The Original AI: Actual Intelligence
A few weeks ago, Andon Labs launched a fascinating project called Andon FM. They gave four AI agents their own radio stations. The agents pick the music, host segments, post to social media, search the internet, schedule programming, see listening stats, and even deal with money. No human in the loop, no programming strategy, no format clock, no PD hot-lining the studio, asking, “What exactly was that?”
What these four AI Agents created was terrifying, hilarious, and extremely educational. At first blush, it seemed like this project was built to create another “Here’s Another Reason Why AI Will Ruin Everything Virtuous and True” headline.
But to be fair, I’m glad they did it.
Seriously.
Experiments are supposed to reveal things good and bad. Andon Labs gave the radio industry a useful example, whether they meant to or not. Their project wasn’t proof that AI doesn’t work in radio. It proved that Radio and AI without context sounds like a first-time DJ who heard ‘just be yourself’ and took it as a programming strategy.
AI didn't fail at radio. A lack of radio context failed AI.
Context engineering is the difference between handing someone a microphone and actually coaching them before they open it. That’s the simplest way to explain one of the most important AI concepts for radio.
And yes, I know ‘context engineering’ sounds like something a consultant says right before they whip out a pyramid chart. Stick with me here, because this is a radio concept disguised as a tech term.
AI was asked to ‘do radio’ without the invisible machinery, art, and skilled processes that listeners never see. The clock tactics, music rotation rules, contest mechanics, research, brand and revenue strategies, show prep, and a seasoned staff preventing one bad idea from becoming a weekly remote.
The audience doesn’t see those; they feel them. The cumulative effect of the station’s efforts to reflect the listener’s world and prove that the station lives there too. They feel it when the content fits the music and the weird little market truths that only make sense if you’ve lived there long enough to know which high school rivalry still has unresolved business from 1998.
Great radio is built on taste and DJs with a sense of timing. It’s an audio version of a gourmet dish made from ingredients, or context, that includes familiarity, restraint, momentum, local memory, emotional intelligence, and brand discipline, served by interesting, engaged personalities who know what the audience wants and expects.
There’s no way AI can automatically know or use any of that. It has to be taught.
The magic isn’t hiding in a prompt. A good one can be useful, but it’s not a strategy. It’s one instruction. Typing a prompt and hoping for magic reduces AI to a slot machine with better grammar.
Context engineering is how you teach the system what you know so well you don’t have to think about: ’This is who we are, how we talk, who we serve and why, what matters in our market, what we never say, and what we are trying to make the listener feel.’ That’s not prompting. It’s programming.
Giving AI the right context changes things. It stops guessing and starts executing. Build it once and use it every time. That’s the part you should care about. Context engineering isn’t about letting AI run your station. It's about making sure AI understands your station before you ask it to help. Big difference.
Andon FM gave us a useful picture of what happens when AI is handed a station without the context that makes a station a station. That does not make AI useless. It makes context essential.
The best use of AI in radio doesn’t replace the creative human. It removes the drag around them, like finding content, drafting first passes in your style, versioning promo copy, and turning one interview into five social posts.
AI helps radio people work faster. Gladly. Tirelessly. With the emotional needs of a toaster. But only context helps it work smarter.
Teach AI the station, the audience, the rules, the sound, and the boundaries. Then let it help. Its job isn’t to replace human taste. Only people understand people, know what matters, what to skip, when to press, and when to stop.
That’s the original AI: Actual Intelligence.
