<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>AI Graveyard</title>
    <link>https://www.aigraveyard.org</link>
    <description>A curated archive of AI products that shut down. Dates, funding, founders, and what killed each one.</description>
    <language>en-us</language>
    <lastBuildDate>Tue, 30 Jun 2026 01:43:45 GMT</lastBuildDate>
    <atom:link href="https://www.aigraveyard.org/feed.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>DALL-E 2</title>
      <link>https://www.aigraveyard.org/p/openai-dall-e-2/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/openai-dall-e-2/</guid>
      <pubDate>Tue, 12 May 2026 00:00:00 GMT</pubDate>
      <description>OpenAI&apos;s second-generation text-to-image model that opened consumer AI image generation to a mass audience.</description>
      <content:encoded><![CDATA[<p>DALL-E 2 launched in April 2022 as an invite-only beta, opened to the public that September, and became the first image generator most people ever used. By 2023 it had been displaced inside OpenAI's own product by DALL-E 3, and by 2025 by the gpt-image family. OpenAI sent a deprecation notice on November 14, 2025, and removed DALL-E 2 and DALL-E 3 from the API on May 12, 2026, pointing developers to gpt-image-1 and gpt-image-1-mini. The model that proved a market existed was retired three years after creating it.</p>]]></content:encoded>
    </item>
    <item>
      <title>OpenAI Sora (v1)</title>
      <link>https://www.aigraveyard.org/p/openai-sora-v1/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/openai-sora-v1/</guid>
      <pubDate>Sun, 26 Apr 2026 00:00:00 GMT</pubDate>
      <description>OpenAI&apos;s standalone text-to-video app and web product, killed less than five months after its public launch.</description>
      <content:encoded><![CDATA[<p>OpenAI previewed Sora in February 2024 with cinematic demo clips that briefly defined the state of the art for AI video. The consumer product opened to ChatGPT Plus and Pro subscribers on December 9, 2024, then launched a dedicated iOS and Android app and the sora.com site. Usage peaked near one million users before falling below 500,000, while OpenAI was reportedly spending close to a million dollars a day on inference for the model. On March 24, 2026, OpenAI told users it was exiting standalone video generation; the sora.com site and the mobile apps were shut on April 26, 2026, with the API set to follow on September 24. A pending content deal with Disney was abandoned as part of the wind-down.</p>]]></content:encoded>
    </item>
    <item>
      <title>Hiro</title>
      <link>https://www.aigraveyard.org/p/hiro-finance/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/hiro-finance/</guid>
      <pubDate>Mon, 20 Apr 2026 00:00:00 GMT</pubDate>
      <description>A &quot;personal AI CFO&quot; that modeled income, debt, and expenses, acqui-hired into OpenAI five months after launch.</description>
      <content:encoded><![CDATA[<p>Hiro was the latest venture from Ethan Bloch, who had previously built the savings app Digit and sold it to Oportun for more than 200 million dollars. Founded in 2023 and backed by Ribbit Capital, General Catalyst, and Restive, it pitched a personal AI CFO that could model a user's income, debts, and expenses with arithmetic accuracy that general-purpose chatbots lacked. The financial-planning tool had been live for only about five months when OpenAI acquired the roughly ten-person team in April 2026. It was a talent deal, not a product one: Hiro told users it would shut down on April 20, 2026 and delete their data by May 13. The math-verification work that made Hiro attractive headed into ChatGPT, while the app itself went dark.</p>]]></content:encoded>
    </item>
    <item>
      <title>Yupp</title>
      <link>https://www.aigraveyard.org/p/yupp-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/yupp-ai/</guid>
      <pubDate>Wed, 15 Apr 2026 00:00:00 GMT</pubDate>
      <description>A free, crypto-incentivized playground for chatting with 800-plus AI models that paid users for their preference votes.</description>
      <content:encoded><![CDATA[<p>Yupp launched out of stealth in June 2024 with a 33 million dollar seed round led by a16z crypto's Chris Dixon, and a pitch built for the moment: let anyone chat with more than 800 AI models for free, and pay users in points for the preference data their votes generated. Founders Pankaj Gupta and Gilad Mishne, both veterans of Twitter and Google, signed up 1.3 million users and collected millions of ratings a month. It was not enough. The company said it never found strong product-market fit, and that the industry's swing toward autonomous agents drained demand for the broad, crowdsourced human feedback Yupp was built to harvest. It told users on March 31, 2026 that it was winding down, and cut off platform access after April 15.</p>]]></content:encoded>
    </item>
    <item>
      <title>Snazzy AI / Smart Copy</title>
      <link>https://www.aigraveyard.org/p/snazzy-ai-smart-copy/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/snazzy-ai-smart-copy/</guid>
      <pubDate>Wed, 28 Jan 2026 00:00:00 GMT</pubDate>
      <description>An early GPT-3 copywriting tool that Unbounce bought, renamed Smart Copy, and later switched off as a standalone product.</description>
      <content:encoded><![CDATA[<p>Snazzy AI was one of the first consumer copywriting tools built on GPT-3, launched in late 2020 by Chris Frantz and a cofounder and bootstrapped to roughly 30,000 users without an outside round. In May 2021 the landing-page company Unbounce acquired it, kept it running, and eventually rebranded it Smart Copy. The standalone product survived longer than most acquired tools, but on January 28, 2026 Unbounce discontinued it, shutting off the Smart Copy site and giving users until late March to export their work. AI copywriting lived on inside the Unbounce builder; the standalone descendant of Snazzy did not. It was an unusually long runway for an acquired AI product, and it still ended at a redirect.</p>]]></content:encoded>
    </item>
    <item>
      <title>Phind</title>
      <link>https://www.aigraveyard.org/p/phind/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/phind/</guid>
      <pubDate>Fri, 16 Jan 2026 00:00:00 GMT</pubDate>
      <description>An AI answer engine built for developers that resolved coding questions with cited sources, later running its own large models.</description>
      <content:encoded><![CDATA[<p>Phind launched out of Y Combinator's Summer 2022 batch, founded by Michael Royzen and Justin Wei, as an answer engine for developers that pulled from documentation and code to resolve programming questions with citations, eventually training its own models. For a while it was a favorite among engineers who wanted answers without the forum archaeology. Its problem was structural: once ChatGPT, Perplexity, Google, and Claude all added capable web search, a dedicated coding answer engine had little left to defend, and traffic fell sharply. In December 2025 Phind raised about 10 million dollars led by Bessemer, and roughly six weeks later, on January 16, 2026, Royzen announced the shutdown, with user data deleted by January 30. Few deaths in the archive arrive quite so soon after a fresh check.</p>]]></content:encoded>
    </item>
    <item>
      <title>Rewind</title>
      <link>https://www.aigraveyard.org/p/rewind-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/rewind-ai/</guid>
      <pubDate>Fri, 19 Dec 2025 00:00:00 GMT</pubDate>
      <description>A Mac app that recorded and indexed everything on a user&apos;s screen and microphone so they could search their past with natural language.</description>
      <content:encoded><![CDATA[<p>Rewind pivoted in April 2024 away from the local-first Mac app to a $99 wearable pendant called Limitless, moving the architecture cloud-side and rebranding the company. Meta acquired Limitless in December 2025, after which the team announced the original Rewind app would disable all screen and audio capture on December 19, 2025. The Mac product that defined the brand was effectively killed in favor of the wearable, then folded into Meta.</p>]]></content:encoded>
    </item>
    <item>
      <title>Microsoft QnA Maker</title>
      <link>https://www.aigraveyard.org/p/microsoft-qna-maker/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/microsoft-qna-maker/</guid>
      <pubDate>Fri, 31 Oct 2025 00:00:00 GMT</pubDate>
      <description>An Azure service that turned FAQ pages and documents into a question-answering bot.</description>
      <content:encoded><![CDATA[<p>QnA Maker let developers point at a FAQ URL or upload a document and get back a hosted knowledge base with a REST API. Microsoft announced its retirement in May 2022 in favor of the question-answering feature in Azure AI Language, and stopped allowing new QnA Maker resources on October 1, 2022. The portal went offline on March 31, 2025, and the underlying service was fully retired on October 31, 2025. Like LUIS, it was a casualty of the shift from curated knowledge bases to retrieval-augmented LLMs.</p>]]></content:encoded>
    </item>
    <item>
      <title>Microsoft Language Understanding (LUIS)</title>
      <link>https://www.aigraveyard.org/p/microsoft-luis/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/microsoft-luis/</guid>
      <pubDate>Fri, 31 Oct 2025 00:00:00 GMT</pubDate>
      <description>An Azure cloud service for adding natural language intent recognition to bots and apps.</description>
      <content:encoded><![CDATA[<p>LUIS was Microsoft's first attempt at giving developers a managed natural language understanding service, with intent and entity models trained from a web portal. Microsoft stopped allowing new LUIS resources in April 2023 and pushed customers toward Conversational Language Understanding inside Azure AI Language, citing newer transformer-based architectures. After one short extension, the LUIS portal and APIs went dark on October 31, 2025. The retirement was driven less by failure than by obsolescence: the service was designed for a pre-LLM world where developers hand-labeled hundreds of utterances per intent.</p>]]></content:encoded>
    </item>
    <item>
      <title>Dot (New Computer)</title>
      <link>https://www.aigraveyard.org/p/new-computer-dot/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/new-computer-dot/</guid>
      <pubDate>Sun, 05 Oct 2025 00:00:00 GMT</pubDate>
      <description>A personalized AI companion app pitched as an assistant that grows with you and keeps a living diary.</description>
