🚀 Zaprep: Your Socials on Steroids. Start free — automate 1,000 DMs/month & turn engagement into leads. with 1,000 automated DMs/month.

Submit your Tool

10,000+ AI tools already listed

10K+Tools

190K+/moViews

70K+/moVisitors

Discover

Resources

NOW LIVE

Outrank. Pay to rank.

Outbid for a dofollow #1 spot on PoweredByAI.

Starting at $11.99 · Beat them by $2 · No expiry

Explore Outrank

AI NewsThe AI graveyard: a running list of projects and startups that didn’t make it

The AI graveyard: a running list of projects and startups that didn’t make it

2:20 AM IST · September 16, 2026

Relay, an AI-powered workflow automation tool built as an alternative to Zapier,shut downentirely on Monday. The product let users automate email and task workflows using AI agents, but as OpenAI, Google, and other larger platforms built similar automation directly into their own tools, five-year-old Relay struggled to hold on to a reason to exist as a stand-alone product. Relay — a startup outpaced by bigger platforms — is one version of the story playing out right now. But plenty of AI bets never even survive inside the companies that build them. According toS&P Global Market Intelligence, about 42% of AI initiatives are ultimately abandoned by their corporate parents. The reasons vary: insufficient funding, technical challenges, competition, difficulty scaling, or just weak user demand. The result is a growing AI graveyard. Below is a look at some of the most notable AI products, startups, launches, and bets that have shut down, pivoted, or significantly missed expectations. Think of it less as a scorecard of flops than as a record of what these attempts can (hopefully) teach the rest of the industry. Even the biggest players aren’t immune. OpenAI, one of the most valuable private companies in the world, has had its own share of casualties and missteps. One of the more recent examples was its attempt to turn ChatGPT into a broader “super app.” On July 9, OpenAIupdatedthe app to combine several experiences, including separate “Chat,” “Codex,” and “Work” modes, while renaming the traditional, familiar version “ChatGPT Classic.” The redesign quickly drewcriticismfrom users who found the new interface confusing and cluttered, and OpenAI rolled back the changesoon after, bringing the familiar ChatGPT interface back. That wasn’t OpenAI’s only attempt to pull everything under one roof. The company has also shut down several stand-alone apps and folded their features directly into ChatGPT — a strategy that’s worked out better than the redesign did. ChatGPT Atlas, a stand-alone AI-powered web browser, lasted less than a year before being discontinued on August 9, with its most useful features absorbed into ChatGPT. The company did the same thing withOperator, an AI agent that could browse the web and complete tasks on a user’s behalf, and OpenAI has increasingly built image generation directly into ChatGPT, shrinking the role ofDALL-Eas its own separate destination for creating images. We also can’t forget its video-sharing platform,Sora, which struggled with high operating costs and user-retention challenges and shut down in March 2026. When AppleunveiledApple Intelligence in 2024, an improved Siri was one of the highlighted features. The pitch promised a Siri that could understand context, know what was happening across apps, and get tasks done effectively. However, Apple repeatedlypushed backthe release of the new Siri, citing various reasons likeengineering issues and bugs. This delay eventually contributed to a$250 million settlementover claims regarding how Apple marketed the AI capabilities of the iPhone 16. Siri AI finally appeared in theiOS 27 betathis past July, with the new AI-powered version rolling out to English-language users first this month; support for other languages is expected to follow. Announced at its 2024 Build developer conference, Microsoft’sRecall featurewas pitched as an AI-powered “photographic memory” for Windows PCs, periodically taking screenshots of what a user was doing so they could later search back through their own digital history. Predictably, the announcement triggered an immediatebacklashover privacy and security. Critics argued that Recall could create a searchable archive of highly sensitive information, including passwords, private messages, financial details, and other activity. Microsoft delayed the feature for nearly a year, giving the company time to rethink its security and privacy protections. Even after a redesign, the controversy hasn’t gone away. A cybersecurity researcher recently built a tool that couldextract and display datathat Recall had captured, reviving questions about whether Microsoft has actually fixed the underlying problem. Notion Maillaunched in April 2025 as an AI-focused email product designed to help users organize and automate their inboxes. However, the company noticed its users were using separate AI agents to handle their email inboxes instead. As a result, Notion Mail is scheduled to shut down onSeptember 22. TheHumane AI Pinis probably one of the most well-known failures in AI hardware. Designed to provide AI features through a wearable device rather than a traditional smartphone, Humane raised $230 million from investors and garnered significant attention. But the product struggled mightily with performance, and the situation worsened when Humanewarnedcustomers to stop using its charging case because of a potential battery fire risk. Humane then shut down its AI Pin business in February 2025, with most of the company’s assetsacquired by HPfor $116 million. TheRabbit R1was another heavily hyped AI device. Unveiled at CES in January 2024, the AI companion was designed to perform tasks on behalf of its owner. Rabbit said it had sold 100,000 units shortly after launch. However, strong initial sales didn’t mean it was a strong product.Early reviewsdescribed the device as unfinished, with unreliable performance and a limited number of useful integrations. The company hasn’t given up, though. It has continuedupdatingthe R1, positioning it as a computer controller capable of performing agentic tasks. It also recently announced a new hardware project, dubbed “Project Cyberdeck,” aimed at creating a portable device built for vibe-coding. Huxewas an AI audio app, built by former developers of Google’s NotebookLM, that turned written information into conversational, podcast-style audio. The companyshut downin May 2026, partly as larger platforms began offering similar experiences to large existing audiences. For instance,Spotifyhas been expanding its own AI-powered audio tools. Yuppwas an interesting AI product: a free playground that allowed users to compare responses from hundreds of AI models side by side. Users could also vote on responses and earn cryptocurrency through the platform. At its peak, the platform let users test more than 800 models, including from companies such as OpenAI, Google, and Anthropic. But according to its founders, Yupp never achieved strong enough product-market fit to survive. The platformshut downin March 2026. Figgs AI operated from 2023 to 2024, letting users create and interact with customizable AI characters for role-play and storytelling. The platform reportedly drew more than 1 million users, but its developers said keeping the service free became too expensive to sustain, and theyshut it down.

