so you can see what actually works. No hype. No “10x your whatever.” Just the experiment, what happened, and what's worth borrowing. New experiments land here first.
AI Experiment 001: The Podcast Treasure Hunt
September 12, 2026
Eight podcast appearances, scattered across YouTube, Spotify, and a raw MP3 file. I consolidated all of them onto one page on my site, playable right there, with valid PodcastSeries schema so AI search engines can actually find and cite them. Four YouTube embeds, three Spotify players, one native audio player running 34:34. Now every conversation I've had about personal branding, LinkedIn, and AI lives in one place instead of dying in a feed. Full transcripts are on the page too, so the content works as hard as the audio.
AI Experiment 002: I Quit Asking AI to Draw Logos
September 12, 2026
My LinkedIn banner's tool icons looked janky because AI drew the logos from memory. The Claude asterisk was completely wrong, CapCut's mark was a guess, and the whole grid had that unmistakable AI-generated wobble. So I rebuilt all 20 icons from official vector assets, composited in code, tile by tile, at 2x for clean edges. The rule I'm taking away: never ask AI to draw a logo. Use official assets and let AI handle the composition.
AI Experiment 003: I Published a Website on GitHub Without Learning Git
September 15, 2026
I published a website on GitHub Pages without learning git. No command line, no commits, no merge conflicts. Just export, upload two files, flip one toggle, and it is live. The lesson: you do not need a developer's toolchain to ship on a developer's platform.