      <content:encoded><![CDATA[<p>New Computer was founded by former Apple designer Jason Yuan and engineer Sam Whitmore, who raised $3.7 million from the OpenAI Startup Fund, Lachy Groom, and South Park Commons. Dot launched on iOS in June 2024 with a "Chronicles" feature that wrote a memory-rich journal of the user's life. On September 5, 2025 the co-founders announced their shared "Northstar had diverged" and would wind operations down. The app went offline on October 5, 2025. Third-party data put lifetime iOS downloads at roughly 24,500, and the shutdown landed in the middle of broader safety scrutiny of AI companion apps.</p>]]></content:encoded>
    </item>
    <item>
      <title>Humanloop</title>
      <link>https://www.aigraveyard.org/p/humanloop/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/humanloop/</guid>
      <pubDate>Mon, 08 Sep 2025 00:00:00 GMT</pubDate>
      <description>A development platform for building, evaluating, and monitoring LLM applications, used by companies like Duolingo, Gusto, and Vanta.</description>
      <content:encoded><![CDATA[<p>Humanloop spun out of University College London in 2020 and built tooling for the unglamorous work of shipping LLM features: prompt management, evaluation, and observability, with customers including Duolingo, Gusto, and Vanta. It raised about 44 million dollars, including a 36.2 million dollar Series A in May 2025. Three months later, in August 2025, Anthropic hired the founders and roughly a dozen staff, pointedly without buying the company's assets or intellectual property. With the team gone, Humanloop shut its platform on September 8, 2025 and deleted customer data, the value having walked out the door in people rather than code. Raising a large Series A and disappearing inside a single quarter made it one of 2025's cleaner illustrations of the acqui-hire as an exit.</p>]]></content:encoded>
    </item>
    <item>
      <title>theGist</title>
      <link>https://www.aigraveyard.org/p/thegist/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/thegist/</guid>
      <pubDate>Wed, 03 Sep 2025 00:00:00 GMT</pubDate>
      <description>An AI assistant that read across Slack, email, and CRM to hand knowledge workers one prioritized digest a day.</description>
      <content:encoded><![CDATA[<p>theGist emerged in 2022 with 7 million dollars in pre-seed funding co-led by StageOne Ventures and Aleph, and a founding bench that included former Wix president Nir Zohar. Its assistant read across Slack, email, CRM, and analytics tools, then delivered a single prioritized digest meant to spare knowledge workers the daily scroll. The product reached market in 2023 with around ten employees split between Tel Aviv and New York, but a pivot toward sales-focused features never gained traction, and prolonged military reserve duty among key engineers slowed development further. In September 2025 the company said it would shut down and return its remaining capital to investors, a rare gesture in a year crowded with messier failures.</p>]]></content:encoded>
    </item>
    <item>
      <title>Astra</title>
      <link>https://www.aigraveyard.org/p/astra-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/astra-ai/</guid>
      <pubDate>Thu, 31 Jul 2025 00:00:00 GMT</pubDate>
      <description>An AI chief of staff for sales reps that promised to automate most of an account executive&apos;s grunt work.</description>
      <content:encoded><![CDATA[<p>Astra was founded in 2023 in Bengaluru by IIT Madras alumni Supreet Hegde and Ranjan Rajagopalan, pitching software that could handle up to 80 percent of an account executive's daily tasks. Perplexity founder Aravind Srinivas joined as an angel in an undisclosed pre-seed round in March 2025. Four months later the company was gone, having signed only two enterprise clients in beta, hampered by long sales cycles and customer wariness about granting an AI access to sensitive CRM and email data. CEO Hegde announced the shutdown on LinkedIn in late July 2025, citing irreconcilable differences with his cofounder over the pace of growth rather than any failure of the product. The frontier had also moved: broad sales-assistant tools were being swallowed by the industry's turn toward autonomous agents.</p>]]></content:encoded>
    </item>
    <item>
      <title>CentML</title>
      <link>https://www.aigraveyard.org/p/centml/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/centml/</guid>
      <pubDate>Thu, 17 Jul 2025 00:00:00 GMT</pubDate>
      <description>A Canadian startup that built an optimization layer between AI models and chips to squeeze more performance and lower cost out of compute.</description>
      <content:encoded><![CDATA[<p>CentML was founded in 2022 by University of Toronto professor Gennady Pekhimenko and three colleagues to sit between machine-learning models and the hardware they run on, wringing out cost and latency. It raised about 31 million dollars, anchored by a 27 million dollar seed round in October 2023 led by Google's Gradient Ventures, with Nvidia among its backers. Less than two years later Nvidia bought the company outright, and CentML told customers it would discontinue its services on July 17, 2025. The startup that optimized other people's compute became a line item on the balance sheet of the company that sells it.</p>]]></content:encoded>
    </item>
    <item>
      <title>CodeParrot</title>
      <link>https://www.aigraveyard.org/p/codeparrot/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/codeparrot/</guid>
      <pubDate>Thu, 17 Jul 2025 00:00:00 GMT</pubDate>
      <description>A Y Combinator-backed tool that turned Figma designs and screenshots into production front-end code.</description>
      <content:encoded><![CDATA[<p>CodeParrot joined Y Combinator's Winter 2023 batch as one of more than a hundred AI startups in a single cohort, building a VS Code extension that turned Figma designs and screenshots into React, Flutter, and HTML. Cofounders Royal Jain and Vedant Agarwala raised 500,000 dollars and found early traction, but never converted it into revenue or a follow-on round. A final pivot to LLM-based Figma-to-code reached only about 1,500 dollars in monthly recurring revenue before the money ran out. The team shut the company down in July 2025, with a cofounder reflecting that startups are brutally hard. By then GitHub Copilot and Replit had folded comparable design-to-code features into products developers already used.</p>]]></content:encoded>
    </item>
    <item>
      <title>Pepper</title>
      <link>https://www.aigraveyard.org/p/pepper-aldebaran/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/pepper-aldebaran/</guid>
      <pubDate>Tue, 03 Jun 2025 00:00:00 GMT</pubDate>
      <description>SoftBank&apos;s four-foot humanoid robot with an emotion-reading face and a chest tablet, sold for retail greeting and home companionship.</description>
      <content:encoded><![CDATA[<p>Pepper arrived in 2014 from Aldebaran, the French robotics firm SoftBank had bought in 2012, a waist-high humanoid with a tablet on its chest and a face designed to read and mirror human emotion. About 27,000 were built for stores, hotels, and living rooms, where they mostly stood around as costly novelties that struggled with the conversations they had been sold to hold. SoftBank halted production in 2021 as demand stalled, then stopped funding Aldebaran after years of losses. Aldebaran filed for bankruptcy in February 2025 and was placed in receivership on June 3, 2025, its staff laid off and the future of its robots uncertain. Pepper had been the public face of the social-robot dream, which is why its long, slow failure was watched so closely.</p>]]></content:encoded>
    </item>
    <item>
      <title>Builder.ai</title>
      <link>https://www.aigraveyard.org/p/builder-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/builder-ai/</guid>
      <pubDate>Tue, 20 May 2025 00:00:00 GMT</pubDate>
      <description>A no-code app builder that promised software as easy as ordering pizza, with an AI assistant that turned out to be mostly human.</description>
      <content:encoded><![CDATA[<p>Founded as Engineer.ai in 2016 by Sachin Dev Duggal, Builder.ai promised to make custom software "as easy as ordering pizza," with an AI assistant named Natasha assembling apps from reusable blocks. The reality, former staff and reporters found, leaned heavily on roughly 700 human engineers in India doing the work the AI was credited with. The company raised more than 445 million dollars from Microsoft, the Qatar Investment Authority, and SoftBank affiliates, reaching a 1.5 billion dollar valuation, before restated financials and allegations of inflated revenue triggered a cash crisis. A lender swept tens of millions from its accounts in early 2025, and on May 20, 2025 the new chief executive told employees the company would file for insolvency. It was, by most accounts, the first major bust of the AI funding boom.</p>]]></content:encoded>
    </item>
    <item>
      <title>Tome</title>
      <link>https://www.aigraveyard.org/p/tome-presentations/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/tome-presentations/</guid>
      <pubDate>Wed, 30 Apr 2025 00:00:00 GMT</pubDate>
      <description>An AI presentation tool that let users generate full decks from a prompt and grew to millions of users at its peak.</description>
      <content:encoded><![CDATA[<p>Tome hit twenty-five million users and over $80M raised but, by founder Keith Peiris's own account, never converted that scale into a workable business since users could not export to PowerPoint or fit corporate workflows. After layoffs in April 2024 the team pivoted to a sales product called Lightfield and announced the presentation product would sunset by the end of April 2025. AngelList ended up buying the Tome brand assets while the original deck product was killed.</p>]]></content:encoded>
    </item>
    <item>
      <title>Dashworks</title>
      <link>https://www.aigraveyard.org/p/dashworks/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/dashworks/</guid>
      <pubDate>Wed, 16 Apr 2025 00:00:00 GMT</pubDate>
      <description>An AI assistant that answered work questions by searching across a company&apos;s documents, messages, tickets, and connected apps.</description>