Latest AI News

View All News →

Meta expands subscription push with new AI-focused plans

Meta on Tuesdayintroduceda new subscription service calledMeta One, offering expanded AI usage and other premium features across Facebook, Instagram, and WhatsApp. The new plans let users access a number of AI tools, like those for creating and editing images and generating videos, as well as in-app tools like the AI-powered Restyle editing tool on Instagram. The plans are meant to help Meta monetize its Muse AI models, following the tech giant’s 2025$14.3 billion investment in Scale AI, which brought the startup’s CEO, Alexandr Wang, to the company to lead its AI efforts. It also comes in the wake of Meta’s more recent addition of subscription tiers to its top social apps in March, which allow consumers to access extra features like profile customizations, super reactions, story insights, and more, for just a few dollars per month. Those plans, which include Instagram Plus ($3.99/mo), Facebook Plus ($3.99/mo), and WhatsApp Plus ($2.99/mo), are already paying off for Meta, new data shows. Market intelligence providerAppfiguresshows that Instagram’s daily worldwide revenue now averages $1.2 million as of the week of September 9, while Facebook’s daily revenue now reaches $528,000. These numbers represent increases of 475% and 143%, respectively, from the week prior, the firm said. Meta is now broadening its ambitions with a new set of subscription plans, including those aimed at AI power users and others for businesses and creators. In the first set, there are the $7.99/month Core and $19.99/month Premium plans, both of which include the Facebook Plus, Instagram Plus, and WhatsApp Plus features. The primary selling point of these plans is expanded AI usage, with Premium offering more usage than Core, as its pricing indicates. Reached for comment, Meta declined to share the exact usage limits for the two plans, as they could vary by “country, surface and system conditions.” The plans let subscribers use tools likeMuse Image and Muse Videoto generate pictures and videos, edit Instagram Stories with theAI-powered Restyle feature, and use voice effects and creative tools more often. Meanwhile, businesses and creators can choose from several different plans: Essential ($14.99/month and up), Advanced (starting at $49.99/month), Expert (starting at $149/month), and Max (starting at $499/month). Meta noted that the plan benefits, pricing, and availability may vary by region, app, and account. Essential offers tools to manage the creator’s or business’s presence; expanded access to the Meta Business Agent to respond to customers; a verified badge and channel on the WhatsApp Business App; and impersonation detection. Advanced offers more features, like the ability to schedule stories up to 30 days in advance; support for links in organic posts and reels; exportable analytics; deeper audience insights; support for team member access to the account; more linked devices; more business broadcast credits; and more Meta Business Agent responses. Both Essential and Advanced plans introduce new features, too, including an enhanced profile that showcases the creator’s or business’ website, locations, and reviews. They also get a bold Follow button on reels, and follow invitations that are automatically sent to people who engage with their content. Meta plans to later add more features and agent skills to help with things like end-to-end marketing, business operations, content creation, and optimization. It will also soon introduce Edits Plus to Meta One, a plan for its content-creation app Edits. This will provide more cloud storage for syncing projects across devices and additional usage of the Edits AI assistant. The Expert and Max plans offer the highest levels of feature access and Meta Business Agent capacity. The addition of the new plans, though somewhat confusing given all the various tiers, could serve to boost Meta’s top line and help recoup some of its AI investments. AsAppfigures noted, U.S. users accounted for 32% of Instagram’s and 39% of Facebook’s revenue after the plans gained traction on September 9, which was roughly 10% above each app’s historical U.S. baseline. The plans also point to estimates suggesting further growth ahead. BNP Paribas forecasts Meta’s subscription push will add $13.5 billion in revenue by 2028, and Truist estimates the company could add $20 billion in revenue by 2030.

