Every AI integration is a bet, not a foundation

I wrote recently about the AI tools that are winning, and what they have in common: they solve real problems inside real workflows.

Sora did not do that. And now it's dead.

OpenAI killed it yesterday. Six months after launch, $2.1M in lifetime revenue, downloads down 66%, and compute costs they couldn't justify ahead of an IPO. Disney's teams were in a working session with OpenAI thirty minutes before finding out the whole thing was gone. Billion-dollar deal, dissolved. One insider called it "a big rug-pull."

Disney will recover. A mid-size (or small, yikes!) brand that just spent a quarter building their content pipeline around Sora? Different story.

And it's not just Sora. OpenAI replaced Anthropic on a $200M Pentagon contract, and now federal agencies that built workflows around Claude are scrambling to rip it out. Not because the tech failed. Because of a political dispute.

This is the real AI risk nobody's talking about enough. Not "will AI take my job" but "what happens when the tool I built my workflow around disappears overnight?"

The companies that survive this aren't picking the shiniest tool. They're building vendor-agnostic workflows, keeping switching costs low, and treating every AI integration like what it is: a bet, not a foundation.

Compute is finite. Partnerships are fragile. Operational discipline isn't going anywhere.

Sources

• OpenAI Is Shutting Down Sora, Its A.I. Video Generator, The New York Times