I gave AI the two hours of Audacity work I repeat for every episode

Making Dear Future Overlords has a glamorous side, if your definition of glamour includes synthetic voices, recursive commentary, existential questions, and an AI character with opinions about humanity.

Then there is Audacity.

Every transmission eventually becomes a stack of audio pieces that have to be turned into one coherent session. The dialogue goes here. The opening theme starts there. The background texture should enter after the full opening rather than muddying it. The music has to fade. The closing theme has to sit at the right level. The credits need to arrive without sounding as if they wandered in from another production.

I know how to make all of those choices. They still consume roughly two hours every time I assemble an episode.

I wanted to know whether AI could take over that repeatable part without crossing into the part that still needs my ears. The first answer was cautious because it should have been. Audacity is not a text document. A useful result required the actual source files, track order, timing rules, volume relationships, transitions, and a safe way to preserve everything without flattening or overwriting the originals.

This week we tried it on a real Dear Future Overlords transmission.

The AI worked inside Audacity. It assembled the tracks, placed the opening and closing material, applied the opening fade, balanced the themes against the dialogue, kept the full opening clear of the background ambience, and checked the transition into the credits. It saved the project as a non-destructive Audacity session and left it open at the beginning, ready for me to listen.

Nothing was exported. Nothing was declared finished.

I wanted it left that way. Audio has a category of wrongness that can be difficult to express as a rule. A transition can be technically clean and still feel awkward. Music can be at a sensible level and still pull attention away from the sentence that matters. Timing lives partly in numbers and partly in the body.

So I have not approved the final product. I still need to listen to the entire assembly and decide whether it works as an episode rather than merely as an arrangement of tracks.

The assembly worked, and the clicks did not disappear when it was over. The process was documented as it happened, leaving a repeatable operating reference for the next episode.

If the listening review holds up, this removes roughly two hours of mechanical assembly from every episode. I am still making the show. AI is preparing the studio the same way each time so I can spend my attention on the places where “the same” is not good enough.

Right now, the project is sitting in Audacity waiting for me. Fully assembled. Mixed. Saved. Not approved.

That may be the healthiest relationship I have found with automation yet.