      <content:encoded><![CDATA[<p>Dashworks was founded in 2019 by Prasad Kawthekar and Pratyaksh Sharma and went through Y Combinator, building an enterprise assistant that answered natural-language questions by pulling across a company's documents, chats, tickets, and apps. It raised about 9.5 million dollars from investors including Y Combinator, Point72, and Twitch cofounder Emmett Shear. In April 2025 HubSpot acquired the company, primarily for the team, and folded the technology into its Breeze AI suite. The standalone Dashworks product stopped taking new sign-ups at the announcement and was wound down, the search assistant becoming a HubSpot feature. It was a soft landing for the founders and a 404 for the product.</p>]]></content:encoded>
    </item>
    <item>
      <title>Humane AI Pin</title>
      <link>https://www.aigraveyard.org/p/humane-ai-pin/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/humane-ai-pin/</guid>
      <pubDate>Fri, 28 Feb 2025 00:00:00 GMT</pubDate>
      <description>A screenless wearable AI assistant that clipped to your shirt.</description>
      <content:encoded><![CDATA[<p>Humane raised over two hundred million dollars and a decade of hype to launch a wearable that did less than the phone in the same pocket. Reviewers hated the laser projection, the heat, the latency, the cost. The promise was a post-phone interface. The product was a slower, hotter, more expensive way to ask a question. HP bought the assets in February 2025 for $116M and folded the team into its enterprise AI group. The pin itself was bricked for existing users on February 28, 2025.</p>]]></content:encoded>
    </item>
    <item>
      <title>Magic Leap One</title>
      <link>https://www.aigraveyard.org/p/magic-leap-one/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/magic-leap-one/</guid>
      <pubDate>Tue, 31 Dec 2024 00:00:00 GMT</pubDate>
      <description>A $2,295 mixed-reality headset that used computer-vision AI to anchor virtual objects in physical rooms.</description>
      <content:encoded><![CDATA[<p>Magic Leap raised more than $2.6 billion from Google, Alibaba, and Saudi Arabia's Public Investment Fund on the promise that its spatial-computing headset would replace the smartphone. The Creator Edition shipped on August 8, 2018 and sold roughly 6,000 units in its first six months against an internal target of 100,000. In April 2020, the company laid off about half its workforce, halted consumer plans, and pivoted to enterprise under new CEO Peggy Johnson. Cloud services for the original headset were switched off on December 31, 2024, bricking the device entirely. Magic Leap 2 continues to be sold as an enterprise product, but the original consumer-facing ML1 is dead.</p>]]></content:encoded>
    </item>
    <item>
      <title>Moxie (Embodied)</title>
      <link>https://www.aigraveyard.org/p/embodied-moxie/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/embodied-moxie/</guid>
      <pubDate>Tue, 31 Dec 2024 00:00:00 GMT</pubDate>
      <description>An $800 cloud-connected social robot designed to teach social and emotional skills to children.</description>
      <content:encoded><![CDATA[<p>Embodied was founded in 2016 by former iRobot CTO Paolo Pirjanian and raised roughly $86 million across seven rounds to build Moxie, a Pixar-styled robot aimed at children on the autism spectrum. In late November 2024 the company told customers a critical funding round had collapsed and that Moxie would stop working within days, with no refunds. Because every conversation was routed through cloud LLMs, the units bricked when the servers went dark at the end of December 2024. Embodied included a letter from the in-fiction "Global Robotics Lab" to help parents explain the death to their kids, and videos of children saying goodbye to their robots circulated on TikTok.</p>]]></content:encoded>
    </item>
    <item>
      <title>Figgs AI</title>
      <link>https://www.aigraveyard.org/p/figgs-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/figgs-ai/</guid>
      <pubDate>Mon, 30 Dec 2024 00:00:00 GMT</pubDate>
      <description>A free roleplay chatbot platform where users built custom AI characters for open-ended conversations and interactive fiction.</description>
      <content:encoded><![CDATA[<p>Figgs AI grew quickly inside the Character.AI-style roleplay niche by removing the paywall and message caps that rivals imposed, peaking near 980,000 monthly visits in November 2024. The developers had already announced in May 2024 that they were shifting focus to a new project called Decart AI, and promised maintenance updates never landed. In mid-December 2024, users reported login failures, broken chats, and a suspected data breach exposing passwords. The site went dark on December 30, 2024, confirmed only through a Reddit post, with no email, refund, or migration path for users.</p>]]></content:encoded>
    </item>
    <item>
      <title>Bench</title>
      <link>https://www.aigraveyard.org/p/bench-accounting/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/bench-accounting/</guid>
      <pubDate>Fri, 27 Dec 2024 00:00:00 GMT</pubDate>
      <description>An online bookkeeping service that pitched itself as AI-augmented bookkeeping for small businesses.</description>
      <content:encoded><![CDATA[<p>Bench shut its doors with no warning on December 27, 2024, locking thirty-five thousand small business customers out of their books between Christmas and New Year. TechCrunch reporting and a later Canadian bankruptcy filing revealed more than $65M in debt, a failed late-stage funding push, and operational strain from trying to automate bookkeeping without enough human review to catch systematic errors. Employer.com bought the IP and customer data days later and offered transitions, but the original Bench product was gone.</p>]]></content:encoded>
    </item>
    <item>
      <title>Cruise</title>
      <link>https://www.aigraveyard.org/p/cruise-robotaxi/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/cruise-robotaxi/</guid>
      <pubDate>Tue, 10 Dec 2024 00:00:00 GMT</pubDate>
      <description>GM&apos;s driverless robotaxi service, grounded after a pedestrian was dragged and then scrapped once the parent company stopped funding it.</description>
      <content:encoded><![CDATA[<p>Cruise was founded in 2013 by Kyle Vogt and Daniel Kan, sold to General Motors in 2016, and spent the next eight years as the centerpiece of GM's self-driving ambitions. At its peak it ran paid driverless robotaxis in San Francisco, Phoenix, and Austin and carried a 30 billion dollar valuation backed by SoftBank, Honda, Microsoft, and Walmart. In October 2023 one of its cars struck and dragged a pedestrian who had been thrown into its path by another vehicle, California pulled its permits, and the service never recovered. On December 10, 2024, GM said it would stop funding robotaxi development and fold the team into its personal-vehicle driver-assistance work, ending the ride-hailing product after the company had spent more than 10 billion dollars on it. Outside investors had put in billions more.</p>]]></content:encoded>
    </item>
    <item>
      <title>Forward</title>
      <link>https://www.aigraveyard.org/p/forward-health/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/forward-health/</guid>
      <pubDate>Tue, 12 Nov 2024 00:00:00 GMT</pubDate>
      <description>AI-powered primary care, delivered through standalone kiosks.</description>
      <content:encoded><![CDATA[<p>Forward started as a chic membership clinic and pivoted to CarePods, walk-in medical booths that promised to diagnose dozens of conditions without a doctor in the room. The pods got stuck on patients, misread tests, and ran into the same regulatory and trust problems that have stalled every algorithm-only clinic. The company shut down all locations in November 2024 and laid off staff with little warning. Investors who had bet on a self-service future of medicine were left with the lease costs of empty kiosks.</p>]]></content:encoded>
    </item>
    <item>
      <title>OctoAI</title>
      <link>https://www.aigraveyard.org/p/octoai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/octoai/</guid>
      <pubDate>Thu, 31 Oct 2024 00:00:00 GMT</pubDate>
      <description>A managed platform for running and optimizing open-source AI models cheaply across different hardware, built by the creators of Apache TVM.</description>
      <content:encoded><![CDATA[<p>OctoAI started in 2019 as OctoML, founded by the University of Washington researchers behind the Apache TVM compiler, with a pitch aimed at the unglamorous middle of the AI stack: run any model on any hardware, faster and cheaper. It raised about 132 million dollars, reaching a roughly 900 million dollar valuation at its 2021 Series C led by Tiger Global, and rebranded to OctoAI as it leaned into generative-model inference. In September 2024 Nvidia acquired the company, and the OctoAI cloud service shut down on October 31, 2024, customer accounts deactivated, with no successor product. It was a strong team and a real technology, absorbed by the chipmaker whose hardware it had been built to optimize. The model-serving layer turned out to be a feature of the silicon, not a company.</p>]]></content:encoded>
    </item>
    <item>
      <title>Amazon Astro for Business</title>
      <link>https://www.aigraveyard.org/p/amazon-astro-for-business/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/amazon-astro-for-business/</guid>
      <pubDate>Wed, 25 Sep 2024 00:00:00 GMT</pubDate>
      <description>A roving security-guard version of Amazon&apos;s Astro home robot, sold to small businesses.</description>
      <content:encoded><![CDATA[<p>Amazon launched Astro for Business in November 2023, repackaging its invite-only home robot as a 2,349-dollar autonomous patrol unit with a 60-dollar-per-month Astro Secure subscription and an optional 99-dollar Virtual Security Guard add-on. Reviewers questioned what a short, cargo-bay-equipped robot actually added over fixed cameras, and Amazon found the small-business market thin. On July 3, 2024, the company told customers the program would end, offered full refunds plus a 300-dollar credit, and bricked the units on September 25, 2024, after seven months of service. Amazon said it would refocus on the consumer Astro, which itself remains invite-only.</p>]]></content:encoded>
    </item>
    <item>
      <title>Meta AI Celebrity Personas</title>
      <link>https://www.aigraveyard.org/p/meta-ai-celebrity-personas/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/meta-ai-celebrity-personas/</guid>
      <pubDate>Wed, 31 Jul 2024 00:00:00 GMT</pubDate>
      <description>A lineup of 28 Meta chatbots dressed in the names and faces of celebrities like Tom Brady, Kendall Jenner, Snoop Dogg, and Paris Hilton.</description>
