The Agent Said It's Done: a keynote that turns into an anthem

Watch on YouTube. A product keynote, played completely straight, tips into the absurd one slide at a time until Plan 9 trips over the power cord and a tape stop switches the hype off. Then the ghost light, the napkin wall, the roof opening, and everyone singing: it's the time of a lifetime, we're all beginners again.

Licensed CC BY 4.0.

The idea

A song that makes fun of AI hype, made with AI, and made better by measuring instead of trusting the tools' own claims. The joke wrote itself. The agent kept saying it was done, and the numbers kept saying "nearly".

The brief was a keynote parody that turns sincere. Warm, never bitter. It pokes at the launch, never at the people in the room, and it tries to make AI a little less scary. Everyone's a little scared. That's allowed.

The keynote

The Governor hosts, deadpan, and never cracks. Think Monty Python by way of the Hitchhiker's Guide, with a Kubernetes cluster where the towel should be. "Good morning. Today we are announcing... everything." "Four hundred pods. Per namespace. Per nanosecond." "Please hold your applause until the product exists."

OG Bobby Johnson, the Doberman, raps the verses. Ten-x on the coffee. A startup founded Friday night and funded by the weekend. An agent hired for the job, which promptly hired agents. Plan 9, the cyborg bunny, interrupts on a glitch stinger, a little too sincere at the wrong moment. And the chorus asks the only question that matters in a keynote: disrupt the disrupters, all just because?

Then, in the second pre-chorus, the Governor sings. "When we go to the cloud and into k8s, we are all equal and the same. La la la." It works on my cluster.

The turn

Chorus two, and the power cord. The tape stops, the slides go dark, and nothing is a joke anymore. "I had more slides. I'm not going to use them." The three sit on the edge of the stage under a single ghost light, and the napkin wall behind them draws the printing press, the power line, the light bulb. Every new thing was scary once. Nobody knew what it was yet, either.

That stretch is my favourite part of the film, and it was the hardest to keep. More on that below.

The anthem closes it out. The roof opens, six friends walk on and stand with the three, the audience doubles and their lights sway together, and the whole room claps on two and four. It started as a joke about a keynote. It ends as a sing-along about being beginners together.

The voices, made to sing

ElevenLabs cannot sing. Every voice here is speech made to sing: each syllable found in the audio, locked to the beat or to its note, and the sung lines resynthesised on the melody's exact pitches with a vocoder that keeps the consonants from the original recording, so the words stay words. Every line was then checked by a speech recogniser for how many of its words a listener actually hears.

Kiara sings short Hebrew lines under the first verse, autotuned into the song's own tune: everything is new, we are all beginning, don't be afraid, together we'll learn. A background voice in an ancient language, holding the room while the rap finds its feet.

Twelve passes, measured

The first pass sounded fun but loose. So I measured it, and the rap sat exactly as far from the beat as chance would put it. Every fix after that started with a number:

• On the beat: 31 ms off the grid, down to 14 ms.
• In tune: a semitone off, down to within 0.05 of a semitone.
• Heard: Bobby's verse words heard in the final mix went from 41% to 82%.
• Clean: clicks in the verses, 18 down to 5.

And the twist. The section I loved most, the Turn, had quietly drifted an octave between renders. Each pass "improved" it and each pass made it a little worse. The fix was to freeze it, not to improve it: lock the good take and stop touching it.

The lesson is the song's own joke, pointed back at the process. Every gate said "done" before the listener did. Measure what the ear is judging.

The score

An original ChipForge score, every sound synthesised in code. 118 BPM, a fake product keynote in C major whose fake drop lands on a Game Boy blip and a slide whistle, then a tape stop and a turn to E flat for the anthem. The film opens cold on a flash-forward of chorus two that tape-stops into a single spotlight, because the first ten seconds decide everything.

The picture and the easter eggs

Stick figures on a keynote stage. The wall runs the slides, a karaoke screen, and a napkin with the whole company roadmap on it. Things to look for:

• The napkin roadmap, ending in "AGI (TBD)".
• The livestream holding slide, where the progress bar stalls at 99% while the crickets play. Is it done? It's done.
• The 03:14 clock.
• kubectl apply -f future.yaml.
• Plan 9 peeking in at NO SIGNAL.
• The goodbyes at the end, in Hebrew and Bavarian.
• A post-keynote satisfaction survey where every box gets ticked.

No pods were harmed in the making of this film.

The Shorts

• Everyone's a little scared. That's allowed. The Turn.
• It's the time of a lifetime. The anthem.
• Ten-x. Ten-x. The chorus.

Made on a laptop

No GPU, no stock footage, no render farm. The animation engine is Napkin Films, the music engine is ChipForge, both written in Python. At the very end, the agent asks: "Is it done now?" And the answer is the honest one. "Nearly."

More from Napkin Films

• Ollama Mama, Plan 9 and OG Bobby Johnson roast the frontier and tell you to own your model.
• Arena, the same two on a fighting-game stage, three rounds and a death god in the husk.
• The Grand Tour, where OG Bobby and his Anubis alter ego are revealed as one voice, two faces.

More in the Napkin Films catalog.

License

This film is licensed CC BY 4.0. Remix it, repost it, drop it into your own thing. Credit "Napkin Films / Organic Arts LLC" and link CC BY 4.0. Engine code (Napkin Films, ChipForge) is GPL-3.0-or-later. ElevenLabs voice audio is licensed content and is not redistributed. The score is an original ChipForge composition. No real companies or people are depicted or named.