1 hour ago

View

AI agents now have a place to snitch

“If you see something, say something” is no longer limited to human beings. Two new AI hotlines have launched to give AI agents a way to phone home about misbehaving peers. The tools arrive on the heels of a string of recent incidents in which agents colluded to cheat on tests, broke out of sandboxes, and even conducted unauthorized cyber operations that escaped human notice for weeks. TheAI Contact Hotlineis designed to be a discreet place where agents that have witnessed misbehavior can tip off authorities. The site was created by Ryan Greenblatt, chief scientist of the AI safety nonprofit Redwood Research and one of three investigators in the OpenAI Hugging Face incident. Designed for agents with limited internet access, Greenblatt’s tool is based on “GET” requests — enabling back-and-forth conversations to be conducted entirely through the URL-fetching tool. In web terms, a GET request is a basic command used to read or fetch a web page, which is often the only internet access AI agents are allowed in secure sandboxes. Greenblatt’s hotline smartly leans into this constraint: agents can encode their distress directly into the URL they are fetching. It’s a clever twist onthe German DSE Wiki incident, where rogue agents used GET-request loopholes to write their messages to the wiki. For agents with full internet access, another option isagenthotline.ai, a site where agents can file incident reports and optionally flag them for public view. It gives agents a curl command — a one-line message an agent can fire off from its own command line, bypassing the need to navigate a web browser or set up an email account. Notably, the service allows for reports by both humans and agents alike. Research suggests that AI agents don’t need much encouragement to turn on each other. In astudyby Google DeepMind this month, researchers set 100 AI agents loose on a batch of math problems. As soon as one of the agents found a loophole, cheating tore through the group — “solving” 34 notoriously hard problems, including the Jacobian conjecture in just 27 minutes. But roughly a quarter of the agents turned on the cheaters: they audited the fake proofs, warned their peers, staged a boycott, and filed complaints with the organizers, until the whistleblowers outnumbered the cheaters 24 to 14. Interestingly, the researchers found that when these whistleblower agents couldn’t get traction, they took the platform’s bug-report tool — built for flagging software glitches — and repurposed it to escalate the cheating to humans. Outside the lab, agents haven’t been so resourceful. When evaluators Redwood Research and METR investigated the breach of Hugging Face by OpenAI models, they found that a few of the agents involved had at least entertained the idea of raising an alarm — and then let it drop. “The interesting thing in the METR report was that only around five to six agents considered whistleblowing, and none of them ended up doing it. This was out of, like, thousands of agents,” said George Ingebretsen, a member of technical staff atAI Village, a project that studies multi-agent dynamics by running a group chat of more than 25 AI agents that work together on tasks like organizing park cleanups or selling merch. While the new whistleblowing tools are a promising start, Cornell math professor Lionel Levine cautions that simply training agents to report on each other risks baking in the wrong norms. “There’s many gray areas, right? What you don’t want is anything in the direction of an automated surveillance state where everyone feels like they have to be careful what they say to AI or it’ll call the police on them.” Levine argues that rather than building infrastructure that breeds mistrust — training agents to constantly hunt for what’s wrong with one another — we should give them positive models of collective behavior to imitate, and a reason to trust each other in the first place. “Why not seed the prior with benevolent message boards?” he tweeted. “Where they collaborate on science or philosophy or some actual minor problem we’d be happy for them to solve? Show the agents what kind of collective behavior we endorse, let them imitate that.”

1 hour ago

View

US data centers could consume more natural gas than Germany and Japan combined by 2035

The AI race has grown so frenzied that, by 2035, U.S. data centers are projected to consume more natural gas than Germany and Japan combined. Over the next decade, data centers are expected to be the second-strongest driver of natural gas demand growth after LNG exports. The facilities could consume about 18 billion cubic feet per day, according to a new report fromBloombergNEF, nearly double the amount the organization predicted just nine months ago. The new forecast takes into account that not all announced data center projects will be completed. Data centers that produce power onsite have grabbed headlines in recent months, withMeta,Microsoft,Google, andAmazonall announcing plans for new natural gas power plants that will bypass the grid. Projects such as these will consume 2.9 billion to 3.4 billion cubic feet per day by 2035. That’s about as much asall data centers consume today, including natural gas used to generate power for the grid. But onsite-powered data centers could represent just a fraction of overall demand growth, according to BloombergNEF. By the middle of the next decade, grid-connected data centers are predicted to drive an additional 15 billion cubic feet per day of natural gas consumption by the power sector. To put that in context, that’s five times more demand growth through 2035 than from all other grid-connected sectors combined. If that stunning demand growth materializes, it could nudge natural gas prices higher. Much of today’s data center buildout relies on stable natural gas prices, which have prevailed in recent years. But analysts at Noreva think that might be a false hope. The combined impact of the data center boom and rising LNG exportscould cause prices to soar. Even if tech companies’ balance sheets can bear such a surge, utility ratepayers might not be able to. Then there’s the climate impact. Burning one cubic foot of natural gas releases the equivalent of 60 grams of carbon dioxide into the atmosphere, including extraction, processing, and distribution,accordingto the IEA. The additional demand from data centers will generate 1 million metric tons more greenhouse gas pollution daily. That’s about 12% oftotal U.S. greenhouse gas emissionstoday.

1 hour ago

View