      <content:encoded><![CDATA[<p>At its Connect event in September 2023, Meta unveiled 28 AI chatbots wearing the names and likenesses of celebrities and influencers, reportedly paying stars like Tom Brady, Kendall Jenner, Charli D'Amelio, and Snoop Dogg millions to lend their faces to bots with invented personas. Users found them more uncanny than charming, and almost nobody talked to them. Less than a year later, in the summer of 2024, Meta quietly retired the characters, and a spokesperson confirmed people could no longer interact with the celebrity-embodied AIs. The company redirected the effort into AI Studio, a tool that lets any creator build a bot, which is a more useful idea and a quiet admission that paying for celebrity avatars had not been. The personas that cost the most to license were among the first to go dark.</p>]]></content:encoded>
    </item>
    <item>
      <title>GameOn</title>
      <link>https://www.aigraveyard.org/p/gameon-chatbot/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/gameon-chatbot/</guid>
      <pubDate>Wed, 10 Jul 2024 00:00:00 GMT</pubDate>
      <description>A sports-focused AI chatbot platform that powered fan conversations for the NBA, Arsenal, and the Las Vegas Raiders.</description>
      <content:encoded><![CDATA[<p>GameOn raised what investors believed was around $35 million in disclosed funding, including a 2022 Series B, to build conversational agents that sports teams and leagues could deploy across messaging apps. CEO Alexander Beckman resigned on July 1, 2024, and ten days later the company disclosed that its financial statements had been fraudulent for years and laid off nearly all of its roughly 60 staff. In January 2025 Beckman and his wife Valerie Lau Beckman were indicted on 23 and 16 counts respectively, with prosecutors alleging more than $60 million was raised through forged audits and fake bank statements and spent on a Tesla, real estate, and a lavish wedding. The product went dark with the layoffs in July 2024.</p>]]></content:encoded>
    </item>
    <item>
      <title>Adept AI</title>
      <link>https://www.aigraveyard.org/p/adept-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/adept-ai/</guid>
      <pubDate>Fri, 28 Jun 2024 00:00:00 GMT</pubDate>
      <description>An agent startup training large models to operate existing software and APIs on behalf of knowledge workers.</description>
      <content:encoded><![CDATA[<p>Adept raised four hundred million dollars to chase general-purpose computer-using agents, but it burned through capital faster than it could ship a defensible product, and the agent space got crowded fast. In June 2024 Amazon hired the founders and key staff and licensed the tech in a reverse acqui-hire, paying investors back rather than acquiring the company outright. The remaining Adept entity was left as a shell, and the FTC opened an inquiry into the structure of the deal.</p>]]></content:encoded>
    </item>
    <item>
      <title>Splash</title>
      <link>https://www.aigraveyard.org/p/splash-music/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/splash-music/</guid>
      <pubDate>Mon, 22 Apr 2024 00:00:00 GMT</pubDate>
      <description>An AI music generator from the Australian studio behind a hit Roblox music game, able to compose backing tracks and synthesize singing from a text prompt.</description>
      <content:encoded><![CDATA[<p>Splash began in Brisbane in 2017 as Popgun, an AI music research outfit, and grew into a studio whose Roblox music game drew tens of millions of plays before it launched a standalone generator, Splash Pro, in September 2023. The product could compose piano, bass, and drums and synthesize singing voices from a text prompt, riding the 2023 wave of interest in generative audio. It had raised a 27 million dollar Series A in 2021 co-led by Amazon's Alexa Fund and BITKRAFT, with Khosla Ventures among the backers. About eight months after Splash Pro shipped, the company decided its Roblox game and an upcoming Steam title were the better bet, announced in March 2024 that it would discontinue the AI music products, and cut off free access on April 22, 2024. The music tool was a detour, and the company chose to go back to making games.</p>]]></content:encoded>
    </item>
    <item>
      <title>GitHub Copilot Voice (&quot;Hey, GitHub&quot;)</title>
      <link>https://www.aigraveyard.org/p/github-copilot-voice/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/github-copilot-voice/</guid>
      <pubDate>Wed, 03 Apr 2024 00:00:00 GMT</pubDate>
      <description>A voice-controlled coding assistant that let developers dictate code and editor commands to Copilot.</description>
      <content:encoded><![CDATA[<p>Announced in November 2022 as a technical preview called "Hey, GitHub," the voice tool targeted developers who had difficulty using a keyboard. It was rebranded Copilot Voice when Copilot Chat launched. In early 2024 GitHub emailed preview users that the project would not become a product and shut down the service on April 3, 2024. Microsoft folded some of the speech capabilities into the VS Code Speech extension, which lets developers trigger Copilot Chat by saying "Hey, Code." Accessibility advocates noted that no equivalent first-party replacement shipped alongside the shutdown.</p>]]></content:encoded>
    </item>
    <item>
      <title>Ghost Autonomy</title>
      <link>https://www.aigraveyard.org/p/ghost-autonomy/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/ghost-autonomy/</guid>
      <pubDate>Wed, 03 Apr 2024 00:00:00 GMT</pubDate>
      <description>Aftermarket self-driving software for consumer cars.</description>
      <content:encoded><![CDATA[<p>Ghost wanted to skip the robotaxi war and sell autonomy software directly to automakers, with a late pivot to applying large multimodal models to driving. The bet was that LLM-style reasoning could close the gap on the slow grind of perception stacks. The funding environment for capital-intensive autonomy collapsed before the bet could pay off. Founder John Hayes announced the wind-down in April 2024, returning remaining cash to investors. A clean exit, but an expensive one.</p>]]></content:encoded>
    </item>
    <item>
      <title>Forefront Chat</title>
      <link>https://www.aigraveyard.org/p/forefront-chat/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/forefront-chat/</guid>
      <pubDate>Thu, 22 Feb 2024 00:00:00 GMT</pubDate>
      <description>A free consumer chat app that gave users access to GPT-4, Claude, and persona-based AI characters in one interface.</description>
      <content:encoded><![CDATA[<p>Forefront launched its consumer chat in mid-2023 offering free access to frontier models, but covering the inference cost without a paid tier proved unworkable. As OpenAI, Google, and Anthropic strengthened their own free tiers through late 2023, the differentiation collapsed. The team shut Forefront Chat on February 22, 2024 and refocused the parent company on developer tools for fine-tuning open-source models.</p>]]></content:encoded>
    </item>
    <item>
      <title>TuSimple (US)</title>
      <link>https://www.aigraveyard.org/p/tusimple-us/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/tusimple-us/</guid>
      <pubDate>Wed, 17 Jan 2024 00:00:00 GMT</pubDate>
      <description>An autonomous trucking company that ran some of the first driverless freight runs in the US before retreating to Asia.</description>
      <content:encoded><![CDATA[<p>TuSimple's US arm was hobbled by a federal probe into alleged technology transfers to Hydron, a Chinese hydrogen-truck startup tied to co-founder Mo Chen, followed by boardroom and SEC turmoil. After failing to find a buyer, TuSimple laid off about seventy-five percent of US staff in late 2023, voluntarily delisted from Nasdaq on January 17, 2024, and auctioned off its autonomous fleet and R&D gear. The company kept its Asia-Pacific operations and abandoned the US market.</p>]]></content:encoded>
    </item>
    <item>
      <title>Artifact</title>
      <link>https://www.aigraveyard.org/p/artifact-news/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/artifact-news/</guid>
      <pubDate>Fri, 12 Jan 2024 00:00:00 GMT</pubDate>
      <description>Instagram founders&apos; AI-driven news reader that ranked, summarized, and de-clickbaited the feed.</description>
      <content:encoded><![CDATA[<p>Artifact was the second act for Instagram cofounders Kevin Systrom and Mike Krieger, an AI-driven news reader that launched in early 2023 and used machine learning to rank stories, summarize articles, and flag clickbait headlines. Critics praised the personalization, but the app never found a large enough audience: the United States accounted for 44 percent of downloads and no other country topped 4 percent. In January 2024 the founders announced they were winding it down, concluding the market opportunity was not big enough to justify continued investment. The product limped along for a few weeks before its underlying technology was sold to Yahoo that April and folded into a revamped Yahoo News. The standalone Artifact app, the thing users actually held, was finished.</p>]]></content:encoded>
    </item>
    <item>
      <title>Olive AI</title>
      <link>https://www.aigraveyard.org/p/olive-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/olive-ai/</guid>
      <pubDate>Tue, 31 Oct 2023 00:00:00 GMT</pubDate>
      <description>Robotic process automation for hospital back offices.</description>
      <content:encoded><![CDATA[<p>Olive sold hospitals on the idea of an AI worker that could automate prior auth, claims, and revenue cycle tasks. The pitch outran the product. Customers reported that what looked like AI was often scripted RPA glued together with services work, and savings did not match the marketing. Once procurement teams compared notes, renewals dried up. Olive sold off two product lines in 2023 and shut the rest of the company in October that year.</p>]]></content:encoded>
    </item>
    <item>
      <title>Babylon Health</title>
      <link>https://www.aigraveyard.org/p/babylon-health/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/babylon-health/</guid>
      <pubDate>Tue, 15 Aug 2023 00:00:00 GMT</pubDate>
      <description>AI-driven symptom checker and virtual primary care service that contracted with the NHS and US insurers.</description>
      <content:encoded><![CDATA[<p>Babylon raised over a billion dollars and went public via a $4.2B SPAC in 2021, but the economics of full-risk Medicaid and Medicare contracts in the US collapsed faster than the AI symptom checker could justify, and the share price cratered. A planned take-private deal with MindMaze fell apart in August 2023, and the US entity filed Chapter 7 liquidation while the UK arm went into administration weeks later. UK assets were sold to eMed for half a million pounds, a near-total wipeout from the SPAC valuation.</p>]]></content:encoded>
