Relay, an AI-powered workflow automation platform created as an alternative to Zapier, ceased operations completely in August. The service allowed users to automate email and task workflows using AI agents. However, as major platforms like OpenAI and Google integrated similar automation features directly into their existing tools, the five-year-old Relay struggled to maintain a viable niche as an independent offering.
More recently, the chatbot platform Character.ai declared bankruptcy just five years after its launch.
Beyond Character.ai and Relay, numerous AI ventures fail to thrive even within their parent companies. According to S&P Global Market Intelligence, approximately 42% of AI initiatives are eventually discontinued by their corporate sponsors. Reasons for these failures vary, including insufficient funding, technical hurdles, intense competition, difficulties in scaling, or simply inadequate user demand.
The cumulative effect is a growing collection of defunct AI projects. Below, we examine some of the most significant AI products, startups, launches, and strategic bets that have either shut down, undergone major pivots, or fallen short of expectations. This compilation serves not as a tally of failures, but rather as a record of lessons that (hopefully) the broader industry can learn from these attempts.
OpenAI Faces Several Challenges
Even the industry’s largest players are not immune to setbacks. OpenAI, currently one of the world’s most valuable private companies, has experienced its share of casualties and missteps. A recent instance was its endeavor to transform ChatGPT into a broader “super app.” On July 9, OpenAI updated the application to merge various experiences, including distinct “Chat,” “Codex,” and “Work” modes, while renaming the traditional, familiar version “ChatGPT Classic.”
This redesign quickly attracted criticism from users who found the new interface confusing and cluttered. Consequently, OpenAI reversed the change shortly thereafter, restoring the familiar ChatGPT interface.
This was not OpenAI’s only attempt to centralize features. The company has also discontinued several standalone applications, integrating their functionalities directly into ChatGPT—a strategy that has proven more successful than the aforementioned redesign. ChatGPT Atlas, a standalone AI-powered web browser, operated for less than a year before being decommissioned on August 9, with its most valuable features incorporated into ChatGPT.
The company applied the same approach to Operator, an AI agent capable of web browsing and task completion on behalf of users. Furthermore, OpenAI has progressively integrated image generation directly into ChatGPT, diminishing the role of DALL-E as a separate platform for image creation.
We also cannot overlook its video-sharing platform, Sora, which faced high operational costs and user retention issues, leading to its closure in March 2026.
Apple’s Siri AI Evolution
When Apple introduced Apple Intelligence in 2024, an enhanced Siri was among its highlighted features. The promise was a Siri capable of understanding context, recognizing activities across applications, and efficiently completing tasks.
However, Apple repeatedly postponed the release of the new Siri, citing various reasons such as engineering challenges and bugs. This delay ultimately led to a $250 million settlement stemming from claims regarding Apple’s marketing of the iPhone 16’s AI capabilities.
Siri AI eventually debuted in the iOS 27 beta last July, with the updated AI-powered version rolling out to English-language users first this month. Support for other languages is anticipated to follow.
Microsoft’s Recall Feature Under Scrutiny
Unveiled at its 2024 Build developer conference, Microsoft’s Recall feature was marketed as an AI-driven “photographic memory” for Windows PCs. It was designed to periodically capture screenshots of user activity, allowing users to later search through their digital history.
Predictably, the announcement provoked an immediate backlash concerning privacy and security. Critics argued that Recall could create an indexable archive of highly sensitive information, including passwords, private communications, financial data, and other activities. Microsoft postponed the feature’s release for nearly a year, using the time to re-evaluate its security and privacy measures.
Even after a redesign, the controversy persists. A cybersecurity researcher recently developed a tool capable of extracting and displaying data captured by Recall, reigniting questions about whether Microsoft has truly resolved the underlying issues.
Notion Mail Discontinued
Notion Mail launched in April 2025 as an AI-focused email solution aimed at helping users manage and automate their inboxes. However, the company observed that its users were opting for separate AI agents to manage their emails instead. As a result, Notion Mail is slated to cease operations on September 22.
The Humane AI Pin’s Downfall
The Humane AI Pin is arguably one of the most prominent failures in AI hardware. Designed to deliver AI functionalities through a wearable device rather than a conventional smartphone, Humane secured $230 million from investors and attracted considerable attention.
However, the product encountered severe performance issues, and the situation worsened when Humane advised customers to discontinue using its charging case due to a potential battery fire hazard. Humane subsequently shut down its AI Pin division in February 2025, with the majority of the company’s assets acquired by HP for $116 million.
Rabbit R1: A Hyped Device’s Mixed Reception
The Rabbit R1 was another highly anticipated AI gadget. Unveiled at CES in January 2024, this AI companion was designed to execute tasks on behalf of its owner. Rabbit reported selling 100,000 units shortly after its launch.
However, robust initial sales did not translate to a robust product. Early reviews characterized the device as underdeveloped, exhibiting unreliable performance and a limited array of useful integrations.
Despite these challenges, the company has not given up. It has continued updating the R1, repositioning it as a computer controller capable of performing agentic tasks. Rabbit also recently announced a new hardware project, named “Project Cyberdeck,” which aims to create a portable device tailored for