With Sora’s death, AI’s age of frivolity may be ending

Hello again, and welcome back to Fast Company’s Plugged In. Before we get underway, a little self-promotion: Apple’s 50th anniversary is on April 1. As the big day approached, I realized that many people present at the company’s creation were still very much with us. So I interviewed 23 of them for an oral history, “How Apple Became Apple: The Definitive Oral History of its Earliest Years.” It’s chock-full of great tales as told by everyone from cofounder Steve Wozniak to Liza Loop, the first Apple user. Hearing these pioneers reminisce, I felt like I had been there, too—and so will you, I think. Here’s the article.

When OpenAI launched its Sora app last September, the video-centric social network arrived on a tide of buzzy goodwill. Its feed of 10-second video clips had a TikTok-esque vibe—except that it was filled with AI-generated stuff instead of anything remotely real. In less than a minute, Sora users could create digital doppelgängers of themselves that were eerily convincing for use in their own clips and, optionally, those created by others. The result was playful, goofy fun, and far more intriguing than Meta’s theoretically similar but painfully bland Vibes.

But if Sora ends up being remembered for anything, it won’t be for existing. Instead, it will have made its mark by going away.

On March 25, OpenAI announced that it was killing the app, along with the Sora API that let developers generate their own videos using the company’s technology. The decision appeared hasty: OpenAI still hasn’t shared details on when, exactly, Sora will cease to exist, or how users can download their videos for preservation.

Most of the insta-reaction I’ve seen to Sora’s demise amounts to grave tap-dancing of one sort or another. People are helpfully explaining that the app was a stupid idea from the start, or assailing it as a slop machine that deserved its fate. But I’m not ashamed to admit that I will miss it. For reasons I wrote about shortly after its debut, escaping to Sora’s weird little world always brightened my day.

For one thing, I found the app to be a genuine canvas for creativity, albeit in brief, inherently inconsequential bursts. My feed was full of fake commercials, fabricated vintage news clips, and other snippets of fantasy content that were like glimpses of bizarre alternate realities. An oddball crew of deceased celebrities—Larry King, Richard Nixon, Queen Elizabeth II—often starred in them, sometimes in uncannily convincing form and sometimes as vague approximations. On an internet that can feel unrelentingly grim, Sora’s essential absurdity made me laugh.

Counterintuitively, I also found comfort in the fact that the app was all AI, all the time. Conventional social media such as Facebook, Instagram, and TikTok is now befouled by true AI slop, generated solely to try and attract eyeballs without working very hard. Being exposed to it always feels like an imposition. On Sora, however, I never had to wonder if something was real or not. It wasn’t, and that was the point.

I do acknowledge that the app peaked early. The world needs only so many silly imaginary gadget commercials and clips of unlikely celebrities rapping—both my feed and my own ideas for prompts grew repetitive over time. If OpenAI had added more features, or let us create videos longer than 10 seconds, it might have helped the platform develop more substance. Now we’ll never know.

Still, I’m not going to make the case that OpenAI’s seemingly abrupt decision to shutter Sora is a terrible mistake. It might actually be a commendable, responsible act—or even the beginning of a trend for the entire AI industry.

On March 16, before that move was public, The Wall Street Journal’s Berber Jin reported that Fidji Simo, OpenAI’s CEO of applications, had sent a memo to employees declaring that it was time to get down to business. She sent it at a time when archrival Anthropic had made enormous inroads with its Claude Code software-generation tool, the hottest product in AI’s hottest category.

“We cannot miss this moment because we are distracted by side quests,” Simo wrote. “We really have to nail productivity in general and particularly productivity on the business front.” One element of this strategy recalibration involves OpenAI releasing a “super app” that rolls ChatGPT, Codex, and the Atlas web browser into one piece of software, roughly akin to what Anthropic has already done with the Claude desktop app.

As excerpted by Jin, Simo’s memo did not name-check Sora. In retrospect, though, her call to action left it a dead app walking. Rather than facilitating productivity, Sora was frivolous to its core. I certainly got sucked into it any number of times when I had better things to do.

But OpenAI isn’t terminating Sora because it might divert users from more productive tasks. It’s doing it because it’s a pricey distraction for the company itself. That OpenAI is suddenly interested in self-discipline is news in itself.

Until now, after all, its strategy has seemingly been to do, well, everything. ChatGPT in its current form is just the beginning. The company is also into enterprise agents! Health advice! Epoch-shifting gadgets! Browsers! Chips! Smut, though it’s been delayed! That’s before you get to the unprecedented investment in data center infrastructure it will have to build out to generate all that AI.

Maybe a huge, wildly profitable company could reasonably attempt to digest such a sprawling menu of projects simultaneously. OpenAI is not that company.

Like much of the AI in our lives, Sora has been running on a gigantic subsidy provided by venture-capital dollars. In November, Forbes’s Phoebe Liu guesstimated that OpenAI might be spending $15 million per day spitting out Sora videos. No analysis performed by an outsider stands a chance of nailing the precise cost, but this we know: Video generation is among the most computationally expensive AI tasks, and OpenAI had yet to book its first nickel of Sora user revenue. (It had inked a “landmark agreement” with Disney to use that company’s characters inside Sora, but that $1 billion deal is now off.)

If Sora stood a chance of being a profit machine someday, absorbing its current losses—which, over the course of a year, would have likely been in the billions—might not have been wholly irrational. But substantial profit would have come only if the app’s user base had grown gigantic and OpenAI figured out a brilliant way to weave ads into the experience. Though not impossible, that feat would have required vast intellectual capital and tolerance for risk.

By comparison, OpenAI focusing on ensuring that its Codex AI software-generation tool is a compelling alternative to Claude Code—one companies are happy to pay for—sounds dead easy. Who can blame Simo for opting not to pursue “side quests” when it’s imperative to get the core ones right?

As rational as OpenAI giving up on Sora may be, I hope that it doesn’t represent an end to the theory that a rewarding social network might someday be built around AI-fueled content. The evidence that AI can make social experiences much, much worse is all around us. Given that the technology isn’t going anywhere, I choose to cling to the possibility that someone will figure out how to adopt it in a constructive manner. Maybe even one that won’t bankrupt the company that offers it.

You’ve been reading Plugged In, Fast Company’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on fastcompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky, Mastodon, and Threads, and you can follow Plugged In on Flipboard.

More top tech stories from Fast Company

This Microsoft security team stress-tests AI for its worst-case scenarios
The company’s Red Team simulates attacks to uncover risks before bad actors do. Read More →

A top AI researcher explains the limitations of current modelsFrançois Chollet talks about his deceptively simple new benchmark test for AI models. Read More →

Manus AI cleaned up my computer—for a price
The desktop app can automate all kinds of tedious computing tasks, but the costs can quickly get out of hand. Read More →  

Exclusive: This new benchmark could expose AI’s biggest weakness
ARC-AGI-3 tests whether models can reason through novel problems, not just recall patterns, a task even top systems still struggle to do. Read More → 

Writer wants to be the go-to AI tool kit for the enterprise With customizable ‘skills’ and step-by-step ‘playbooks,’ the company aims to help employees automate workflows without touching code. Read More → 

This brilliant browser tool purposely makes AI chatbots worseThe extension’s designer calls it a ‘tiny tool of digital sabotage.’ Read More →

View the full article