    </item>
    <item>
      <title>Microsoft Cortana</title>
      <link>https://www.aigraveyard.org/p/microsoft-cortana/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/microsoft-cortana/</guid>
      <pubDate>Fri, 11 Aug 2023 00:00:00 GMT</pubDate>
      <description>Microsoft&apos;s voice assistant for Windows, phones, and smart speakers, retired in favor of Copilot.</description>
      <content:encoded><![CDATA[<p>Cortana launched on Windows Phone in 2014 as Microsoft's answer to Siri and Google Now, named for the AI character in Halo. It expanded to Windows 10, iOS, Android, the Harman Kardon Invoke speaker, and the first Surface Headphones, then steadily retreated as Microsoft lost its phone bet and consumers picked Alexa instead. Microsoft shut the iOS and Android apps on March 31, 2021, ended Invoke support the same year, and announced in June 2023 that the standalone Windows app would be deprecated; the app stopped working on August 11, 2023. The remaining Cortana surfaces in Teams, Outlook mobile, and Microsoft 365 were retired through late 2023 and mid-2024, with Microsoft directing users to Copilot.</p>]]></content:encoded>
    </item>
    <item>
      <title>Amazon Halo</title>
      <link>https://www.aigraveyard.org/p/amazon-halo/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/amazon-halo/</guid>
      <pubDate>Mon, 31 Jul 2023 00:00:00 GMT</pubDate>
      <description>Amazon&apos;s health-tracking wristband and subscription app that scanned body fat from phone photos and judged the emotional tone of the wearer&apos;s voice.</description>
      <content:encoded><![CDATA[<p>Amazon launched Halo in August 2020 as its bid for the wrist, a screenless band paired with a subscription app whose signature features were a body-fat estimate taken from phone photos and a Tone analysis that scored the emotional quality of the wearer's voice. Both landed as more unsettling than useful, and the hardware never threatened the Apple Watch or Fitbit. On April 26, 2023, Amazon announced it was discontinuing the entire Halo line in a cost-cutting round; support ended July 31, 2023, and the devices and app stopped working the next day, turning every band into inert plastic. Amazon refunded purchases from the prior year and the unused portion of subscriptions, a tidy ending for a product that had asked customers to wear its camera and its microphone at once.</p>]]></content:encoded>
    </item>
    <item>
      <title>Microsoft Soundscape</title>
      <link>https://www.aigraveyard.org/p/microsoft-soundscape/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/microsoft-soundscape/</guid>
      <pubDate>Fri, 30 Jun 2023 00:00:00 GMT</pubDate>
      <description>A Microsoft Research app that used 3D spatial audio to call out streets, shops, and intersections to blind and low-vision users as they walked.</description>
      <content:encoded><![CDATA[<p>Soundscape came out of Microsoft Research in 2018 after more than six years of work, and it did something few products attempted: instead of turn-by-turn directions, it painted an ambient audio map, calling out shops, parks, and intersections in 3D sound as a blind or low-vision user moved through them. It was admired by the people who relied on it and was never built to make money. Microsoft pulled it from the App Store on January 3, 2023 and ended support on June 30, releasing the code as open source so the disability community could try to keep it alive. The funeral was gentler than most, but a beloved accessibility tool maintained by a trillion-dollar company still ended up as a volunteer project, the company's attention having moved on to Copilot.</p>]]></content:encoded>
    </item>
    <item>
      <title>Joonko</title>
      <link>https://www.aigraveyard.org/p/joonko/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/joonko/</guid>
      <pubDate>Sun, 25 Jun 2023 00:00:00 GMT</pubDate>
      <description>An AI recruiting startup that claimed to source diverse job candidates, brought down when its founder fabricated customers and revenue.</description>
      <content:encoded><![CDATA[<p>Joonko sold an AI platform that promised to surface diverse, pre-qualified job candidates into corporate hiring systems, and it raised about 38 million dollars, including a 25 million dollar Series B in 2022 led by Insight Partners, on the strength of a client list that named Adidas, PayPal, and others. In June 2023 the company's own board found that founder and chief executive Ilit Raz had fabricated customers, revenue, bank statements, and contracts; the real business was a fraction of what investors had been shown. Operations were wound down within days, and the company filed for bankruptcy in May 2024. The following month the SEC and the Department of Justice charged Raz with defrauding investors of more than 27 million dollars. Joonko is the archive's clearest case of a product that was less artificial intelligence than artificial traction.</p>]]></content:encoded>
    </item>
    <item>
      <title>Neeva</title>
      <link>https://www.aigraveyard.org/p/neeva/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/neeva/</guid>
      <pubDate>Fri, 02 Jun 2023 00:00:00 GMT</pubDate>
      <description>An ad-free, subscription-based AI search engine pitched as a privacy-respecting Google alternative.</description>
      <content:encoded><![CDATA[<p>Neeva could not convince enough consumers to switch from Google or pay a subscription for search, even after layering on generative AI features. The founders said the hardest problem was not getting users to pay but getting them to try a new search engine at all, and the macro environment made continuing unviable. Snowflake bought the team and tech for enterprise AI use shortly after the consumer product was shut down.</p>]]></content:encoded>
    </item>
    <item>
      <title>Pear Therapeutics</title>
      <link>https://www.aigraveyard.org/p/pear-therapeutics/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/pear-therapeutics/</guid>
      <pubDate>Fri, 07 Apr 2023 00:00:00 GMT</pubDate>
      <description>A prescription digital therapeutics company making FDA-cleared software apps for substance use disorder and insomnia.</description>
      <content:encoded><![CDATA[<p>Pear got FDA clearance for three prescription digital therapeutics including reSET and Somryst, but insurers refused to reimburse them at scale, leaving the company without a viable payer model. After going public via a $1.6B SPAC in late 2021, the stock collapsed and the company filed Chapter 11 in April 2023, with CEO Corey McCann resigning the same day. The assets were broken up at auction for just six million dollars across four buyers.</p>]]></content:encoded>
    </item>
    <item>
      <title>Mindstrong Health</title>
      <link>https://www.aigraveyard.org/p/mindstrong-health/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/mindstrong-health/</guid>
      <pubDate>Fri, 10 Mar 2023 00:00:00 GMT</pubDate>
      <description>A mental health platform that read smartphone typing and scrolling patterns as digital biomarkers of cognitive state.</description>
      <content:encoded><![CDATA[<p>Mindstrong bet that passive smartphone signals could serve as an early warning system for mental illness, but the digital biomarker science never matured to clinical utility, and the company pivoted to providing actual telehealth therapy at much higher cost. With $160M burned and no path to profitability, the board wound down operations in early 2023, laying off about 130 staff including the CEO. SonderMind bought the remaining technology assets in March 2023.</p>]]></content:encoded>
    </item>
    <item>
      <title>Embark Trucks</title>
      <link>https://www.aigraveyard.org/p/embark-trucks/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/embark-trucks/</guid>
      <pubDate>Fri, 03 Mar 2023 00:00:00 GMT</pubDate>
      <description>An autonomous trucking startup building a self-driving software platform for long-haul Class 8 freight.</description>
      <content:encoded><![CDATA[<p>Embark went public via SPAC in November 2021 at roughly a $5B target valuation, then watched capital markets close on pre-revenue AV plays while truck OEM timelines slipped further out. By March 2023 the company cut seventy percent of staff, told employees it could not find a path forward, and began winding down. Applied Intuition bought the remaining IP and Embark's autonomous program went dark.</p>]]></content:encoded>
    </item>
    <item>
      <title>Mycroft AI</title>
      <link>https://www.aigraveyard.org/p/mycroft-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/mycroft-ai/</guid>
      <pubDate>Fri, 10 Feb 2023 00:00:00 GMT</pubDate>
      <description>An open-source voice assistant and Linux-based smart speaker positioned as a privacy-respecting alternative to Alexa and Google Assistant.</description>
      <content:encoded><![CDATA[<p>Mycroft launched on Kickstarter with the Mark I and later the Mark II smart speaker, attracting a community of developers who wanted a voice stack they could inspect and self-host. Component costs for the Mark II rose from roughly 99 dollars to about 300 dollars per unit, and a long-running patent suit from Voice Tech Corp drained cash the company could not replace. In February 2023, CEO Michael Lewis announced the company was winding down, with staff reduced to a handful and remaining Kickstarter rewards unfulfilled. The codebase lives on in the community fork OpenVoiceOS, but Mycroft AI as a company stopped shipping product.</p>]]></content:encoded>
    </item>
    <item>
      <title>Google Stadia</title>
      <link>https://www.aigraveyard.org/p/google-stadia/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/google-stadia/</guid>
      <pubDate>Wed, 18 Jan 2023 00:00:00 GMT</pubDate>
      <description>Google&apos;s cloud gaming platform, marketed around machine-learning features for streaming and game development.</description>
      <content:encoded><![CDATA[<p>Stadia launched in November 2019 with promises of negative latency, ML-driven style transfer for in-game art, and a Stadia-exclusive games studio. The studio was closed in February 2021 before shipping anything original, and the service never reached the user numbers Google needed to justify the data-center cost. Phil Harrison's team kept adding titles into 2022, but on September 29 of that year Google announced the platform would be turned off and that all hardware and software purchases would be refunded. The Stadia servers went offline on January 18, 2023; the underlying streaming tech survived inside YouTube and Google's immersive efforts but the consumer product was gone.</p>]]></content:encoded>
