Why AI Can't Replace Human Curation (Yet)
One of the joys of this AI era is experimenting with things that would otherwise be completely out of reach.
A few months ago, we used Claude to experiment with YOLO — an object detection model. The process involved installing elements in the terminal and changing code. Two years ago, this felt impossible for a non-technical team.
It worked. And it didn't.
The Logo Detection Dream
We had this idea: use computer vision to automatically identify and count logos in video content. We could go to brands with their advertising assets and ask:
"What is the ratio of euros per logo appearance in your campaigns?"
And tell them: "The benchmark is here."
It's an old advertising truth — brands always want their logo bigger, while agencies and directors work to minimize it.
Computer vision promised to quantify this tension. Automatically detect every brand mention, every product appearance, every logo flash.
The technology exists. Roboflow's latest version can find objects just by typing words — no labeling required.
So why aren't we using it?
Three Problems AI Can't Solve
1. Context
Video — especially creative video — is full of choices that carry cultural meaning. References, subtext, shared understanding. A director puts a specific product in a specific scene for reasons that go beyond "object detected."
Computer vision sees a coffee cup. It doesn't know the cup signals the character's social class, or references a famous film, or carries ironic weight.
Context is what makes content interesting. AI can detect objects. It can't detect meaning.
2. Hierarchy
Even when AI correctly identifies elements, it tends to "cloud" results by treating every detection equally. The hero product gets the same weight as background clutter.
The question isn't just "what's in the frame?" It's "what matters in the frame?"
Answering that requires judgment. Which detections are significant? Which are noise? These are editorial decisions, not detection problems.
3. Taste
Here's the real issue: by the power of human editorial capacity, what one person suggests can be more enlightening, original, and personal than what an AI suggests.
When you watch a video and notice something worth sharing — a detail, a product, a reference — that noticing is valuable. It reflects your perspective, your knowledge, your curation.
AI can tell you what's there. It can't tell you what's interesting.
Why This Matters for Vidlink
At Vidlink, we thought hard about whether to use AI for suggesting what cards to add to videos.
We decided against it. At least for now.
Not because the technology doesn't work. It does, within limits. But because human curation produces better results.
When you add a card to your video at a specific moment — "this is the jacket I'm wearing" or "this is the tool I'm using" — you're making an editorial choice. You're saying: this matters, at this moment, for this reason.
That judgment is what makes interactive content valuable. An AI suggesting random product detections would be noise. A creator choosing what to highlight is signal.
The Human Advantage
We're aware that not including AI in our selling point might seem risky in 2026. Everyone wants AI features. AI is the magic word.
But we believe human intelligence is more relevant, brighter, and more engaging than automated curation. At least at this stage.
At the end of the day, we want to be entertained. We want to discover things we didn't know we wanted. We want someone with taste to point us in the right direction.
Humans are still better at this than computers.
Where AI Helps
We do use AI — just not for the core curation task.
AI helps us understand what elements might be interesting in a video. It helps with suggestions and background analysis. It handles technical tasks that don't require taste.
But the decision about what to make clickable, when to make it clickable, and why it matters — that's human work.
The creator knows their content. They know their audience. They know what's worth highlighting.
Our job is to make it easy for them to act on that knowledge.