600+ Days of an AI Podcast I Accidentally Got Emotionally Attached To

Hello fellow dev.to readers! I don't usually post much but I wrote this for a specific openAI related community, and I thought I would also share it because I just like the vibe from dev.to and I always thought I wanted to contribute something - but I never find the time. So today I did, and here we are! This is a super short story about a fun little experiment I did. Almost two years ago, I was playing around with LLM apis building small things. Around the same time a friend and I kept talking about starting a podcast - but despite meeting every Friday for pizza, we never actually recorded anything. Instead of doing the obvious - like idk, recording over pizza? - I thought, "What if we get AI to write the podcast, clone our voices, and have it made automagically?" Back then, AI podcasts weren't really a thing. Google's NotebookLM wasn't making headlines, and it felt like a fun project to tackle. So it was still cool, in some nerdy way :P So I made a script that generated a short episode using AI to write dialogues for two hosts with a personality loosely modeled after me and my friend. I used TTS voices (not super natural, but decent and cheap enough to tinker with). The results were fun, so I took it a step further. I built an automated workflow that once per day picks a topic from a list, writes an episode script, generates audio, adds intro/outro music, and publishes it to Spotify and Apple Podcasts. The only human input is adding new topics via email or WhatsApp - and paying the bills (I'm still looking for a way to get AI to pay my bills!). I didn't automate the topics-list on purpose: That small creative task keeps me involved and makes me feel like I'm still part of the show. The episodes started to get auto-published and I began listening to them every day during breakfast or while driving. Each episode is about 4–5 minutes (I like to call it a micropodcast), so it easily fits into my day. At first, I was just curious to see what the AI would come up with. I always planned to upgrade the voices with our own cloned ones (ElevenLabs API was on the list), but here's the weird thing: I got used to the synth voices. I started to like them. I got attached to these synthetic, artificial hosts. One of them has an odd and lovely German accent that I find charming. Upgrading them now would feel like firing the hosts and hiring replacements. I just… can't do it. What surprises me is that even after 600+ daily episodes, I still enjoy listening every day. Even though I know exactly how it's made - or perhaps because I know how it's made. I never swapped out the voices for more natural ones, and somehow, in today's world of almost perfect voice models, these clunky, imperfect ones feel refreshing - perhaps almost like how many of us have re-discovered a love for vinyl records in an age of perfect high quality digital music. I'm the only one who listens. Spotify's analytics certainly suggest that. It makes sense, I'm probably the only one who can actually enjoy listening to this weird thing. Not even my friend, who is technically one of the hosts, tunes in regularly. It's like self-generated entertainment. Like if you would create a Netflix show just for your own entertainment. It makes me wonder: could it be that personal, self-catering entertainment is going to be a thing in the future? Or am I just weird?

Apr 11, 2025 - 17:04

600+ Days of an AI Podcast I Accidentally Got Emotionally Attached To

Hello fellow dev.to readers!
I don't usually post much but I wrote this for a specific openAI related community, and I thought I would also share it because I just like the vibe from dev.to and I always thought I wanted to contribute something - but I never find the time. So today I did, and here we are!
This is a super short story about a fun little experiment I did.

Almost two years ago, I was playing around with LLM apis building small things. Around the same time a friend and I kept talking about starting a podcast - but despite meeting every Friday for pizza, we never actually recorded anything.

Instead of doing the obvious - like idk, recording over pizza? - I thought, "What if we get AI to write the podcast, clone our voices, and have it made automagically?" Back then, AI podcasts weren't really a thing. Google's NotebookLM wasn't making headlines, and it felt like a fun project to tackle. So it was still cool, in some nerdy way :P

So I made a script that generated a short episode using AI to write dialogues for two hosts with a personality loosely modeled after me and my friend. I used TTS voices (not super natural, but decent and cheap enough to tinker with). The results were fun, so I took it a step further. I built an automated workflow that once per day picks a topic from a list, writes an episode script, generates audio, adds intro/outro music, and publishes it to Spotify and Apple Podcasts. The only human input is adding new topics via email or WhatsApp - and paying the bills (I'm still looking for a way to get AI to pay my bills!). I didn't automate the topics-list on purpose: That small creative task keeps me involved and makes me feel like I'm still part of the show.

The episodes started to get auto-published and I began listening to them every day during breakfast or while driving. Each episode is about 4–5 minutes (I like to call it a micropodcast), so it easily fits into my day.

At first, I was just curious to see what the AI would come up with. I always planned to upgrade the voices with our own cloned ones (ElevenLabs API was on the list), but here's the weird thing: I got used to the synth voices. I started to like them. I got attached to these synthetic, artificial hosts. One of them has an odd and lovely German accent that I find charming. Upgrading them now would feel like firing the hosts and hiring replacements. I just… can't do it.

What surprises me is that even after 600+ daily episodes, I still enjoy listening every day. Even though I know exactly how it's made - or perhaps because I know how it's made. I never swapped out the voices for more natural ones, and somehow, in today's world of almost perfect voice models, these clunky, imperfect ones feel refreshing - perhaps almost like how many of us have re-discovered a love for vinyl records in an age of perfect high quality digital music.

I'm the only one who listens. Spotify's analytics certainly suggest that. It makes sense, I'm probably the only one who can actually enjoy listening to this weird thing. Not even my friend, who is technically one of the hosts, tunes in regularly. It's like self-generated entertainment. Like if you would create a Netflix show just for your own entertainment. It makes me wonder: could it be that personal, self-catering entertainment is going to be a thing in the future? Or am I just weird?