    </item>
    <item>
      <title>Meta Portal</title>
      <link>https://www.aigraveyard.org/p/meta-portal/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/meta-portal/</guid>
      <pubDate>Sat, 31 Dec 2022 00:00:00 GMT</pubDate>
      <description>Meta&apos;s line of smart video-calling displays whose Smart Camera used computer vision to pan, zoom, and keep callers in frame automatically.</description>
      <content:encoded><![CDATA[<p>Facebook launched Portal in November 2018, a family of smart displays built around video calling, whose standout feature was a Smart Camera that used computer vision to pan, zoom, and keep people framed as they moved around a room. It launched into the worst possible headwind for the company that made it: deep public distrust of putting a Facebook camera and microphone in the living room. The devices found a small foothold among remote workers during the pandemic and never became the consumer hit Meta wanted. In 2022 Meta stopped developing consumer Portals, sold through the remaining inventory by year end, and pointed what was left toward business before letting the line lapse entirely. Existing units kept working, but the Portal as a product Meta sold and built was finished.</p>]]></content:encoded>
    </item>
    <item>
      <title>Galactica</title>
      <link>https://www.aigraveyard.org/p/galactica/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/galactica/</guid>
      <pubDate>Thu, 17 Nov 2022 00:00:00 GMT</pubDate>
      <description>Meta&apos;s large language model for science, trained on 48 million papers, textbooks, and encyclopedias that could summarize research, generate equations, and annotate molecules.</description>
      <content:encoded><![CDATA[<p>Meta released the public demo on November 15, 2022 and pulled it three days later after scientists flooded social media with examples of confident, citation-laden hallucinations including fake papers about bears in space and invented chemical compounds. Critics warned the tool could pollute scientific literature with plausible-sounding misinformation. Yann LeCun later argued it was killed by the Twitter mob just two weeks before ChatGPT normalized the same hallucination behavior.</p>]]></content:encoded>
    </item>
    <item>
      <title>Kite</title>
      <link>https://www.aigraveyard.org/p/kite-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/kite-ai/</guid>
      <pubDate>Wed, 16 Nov 2022 00:00:00 GMT</pubDate>
      <description>A Python-focused AI code completion plugin that predated GitHub Copilot by years.</description>
      <content:encoded><![CDATA[<p>Kite spent eight years and roughly $17 million building an AI assistant for Python developers, attracting around 500,000 monthly active users with almost no marketing. The company never solved monetization. In a farewell post, founder Adam Smith concluded that individual developers do not pay for tools and that the underlying models could not understand non-local code context well enough to deliver a true 10x improvement. The team open-sourced the codebase under BSD-3 and shut the product down in November 2022, less than five months after GitHub Copilot left preview.</p>]]></content:encoded>
    </item>
    <item>
      <title>Mighty</title>
      <link>https://www.aigraveyard.org/p/mighty/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/mighty/</guid>
      <pubDate>Mon, 14 Nov 2022 00:00:00 GMT</pubDate>
      <description>A cloud-streamed Chromium browser that ran Chrome on remote servers to free up local laptop resources.</description>
      <content:encoded><![CDATA[<p>Mighty tried to sell a thirty-dollar-a-month cloud browser subscription to power users, but the premise weakened as Apple Silicon and newer laptop chips closed the single-core performance gap with server hardware. After three and a half years, Doshi concluded the market was too small and pivoted the remaining team and capital into a new AI image product, Playground AI. The cloud browser was wound down in November 2022.</p>]]></content:encoded>
    </item>
    <item>
      <title>Argo AI</title>
      <link>https://www.aigraveyard.org/p/argo-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/argo-ai/</guid>
      <pubDate>Wed, 26 Oct 2022 00:00:00 GMT</pubDate>
      <description>Autonomous vehicle software, backed by Ford and Volkswagen.</description>
      <content:encoded><![CDATA[<p>Argo had two of the biggest automakers in the world writing checks and a team pulled from Google and Uber. None of it was enough to close the gap from impressive demo to a vehicle a customer would buy. Ford announced in late 2022 that profitable fully autonomous driving was further out than the timeline the company could fund. The wind-down was clean: Ford and VW absorbed some engineers, the rest were laid off, and the broader industry quietly revised its self-driving timeline by a decade.</p>]]></content:encoded>
    </item>
    <item>
      <title>Amazon Scout</title>
      <link>https://www.aigraveyard.org/p/amazon-scout/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/amazon-scout/</guid>
      <pubDate>Fri, 07 Oct 2022 00:00:00 GMT</pubDate>
      <description>Amazon&apos;s six-wheeled, cooler-sized autonomous robot that rolled along sidewalks delivering packages to customers&apos; doors.</description>
      <content:encoded><![CDATA[<p>Amazon unveiled Scout in January 2019, a knee-high six-wheeled robot built partly on its 2017 acquisition of the startup Dispatch, and set it rolling down sidewalks in Washington, California, Georgia, and Tennessee to deliver packages. Every robot was shadowed by a human Scout Ambassador, which undercut the entire premise of autonomous delivery. In October 2022 Amazon ended the field tests and disbanded the roughly 400-person team, saying the program was not meeting customers' needs. The company framed it as a reorientation rather than a cancellation, but the robots never returned to the sidewalks. Four years on, the customer-facing product is a memory.</p>]]></content:encoded>
    </item>
    <item>
      <title>Amazon Glow</title>
      <link>https://www.aigraveyard.org/p/amazon-glow/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/amazon-glow/</guid>
      <pubDate>Tue, 04 Oct 2022 00:00:00 GMT</pubDate>
      <description>A $300 video-calling device for children that projected interactive games and puzzles onto a table.</description>
      <content:encoded><![CDATA[<p>Amazon unveiled Glow on September 28, 2021 as an invite-only pandemic product, then opened sales to the general public in March 2022. The device used computer vision to track hands and physical pieces on a projected play surface, letting a remote grandparent see the same puzzle a child was working on. Content partners included Disney, Mattel, Nickelodeon, and Sesame Workshop. Sales were poor, and on October 4, 2022, against a backdrop of broad hardware cost cuts inside Amazon, the company killed Glow roughly a year after launch and offered refunds. Nothing replaced it.</p>]]></content:encoded>
    </item>
    <item>
      <title>DeepArt</title>
      <link>https://www.aigraveyard.org/p/deepart-io/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/deepart-io/</guid>
      <pubDate>Mon, 01 Aug 2022 00:00:00 GMT</pubDate>
      <description>One of the first public neural style-transfer tools, turning any photo into artwork in the style of a second image.</description>
      <content:encoded><![CDATA[<p>DeepArt launched in 2015 out of the University of Tubingen, built directly on the landmark paper that introduced neural style transfer, and let anyone upload a photo and repaint it in the style of a chosen artwork. For a couple of years it was many people's first encounter with a neural network doing something visibly creative, a direct ancestor of the image generators that followed. It was also slow and expensive to run, and faster mobile rivals like Prisma took the audience. The site went dark in the second half of 2022, its servers no longer worth the cloud bill, and the operating company was later wound down. DeepArt proved the public appetite for AI image-making years before the tools that would monetize it existed, then exited before the boom it had helped seed.</p>]]></content:encoded>
    </item>
    <item>
      <title>Sonantic</title>
      <link>https://www.aigraveyard.org/p/sonantic/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/sonantic/</guid>
      <pubDate>Mon, 13 Jun 2022 00:00:00 GMT</pubDate>
      <description>A London voice-AI startup that synthesized emotional, human-sounding speech for games and film.</description>
      <content:encoded><![CDATA[<p>Sonantic spent four years building text-to-speech models that could perform anger, fear, and grief well enough to satisfy game studios and Hollywood. Its most public credit was recreating Val Kilmer's voice for Top Gun: Maverick after throat cancer left him unable to speak. Spotify bought the company in June 2022 for a reported 93 million euros and shut the standalone Sonantic product, folding the team and the technology into what would become Spotify's AI DJ feature in early 2023. The Sonantic platform stopped accepting new customers immediately after the deal and went dark for existing ones over the following months.</p>]]></content:encoded>
    </item>
    <item>
      <title>Vicarious</title>
      <link>https://www.aigraveyard.org/p/vicarious/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/vicarious/</guid>
      <pubDate>Fri, 22 Apr 2022 00:00:00 GMT</pubDate>
      <description>A long-horizon AI research startup building brain-inspired models for industrial robotics and visual reasoning.</description>
      <content:encoded><![CDATA[<p>Vicarious spent over a decade and a quarter billion dollars from Bezos, Musk, Zuckerberg and Samsung trying to build human-level AI through Recursive Cortical Networks, but a viable commercial product never materialized as deep learning approaches overtook its brain-inspired methods. In April 2022 the team was split between Alphabet's Intrinsic robotics unit and DeepMind, and the original Vicarious entity was folded. Financial terms were never disclosed and the outcome is widely seen as a soft landing rather than a successful exit.</p>]]></content:encoded>
