AI products are not only failing in public. Increasingly, they are disappearing into the platforms that made them unnecessary.
A new TechCrunch report catalogs the growing list of AI products and startups that have shut down, pivoted, been absorbed, or failed to meet expectations. The examples range from Relay, an AI workflow tool that closed after larger platforms built similar automation into their own products, to Character.ai, which filed for bankruptcy only five years after launching.
The standalone AI product is becoming harder to defend
The pattern is bigger than a few weak launches. OpenAI reportedly rolled back a ChatGPT redesign that combined Chat, Codex, and Work modes after users found the interface confusing. The company has also folded the functions of Operator and image-generation tools more deeply into ChatGPT, reducing the need for separate destinations. Sora, its video-sharing platform, shut down in March after struggling with operating costs and retention.
The same pressure appears outside software. Humane’s AI Pin business ended after performance problems and a battery-fire warning involving its charging case. Rabbit’s R1 generated strong initial interest, with the company saying it sold 100,000 units, but early reviews described an unfinished product with unreliable performance and limited integrations. Notion Mail, launched as an AI-first email product, is scheduled to shut down after users increasingly relied on separate agents to manage their inboxes.
AI is being judged by distribution, not novelty
The common thread is not simply that the products were imperfect. It is that their most useful ideas could be copied, absorbed, or offered more conveniently by companies with a larger installed base.
That creates a difficult standard for AI startups and product teams. A feature can attract attention without earning a durable place in a user’s routine. A device can sell quickly without proving that it deserves to exist beside the smartphone. Even a major company can discover that a broader AI interface creates more friction than value, as Apple’s repeatedly delayed Siri and Microsoft’s still-contested Recall feature demonstrate.
For brands and marketers, the lesson is less about avoiding experimentation than recognizing where the experiment lives. AI features attached to an existing product may survive because they benefit from distribution. Standalone products need a stronger reason to remain separate. The AI graveyard is filling up with ideas that were interesting enough to launch, but not distinct enough to last.