When the Toolmaker Didn't Know the Work on the Floor
When building our video search product atoindex, a client asked to "search material by technique," and I couldn't grasp what they meant. Around the same time, a film director in Singapore advised me to hire an industry veteran. If that was the answer, I could become a maker myself. Five months into running a YouTube channel, I finally understand, through my own struggles, what that meeting was about.
We are building a video analysis and search product called atoindex. It analyzes video with AI and makes the contents searchable by meaning (I wrote about the thinking behind it in “Your Most Underused Asset Might Be Your Video”).
This article is about how I realized that as someone building a tool, I didn’t actually know the work the tool was for — and what happened when I decided to cross over to the maker’s side myself.
The Meeting I Couldn’t Understand
When we first developed the atoindex video analysis pipeline, I tried cold outreach through website contact forms as an initial test, sending around 200 messages. One person who responded ran a production company that created YouTube videos for clients, and we set up a meeting.
They shared several needs from their production floor. The one that stuck with me most was their desire to “search material by technique.” They wanted to pull up assets based on editing styles, effects, animation techniques, or camera movements.
Honestly, having never produced video myself, I couldn’t truly grasp that need. While I could build a tool that “analyzes” video, I had no visceral understanding of what assets get created during actual video production, through what steps, or how they might be reused months later. While I could answer technical questions about our system, I couldn’t hold a conversation at the same depth once the topic turned to day-to-day production work.
”Without Common Words, Negotiation Is Hard”
Around the same time, I was reaching out via LinkedIn to video news organizations and production companies overseas. One of the few people who replied was a documentary director in Singapore. He had spent years directing documentaries at a broadcast network before going freelance. When he visited Japan, we went out for drinks and became friends.
I still remember his advice word for word:
Taka, if you can, hire someone who has worked in film or video. Otherwise, when you sit down with industry people, you won’t share the same language, and negotiation will be very difficult.
If you want to build tools for the video industry, bring someone onto the team who speaks its language. It was completely sound advice. Yet at our stage, hiring an industry veteran simply wasn’t realistic. Still, those words stayed with me as a constant reminder: don’t stop at merely being a tool developer.
Becoming a Maker Myself
Some time later, I came across a post on X showcasing video generation using Kling AI. It built compelling scenes starting from just a few reference images. It looked fascinating, so I tried mimicking the workflow.
Over the next few weeks, I thought about what kind of compelling project I could build with this technology. Stripping away characters, dialogue, and plot, I arrived at historical scenes where it simply rains — leading to the YouTube channel Sounds of Rain in Old Japan (I wrote about that motivation in another post).
For five months, I managed this channel entirely by hand, from initial concept to publishing. Along the way, I built an automated agent pipeline so that sending parameters from WhatsApp generates a complete video (the technical side is documented here). But what I want to share here is not the technology, but what became visible only after becoming a creator myself.
Becoming a Maker: The Meeting Finally Made Sense
Running the channel forced me to confront the exact realities I couldn’t picture during that initial meeting — this time, as my own everyday pain.
The “enormous volume of assets” turned out to be genuinely overwhelming. Behind a single 15-second Short lay three generated still images (with only one chosen), several failed video takes, multiple candidate rain audio tracks, and generations of prompt iterations. With 100 scenes produced, the raw material easily numbered in the thousands. Months later, when I thought, “I want that lighting from the pleasure house scene again,” my only clues were timestamps, file names, and fragmented prompt strings.
I also finally understood what “searching by technique” really meant. I found myself desperately wanting to pull up material by criteria like “Kurosawa-style contrast,” “locked-off fixed camera,” or “seamless rain loops.” Tags like era or location weren’t enough. When creators recall assets, their mental index is organized by technique, not just subject matter.
Furthermore, generative video production churns out a massive number of “failed takes” that cannot simply be discarded — because the way a clip fails dictates how the next prompt must be adjusted. Assets carry the context of which stage of the process produced them. This kind of tactile awareness was something I could never have grasped merely by listening to a client describe it.
The Distance Between Toolmakers and the Floor
If the person building a tool has never done the actual work that the tool serves, communicating with customers at the same depth is virtually impossible. You can write down feature requests, but you cannot imagine the unspoken assumptions or the underlying workflow structure behind a request like “search by technique.”
The Singaporean director’s advice was to bridge that gap by hiring someone. But rather than waiting until we could afford to hire, stepping into the industry and becoming a maker myself was infinitely faster.
Using Our Own Tool on Our Own Floor
Which brings me back to that initial meeting with the production company. I have started an experiment: letting our own product, atoindex, manage the thousands of assets accumulated across Sounds of Rain.
Because a rain video doesn’t change scenes internally, atoindex’s primary strength — automated scene detection — doesn’t directly shine here. Even so, as someone now genuinely struggling to organize assets, I can use the tool from a creator’s perspective. How can the tool answer the very frustrations I once failed to understand? Once results emerge, I will share them in another post.
If you build tools, trying out the work your tool is made for might completely change how you see your product. If you are struggling with managing or searching video assets on your production floor, feel free to reach out to us.