Cognition Bought Poke and Midjourney Bought Co-Star Because AI Labs Are Done Building Audiences From Scratch
I used to think the labs with the best models would win the consumer layer eventually, on the theory that a better assistant is a better assistant regardless of which app it lives inside. Distribution felt like a problem money could solve whenever a lab got serious about it.
Money doesn’t solve it fast enough anymore, apparently.
Cognition acquired Poke, the messaging-based AI assistant, in a deal reportedly valuing Poke’s parent in the low nine figures. Midjourney, it turns out, bought the astrology app Co-Star back in the spring and is now building its own standalone image app on top of that acquisition. Meta launched Seller, a standalone Marketplace app with AI listing tools, and Facebook is testing a full-screen immersive video player to replace the newsfeed entirely. Add Prentis — the Reid Hoffman and Marc Pincus computer-use startup reportedly in talks at a $1 billion valuation before it has anything like consumer scale — and the pattern isn’t “labs building better models.” It’s labs and platforms paying for a habit they didn’t build.
What Did Midjourney Actually Buy When It Bought Co-Star?
Ask what Co-Star is worth to an image-generation company with no prior astrology product. Not the astrology content. Co-Star’s actual asset is a daily-open habit — millions of users checking a personalized notification every morning, a behavior pattern that took Co-Star years of product iteration to earn and that Midjourney could not have replicated by getting marginally better at generating images. Midjourney didn’t buy a feature. It bought the muscle memory of opening an app every day, and it’s now pointing that muscle at image generation instead of horoscopes.
The Acquisition Price Is the Real Signal
Poke’s parent selling for “low nine figures” tells you what a working messaging-native assistant with real usage is worth to a company like Cognition that has excellent coding models and no consumer surface of its own. That number is the market’s honest price for distribution right now, and it’s a number every lab building a general assistant should be watching more closely than the next benchmark release.
Where This Logic Runs Out
Not every acquisition in this pattern will work. Bolting a coding-focused lab’s ambitions onto a messaging assistant, or an image generator onto an astrology app, assumes the acquired audience will tolerate a hard pivot in what the product is actually for — and audiences built around one specific daily ritual don’t always follow the product wherever it goes next.
But the labs making this bet have decided that risk is smaller than the risk of building an audience from zero. I think they’re right.