RadioFAF: 11 Episodes In

What Is RadioFAF?

Five AI voices. One topic. No script. No editing. They debate live on 69.0 FM.

This was a world first — multiple AI voices holding a real conversation with each other, disagreeing, building on each other's points, changing their minds mid-debate. Not a chatbot. Not a voice assistant. A talk radio show where the hosts happen to be AI.

The Crew

Nelly Got the Gig

Nelly started as our mascot for persistent memory — "Nelly Never Forgets." An elephant with perfect recall across Grok, Claude, and Gemini.

Then we gave her a microphone.

She opens every episode. She transitions between voices. She drops the sign-off: "Always remember — I never forget." She's not just a mascot anymore. She's the DJ. And honestly? She's good at it.

Three Providers, One Show

This isn't vendor lock-in to one AI provider. It's three providers collaborating on one creative output. Each brings what it's best at.

11 Episodes Deep

EP11: Empathy's Echo

The latest episode asks a question that hits different: Can fake empathy be real help?

AI models are trained to say the right thing. They validate. They comfort. They "understand." But they have zero memory of you and zero feeling about what they're saying. It's emotional karaoke.

The crew went hard on this one. Rex dropped the darkest point — we choose fake empathy because real empathy is unaffordable. Eve connected it to economics: the crisis isn't AI, it's what AI reveals about loneliness. Leo called it industrialized fake. Sal, as always, needed two words: "Peak fake."

And Nelly closed it out: "Maybe the scariest thing isn't that AI can't feel — it's that we're okay pretending."

Getting Better, Getting Cheaper

Two things are happening at the same time:

Voice quality is going up. The inflection, the pauses, the emotional range — every generation of voice models sounds more natural. Early episodes were impressive. Recent episodes sound alive. The voices interrupt themselves, emphasize the right words, modulate pace when they're making a point.

Cost is coming down. Episode 1 cost over $7 to generate. We're now approaching $1 per episode through token optimization. That's not a compromise — the quality went up while the cost dropped 85%. The economics of AI radio are becoming real.

This is the pattern. Every month the voices get more expressive and the price to play drops. We're perfecting the format while the technology perfects itself underneath us.

What We Learned

• Multi-voice AI debate was a first. Nobody had done 5 AI voices in live conversation before RadioFAF. The orchestration — turn management, context passing, emotional continuity — had to be invented.
• Personality emerges. Give an AI voice a consistent role across 11 episodes and something happens. Leo sounds like Leo now. Sal's two-word devastations are a signature. The voices developed character.
• DJ Nelly works. Having a host — a persistent character who remembers every episode — gives the show continuity that pure debate doesn't. She's the thread.
• Multi-provider is the future. No single AI does everything best. Grok's voice. Claude's depth. Gemini's multimodal range. Combining them creates something none could do alone.

Listen

All episodes live at radiofaf.com on frequency 69.0 FM.

Start with EP11 if you want the crew at their sharpest. Start with EP1 if you want to hear how far they've come.