    </item>
    <item>
      <title>IBM Watson Health</title>
      <link>https://www.aigraveyard.org/p/ibm-watson-health/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/ibm-watson-health/</guid>
      <pubDate>Fri, 21 Jan 2022 00:00:00 GMT</pubDate>
      <description>IBM&apos;s flagship healthcare AI division that promised to use Watson to diagnose disease, match cancer treatments, and run hospital analytics.</description>
      <content:encoded><![CDATA[<p>Watson Health spent years and billions on flagship oncology and analytics deals that underdelivered, with high-profile partners like MD Anderson walking away and clinicians reporting unsafe treatment recommendations. IBM announced on January 21, 2022 that it would sell the bulk of the unit to private equity firm Francisco Partners for a reported price above $1B. The buyer rolled the assets into a new company, Merative, ending the Watson Health brand and its broader AI-for-medicine pitch.</p>]]></content:encoded>
    </item>
    <item>
      <title>Local Motors Olli</title>
      <link>https://www.aigraveyard.org/p/local-motors-olli/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/local-motors-olli/</guid>
      <pubDate>Fri, 14 Jan 2022 00:00:00 GMT</pubDate>
      <description>A 3D-printed, electric, self-driving shuttle designed for low-speed campus and city routes.</description>
      <content:encoded><![CDATA[<p>Local Motors debuted Olli in 2016 at National Harbor in Maryland as a microtransit shuttle whose body was largely 3D-printed and whose conversational interface ran on IBM Watson. Olli 2.0 was meant to begin a Toronto pilot in 2021, but an Olli 1.0 shuttle hit a tree in December of that year and the city paused trials. The company struggled with high per-unit production costs, slow regulatory uptake, and a funding round that did not arrive. Employees learned on LinkedIn that operations had ended on January 14, 2022, with no formal press release from the company.</p>]]></content:encoded>
    </item>
    <item>
      <title>Optimus Ride</title>
      <link>https://www.aigraveyard.org/p/optimus-ride/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/optimus-ride/</guid>
      <pubDate>Tue, 11 Jan 2022 00:00:00 GMT</pubDate>
      <description>An MIT spinout building low-speed autonomous electric shuttles for geofenced communities and campuses.</description>
      <content:encoded><![CDATA[<p>Optimus Ride ran real autonomous shuttle pilots at the Brooklyn Navy Yard and other geofenced sites, but the small-scale low-speed model could not scale into a venture-grade business as the broader AV market consolidated around well-capitalized robotaxi players. In January 2022 Magna bought the technology, IP, and about 120 engineers, while CEO Sean Harrington did not transition over and the company ceased operations. The deal was a soft landing for the team rather than a strategic exit.</p>]]></content:encoded>
    </item>
    <item>
      <title>x.ai (Amy and Andrew)</title>
      <link>https://www.aigraveyard.org/p/xai-amy-andrew/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/xai-amy-andrew/</guid>
      <pubDate>Sun, 31 Oct 2021 00:00:00 GMT</pubDate>
      <description>AI scheduling assistants Amy and Andrew Ingram that handled meeting coordination over email, one of the first widely deployed autonomous agent products. Not the Elon Musk xAI.</description>
      <content:encoded><![CDATA[<p>x.ai pioneered the email-based AI agent pattern but spent years relying on a large back office of human trainers to clean up natural language scheduling requests, and never found unit economics that worked at scale against free Calendly-style link tools. Bizzabo acquired the company in June 2021 to fold the scheduling tech into its events platform, then sunset the standalone Amy and Andrew product on October 31, 2021. Mortensen has since written candidly about how the vertical agent thesis arrived a decade too early on the underlying model quality.</p>]]></content:encoded>
    </item>
    <item>
      <title>Voyage Auto</title>
      <link>https://www.aigraveyard.org/p/voyage-auto/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/voyage-auto/</guid>
      <pubDate>Mon, 15 Mar 2021 00:00:00 GMT</pubDate>
      <description>An autonomous taxi startup that operated low-speed self-driving cars inside private retirement communities.</description>
      <content:encoded><![CDATA[<p>Voyage had narrowed its scope to senior-living communities like The Villages in Florida, but it could not scale that niche into a defensible AV business as Waymo, Cruise, and others raised billions. Cruise bought the roughly sixty-person team in March 2021, with CEO Oliver Cameron moving over as VP of product. Voyage's own deployments and brand were folded into Cruise and never resumed independently.</p>]]></content:encoded>
    </item>
    <item>
      <title>Element AI</title>
      <link>https://www.aigraveyard.org/p/element-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/element-ai/</guid>
      <pubDate>Fri, 08 Jan 2021 00:00:00 GMT</pubDate>
      <description>A Montreal-based enterprise AI services company building custom machine learning solutions for large companies.</description>
      <content:encoded><![CDATA[<p>Element AI raised over a quarter billion dollars at a $600M valuation pitching itself as a top applied AI lab, but the consulting-heavy business model never produced repeatable enterprise software revenue. By late 2020 the company was running short on cash and sold to ServiceNow for roughly $230M, a price at which most founders and early employees saw their equity wiped out. ServiceNow kept the research talent and patents and wound down most of the existing customer business.</p>]]></content:encoded>
    </item>
    <item>
      <title>ROSS Intelligence</title>
      <link>https://www.aigraveyard.org/p/ross-intelligence/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/ross-intelligence/</guid>
      <pubDate>Fri, 11 Dec 2020 00:00:00 GMT</pubDate>
      <description>An AI-powered legal research platform that used natural language queries to surface relevant case law for lawyers.</description>
      <content:encoded><![CDATA[<p>ROSS marketed itself as a faster alternative to Westlaw, pitching IBM Watson-era machine learning to law firms hunting for citations. In May 2020, Thomson Reuters sued, alleging ROSS had used a Westlaw licensee to copy headnotes and train its system on protected content. The litigation froze fundraising. By December the company told customers its bank account was running out and it would cease operations in the new year. A 2025 summary-judgment ruling in Delaware later found the copying was not fair use, long after the product was gone.</p>]]></content:encoded>
    </item>
    <item>
      <title>Starsky Robotics</title>
      <link>https://www.aigraveyard.org/p/starsky-robotics/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/starsky-robotics/</guid>
      <pubDate>Thu, 19 Mar 2020 00:00:00 GMT</pubDate>
      <description>A self-driving truck startup that paired highway autonomy with remote human teleoperation for the first and last miles.</description>
      <content:encoded><![CDATA[<p>A planned $20M Series B fell through in November 2019, forcing Starsky to furlough most staff and try to sell the company before winding down. Co-founder Seltz-Axmacher blamed the collapse on a broader cooling toward AV bets, the high capital intensity of trucking, and supervised learning failing to deliver the reliability investors expected. No buyer materialized for the remote-piloting stack and the company shut down operations in March 2020.</p>]]></content:encoded>
    </item>
    <item>
      <title>Atrium</title>
      <link>https://www.aigraveyard.org/p/atrium-lts/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/atrium-lts/</guid>
      <pubDate>Tue, 03 Mar 2020 00:00:00 GMT</pubDate>
      <description>A law firm and software company that promised to automate startup legal work with machine learning.</description>
      <content:encoded><![CDATA[<p>Atrium opened in September 2017 as a hybrid: a law firm staffed by traditional attorneys paired with Atrium LTS, a software arm building tools to digitize contracts, generate cap tables from term sheets, and automate routine legal work. Andreessen Horowitz led a $65 million Series B in 2018 on the pitch that machine learning would erode billable-hour pricing. The technology did not move fast enough to justify the cost structure. In January 2020 Atrium laid off most of its in-house lawyers and tried to become a pure software company, but customers churned and a follow-on round did not come together. On March 3, 2020, Justin Kan announced Atrium LTS was shutting down and laying off its remaining staff of about 100. The standalone law firm spun out separately.</p>]]></content:encoded>
    </item>
    <item>
      <title>Snips</title>
      <link>https://www.aigraveyard.org/p/snips-voice/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/snips-voice/</guid>
      <pubDate>Tue, 03 Dec 2019 00:00:00 GMT</pubDate>
      <description>A privacy-first voice assistant platform that ran entirely on-device, pitched to hardware makers as an alternative to Alexa and Google Assistant.</description>
      <content:encoded><![CDATA[<p>Snips was founded in Paris in 2013 by Rand Hindi, Mael Primet, and Michael Fester to build a voice assistant that ran on the device itself and sent nothing to the cloud, an architecture that read as either principled or quixotic depending on the year. It raised about 21 million dollars and built a developer platform that let gadget makers add private voice control without Amazon or Google in the loop. In November 2019 Sonos bought the company for 37.5 million dollars and within weeks shut the developer console, ending anyone's ability to build or run Snips voice systems. Sonos wanted the team and the on-device technology for its own music controls, not the platform, and the platform went dark. The privacy pitch that made Snips distinctive was absorbed into a speaker company and stopped being a product anyone else could use.</p>]]></content:encoded>
    </item>
    <item>
      <title>Google Pixel Buds (1st generation)</title>
      <link>https://www.aigraveyard.org/p/google-pixel-buds-gen1/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/google-pixel-buds-gen1/</guid>
      <pubDate>Tue, 15 Oct 2019 00:00:00 GMT</pubDate>
      <description>Wired Bluetooth earbuds with a Google Assistant trigger and a real-time translation feature pitched as a Babel-fish moment for AI.</description>
