Someone just put a $250 million price tag on the ability to search inside video, and honestly, that makes complete sense to me.
Clipto is the company. Its pitch, according to TechCrunch, is that it uses AI to search terabytes of video. That’s the whole thing. Not generating video, not editing video, not making talking avatars. Just finding the moment you’re looking for inside a pile of footage so large that no human could reasonably watch it all.
If your first reaction was “that’s it?”, stick with me, because I think this is one of the clearest examples of what AI agents are actually good for.
Why searching video is genuinely hard
Text search feels like a solved problem. You press Ctrl+F, you type a word, the browser highlights it. That works because text is already made of the thing you’re searching for. The word “invoice” is sitting right there in the document as the letters i-n-v-o-i-c-e.
Video is not like that. A video file is a very long sequence of images plus a sound wave. Nowhere in that file does it say “this is the part where the CEO announces the layoffs” or “this is the shot of the red car turning left.” That information exists only in your head after you watch it.
So historically, if you wanted to search video, a person had to watch it and write notes. Broadcasters and film studios employed people to do exactly this, tagging footage by hand so that someone could later find it. That works fine for hundreds of hours. It falls apart at terabytes.
What the AI is actually doing
The rough idea behind a system like this is translation. AI models can look at a frame of video and describe what’s in it. They can listen to audio and turn it into text. They can turn both of those descriptions into a mathematical form that captures meaning rather than exact wording.
That last part is the interesting bit for non-technical readers. Once meaning is stored mathematically, you can search for “someone looking nervous during a presentation” and get useful results even though nobody ever typed those words. The system isn’t matching your letters. It’s matching your intent against its understanding of what happened on screen.
Do this across an entire archive, and the archive stops being a wall of files. It becomes something you can ask questions of.
The pattern worth noticing
Look at the other AI news TechCrunch reported this week and a shape starts to appear.
- Google’s AI Mode can now track flight prices and help book hotels.
- Micro1, an AI data startup, hit a $500 million gross run rate as demand for AI training data climbed.
- Google is giving publishers a new way to fight traffic losses caused by AI.
- Relay, an AI automation startup, shut down, with its staff joining Google’s Chrome team.
Flight price tracking is a boring job humans are bad at doing consistently. Preparing training data is boring work at enormous scale. Searching an archive is boring work at enormous scale. The money is flowing toward AI that absorbs tedium, not AI that performs cleverness.
Relay’s outcome is the counterweight. “AI automation startup” is one of the most crowded descriptions in tech right now, and a broad promise to automate anything is harder to sell than a specific promise to find a clip. Its team landing at Google’s Chrome group is a reasonably soft ending, but the company itself is gone. Meanwhile the video search company gets a nine-figure valuation. Specificity is doing a lot of work in that comparison.
What this means if you’re not building any of this
Two things, I think.
First, expect the searchable-video idea to show up in tools you already use. Once one company proves people will pay for it, the feature tends to spread into meeting recorders, security camera apps, and phone photo libraries. The version you’ll eventually get will feel unremarkable, which is usually how you know a technology arrived.
Second, if you’re trying to figure out which AI companies are real, ask a simple question about each one: what specific job does this replace, and was that job previously done by a person watching, reading, or clicking through things one at a time? Clipto has a clean answer. Google’s flight tracking has a clean answer. The startups with fuzzier answers are the ones having a harder time.
Video search sounds like a small idea. It’s really a bet that the next wave of valuable AI won’t feel magical at all. It will just quietly handle the parts of the job nobody wanted.
🕒 Published: