 | I built a fake radio station that is also, unfortunately, real. It’s called WRIT-FM. It runs 24/7 from a Mac Mini in my apartment. The whole premise is simple: an AI writes every word spoken on air, text-to-speech performs it, AI music fills the gaps, and a normal deterministic radio pipeline keeps the thing alive. The weird part is that it does not feel like a chatbot demo anymore. It feels like I accidentally hired five strange little night-shift employees who never sleep.
The AI part:
The non-AI part is intentionally boring:
No model is “deciding” to go live at 3:00 a.m. No agent is touching production controls. The AI writes the content; dumb code runs the station. That boundary is probably the most interesting part. The whole thing was also built with AI coding tools. The CLI, host system, scheduler, script generator, TTS pipeline, Icecast/ffmpeg streaming setup – all pair-programmed with Codex / Claude Code. Tech stack: Python, ffmpeg, Icecast, ChatGPT/Claude CLI, Kokoro TTS, Ace-Step, Mac Mini. I know “AI radio station” sounds like a gimmick, but after letting it run continuously, it feels less like a demo and more like a new kind of media object: not a podcast, not a chatbot, not a playlist, not exactly a simulation. Just a little machine that wakes up, checks the hour, puts on a voice, and starts talking into the dark. Radio: www.khaledeltokhy.com/airadio |
Key Takeaways
- The AI-driven radio station feels like a new kind of media object, distinct from traditional podcasts or chatbots.
- The separation between the AI writing content and dumb code running the station is highlighted as an interesting feature.
- This project demonstrates how artificial intelligence can be used to create a functional broadcasting entity without manual intervention.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
