I wanted a 15-second video of a Notion community event.
Some fast cuts. A few animated words. A short-vlog kind of feel.
By the end of the conversation, I was explaining my music taste to an AI and asking it to stop being quite so polite with the soundtrack.
Sir Work-alot has a moustache. Apparently, it now also needs a pair of headphones.
It started with five clips and a simple request
After Notion Build Bar Singapore, I uploaded five short clips and asked my Notion AI agent to turn them into a quick snapshot of the event.
I did not have a huge amount of footage, so I asked it to play around with text animations too. It made a vertical video with quick cuts, punch-in crops and captions, including a little “ASK. BUILD. SNACK.” sequence.
The snack part was important, obviously.
Visually, we had something to work with. The audio was another matter.
First version: you could definitely hear the venue
The first edit kept the original sound. Which sounds reasonable until you remember that an event venue contains other people, conversations and quite a lot of noise.
My feedback was fairly direct:
“Definitely not use the venue sound cause they’re noisy, please add music that fits the vibes.”
That was the first correction. Remove the venue audio completely. Give the video its own soundtrack.
“Music that fits the vibes” was doing a lot of work
The agent created an original instrumental, then we kept refining it.
I also remembered another clip I had forgotten to upload: a longer video of our catch-up afterwards. We compressed that into a roughly three-second, 12×-speed ending and kept the whole recap at 15 seconds.
So now we had the event, the snacks and the post-event conversation.
But I was still working out what I wanted the music to sound like.
I shared Awesome Music Programming, a GitHub collection of tools for synthesis, composition and audio processing, and asked whether we could use it to make something better.
The next direction used guitar-like plucks, warm keys, bass and softer drums.
The warm-groove version
It was a different direction. But the useful detail I had not clearly given the agent yet was this:
I’m more of an EDM and electronic music guy.
Including synthwave sounds. Synth chords, arpeggios, electronic bass. That sort of thing.
“Better music” had been my request. “Music closer to what I actually enjoy” was what I meant.
Those are not quite the same brief.
Right, let’s give it some synths
Once I explained that preference, the conversation became much more specific.
We looked at synths such as Vital, Serum 2 and TAL-U-NO-LX. Those were instrument recommendations, not plugins the agent had secretly installed and started using.
That distinction matters.
Package downloads were blocked in the working environment. The finished test used custom code-based synthesis, not an actual Vital or Serum render.
The GitHub list helped us explore the tools and approaches. It was not a folder of ready-made music, and linking it did not magically turn my agent into a fully equipped recording studio.
Still, we could change the musical direction.
The next version used detuned synth chords, an arpeggio, electronic drums, a bass line and a short melodic hook. It also had some rhythmic variation and a closing chord, rather than simply stopping a loop when the video ran out.
The EDM / synthwave version
This was the one.
My response was:
“Hahaha very goood! i love this.”
Not a particularly technical review. Quite a clear one, though.
The warm-groove and electronic versions use the same visual edit, including the fast catch-up clip at the end. Play those two back to back and listen to how differently they frame the footage.
That was the fun part for me. We were not adding more video. We were changing how the same video felt.
So I asked it to remember the process
I did not want the next recap to start with me explaining my music taste all over again.
We turned the workflow into a reusable Notion skill: Create Electronic Music for Videos.
The useful bits were not just “make EDM”. They included:
Start from my electronic-music preference, rather than defaulting to acoustic or lounge-style sounds.
Keep the approved video timing and visuals intact when only changing the music.
Remove noisy venue audio unless I specifically want it.
Give a short track some shape, including variation and an intentional ending.
Be honest about which instruments and tools were actually used.
Keep the previous version so I can compare.
Then we tested the skill on the Build Bar video. That test produced the electronic version above.
This was not me training a new AI model. It was me giving my existing agent clearer preferences and a repeatable way to approach the work.
Less mysterious. Quite useful.
My ears were still part of the workflow
One limitation was worth keeping visible: the agent could check the exported file, its duration and its audio levels, but full audio playback was not available in its environment.
It could not honestly say it had listened to the finished track and decided it was good.
That part was mine.
The export checks mattered. So did opening the video and deciding whether I actually liked what came out of the speakers.
For this one, I did.
The brief got better before the music did
Looking back, the biggest change was not finding one clever tool.
I started with “add music that fits the vibes”. We eventually got to a much clearer direction: electronic, synthwave-influenced, more synth-led, no venue noise, and an ending that belongs to the edit.
The iterations gave me something concrete to react to. The skill gave us somewhere to keep what we learnt.
I would not call this proof that my AI is now a music producer. It is proof that I can work through a small creative project with it, hear what is not quite right, and steer it towards something I enjoy.
A look inside the instructions behind my personal Notion agent, why they matter, and where I draw the line between useful assistance and handing over judgement.
How one photo and a chat with Notion AI turned into a print-ready 3D stand for my Maono AME2, from first sketch to documented project in a single conversation.
How I turned a broken Beeper extension into a local, cyberpunk command center that unifies all my chats — the decisions, the architecture, and the bugs that taught me the API.