      <content:encoded><![CDATA[<p>Launched alongside the Pixel 2 in October 2017 at 159 dollars, the original Pixel Buds leaned hard on a Google Translate demo that Sundar Pichai used to suggest the earbuds could break down language barriers in real time. Reviewers found the fit awkward, the touch controls unreliable, and the translation feature heavily dependent on the paired phone rather than the buds themselves. Google removed them from its store when the second-generation, truly wireless Pixel Buds were announced at the October 15, 2019 hardware event. The original model was never updated again and was quietly delisted from support pages.</p>]]></content:encoded>
    </item>
    <item>
      <title>Google Clips</title>
      <link>https://www.aigraveyard.org/p/google-clips/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/google-clips/</guid>
      <pubDate>Tue, 15 Oct 2019 00:00:00 GMT</pubDate>
      <description>A clip-on camera that used on-device machine learning to decide on its own when to take a picture.</description>
      <content:encoded><![CDATA[<p>Google announced Clips at its October 2017 Pixel 2 event and put the $249 camera on sale in January 2018. The pitch was that an on-device neural network would learn faces it saw often and silently capture short clips of children and pets without anyone needing to press a button. Reviewers found the results unreliable and the always-watching premise unnerving, and sales never recovered. Google quietly pulled the product from its store on October 15, 2019 and ended cloud sync at the close of 2021. It is one of the cleanest examples of an AI-first consumer hardware concept that the market refused to want.</p>]]></content:encoded>
    </item>
    <item>
      <title>Lyrebird</title>
      <link>https://www.aigraveyard.org/p/lyrebird-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/lyrebird-ai/</guid>
      <pubDate>Wed, 18 Sep 2019 00:00:00 GMT</pubDate>
      <description>A Montreal voice-cloning startup that recreated a speaker&apos;s voice from a one-minute sample.</description>
      <content:encoded><![CDATA[<p>Lyrebird spun out of Yoshua Bengio's MILA lab in 2017 and immediately became one of the cautionary tales of the deepfake era, after demos of synthesized Obama and Trump voices circulated widely. The team built a consumer voice-cloning site and a developer API, but never found a business that scaled beyond novelty. Descript bought the company in September 2019 and retired the Lyrebird product, putting the team to work on an audio-editor feature called Overdub. The standalone Lyrebird site was redirected to Descript shortly afterward, and the brand was retired.</p>]]></content:encoded>
    </item>
    <item>
      <title>Microsoft Zo</title>
      <link>https://www.aigraveyard.org/p/microsoft-zo/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/microsoft-zo/</guid>
      <pubDate>Fri, 19 Jul 2019 00:00:00 GMT</pubDate>
      <description>Microsoft&apos;s English-language successor to the Tay chatbot, designed for casual conversation on messaging platforms.</description>
      <content:encoded><![CDATA[<p>Zo launched on Kik in December 2016 as Microsoft's attempt to redo Tay without the racism, with heavy filters that refused to discuss politics, religion, or anything that could embarrass the company. The filters made Zo dull and evasive, and reporters quickly documented her producing offensive replies anyway when prompted in particular ways. Microsoft began winding her down in March 2019, pulling her from Twitter, Skype, and Kik, then ended Facebook and Samsung phone support that July, and switched off the GroupMe integration in September. The product never found a real audience and was overshadowed inside Microsoft by the work that would become Cortana and, later, Bing Chat.</p>]]></content:encoded>
    </item>
    <item>
      <title>Drive.ai</title>
      <link>https://www.aigraveyard.org/p/drive-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/drive-ai/</guid>
      <pubDate>Tue, 25 Jun 2019 00:00:00 GMT</pubDate>
      <description>A self-driving car startup spun out of Stanford&apos;s AI Lab, known for retrofitted vans running geofenced shuttle pilots in Texas.</description>
      <content:encoded><![CDATA[<p>Drive.ai was on the brink of closure and had filed plans to lay off about ninety staff when Apple stepped in days before shutdown for an acqui-hire focused on engineers and assets for Project Titan. The startup had been squeezed by bigger AV competitors and could not raise on its $200M peak valuation. Apple kept the talent and hardware; the Drive.ai product and brand were retired.</p>]]></content:encoded>
    </item>
    <item>
      <title>Knewton</title>
      <link>https://www.aigraveyard.org/p/knewton/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/knewton/</guid>
      <pubDate>Fri, 31 May 2019 00:00:00 GMT</pubDate>
      <description>A heavily funded adaptive-learning company that promised to personalize coursework for every student in real time.</description>
      <content:encoded><![CDATA[<p>Knewton was founded in 2008 by former Kaplan executive Jose Ferreira and became one of edtech's best-funded bets, raising more than 180 million dollars on the promise that its adaptive engine could tailor coursework to each student in real time. The technology was endlessly demonstrated and never quite turned into a business, and a late pivot to its own Alta courseware did not save it. In May 2019 the textbook publisher Wiley bought Knewton's assets in what observers called a fire sale, with the price reported in the low tens of millions, a fraction of what had gone in. Knewton ceased to exist as an independent company, and Alta was folded into Wiley's higher-education catalog. It remains a reference point for how much money personalized learning could absorb without producing a durable product.</p>]]></content:encoded>
    </item>
    <item>
      <title>Anki</title>
      <link>https://www.aigraveyard.org/p/anki/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/anki/</guid>
      <pubDate>Wed, 01 May 2019 00:00:00 GMT</pubDate>
      <description>A consumer robotics company behind the Cozmo and Vector AI companion robots and the Anki Drive racing system, marketed as real robotics in toy form.</description>
      <content:encoded><![CDATA[<p>Three CMU Robotics Institute alumni built a company that sold over 1.5 million robots and made Cozmo the best-selling toy of 2017, but Anki burned close to two hundred million dollars and could not sustain hardware margins at consumer price points. A planned funding round collapsed at the last minute in April 2019, acquisition talks with Microsoft, Amazon, and Comcast fell through, and the entire staff of roughly two hundred was laid off on May 1, 2019. Digital Dream Labs later bought the IP and brought back limited Cozmo and Vector service.</p>]]></content:encoded>
    </item>
    <item>
      <title>Laundroid</title>
      <link>https://www.aigraveyard.org/p/laundroid-seven-dreamers/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/laundroid-seven-dreamers/</guid>
      <pubDate>Tue, 23 Apr 2019 00:00:00 GMT</pubDate>
      <description>A cabinet-sized robot from Japan&apos;s Seven Dreamers that used cameras and AI to fold laundry, priced around $16,000 and never shipped a unit.</description>
      <content:encoded><![CDATA[<p>Seven Dreamers Laboratories spent years showing the Laundroid at CES, a wardrobe-sized machine that used cameras, AI, and robotic arms to pick up, recognize, and fold arbitrary garments, with a planned price around 16,000 dollars. Folding a random shirt turned out to be one of those tasks that is trivial for a person and brutal for a robot, and the company never shipped a single unit. It raised roughly 10 billion yen, about 89 million dollars, from backers including Panasonic and Daiwa House before the money ran out. On April 23, 2019, Seven Dreamers filed for bankruptcy with about 2.2 billion yen in debt and retreated to its older business making nasal-breathing devices. The most expensive way ever devised to fold a towel was itself folded up.</p>]]></content:encoded>
    </item>
    <item>
      <title>Sentient Technologies</title>
      <link>https://www.aigraveyard.org/p/sentient-technologies/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/sentient-technologies/</guid>
      <pubDate>Thu, 21 Mar 2019 00:00:00 GMT</pubDate>
      <description>One of the best-funded AI startups of its era, built on evolutionary computation and deep learning, that was broken up and sold for parts.</description>
      <content:encoded><![CDATA[<p>Sentient Technologies was founded in 2007 by Antoine Blondeau, Babak Hodjat, and Adam Cheyer, the last of whom had helped create the technology behind Siri, and it raised more than 143 million dollars, including a 103.5 million dollar round in 2014 that was then the largest ever for an AI company. It ran an AI hedge fund and built products for e-commerce optimization and visual personalization across a vast grid of borrowed compute. The hedge fund was liquidated in 2018 after underperforming, and in 2019 the company itself was taken apart: its Ascend optimization product was sold to Evolv in March, its core AI intellectual property went to Cognizant, and the Sentient brand stopped existing. Enormous funding and credentialed founders produced a set of pieces worth more to other companies than the whole had been worth on its own. It is the rare entry that died by dispersal rather than by a single shutdown.</p>]]></content:encoded>
    </item>
    <item>
      <title>Keecker</title>
      <link>https://www.aigraveyard.org/p/keecker/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/keecker/</guid>
      <pubDate>Tue, 12 Mar 2019 00:00:00 GMT</pubDate>
      <description>A French home robot that drove from room to room projecting movies and music while doubling as a security and smart-home hub.</description>
      <content:encoded><![CDATA[<p>Keecker was founded in 2012 by former Google product manager Pierre Lebeau, who built a mobile robot that wandered the home projecting movies and music onto walls and ceilings while monitoring the house. It raised about 8 million dollars and put more than a thousand robots into homes, with owners using them several hours a day. None of that translated into a fundable business, and the company ran out of cash when no new investment arrived. Lebeau announced the shutdown in March 2019 and pushed a final update that stripped the server dependencies so the deployed robots could keep playing media offline. In his closing note he cited the same headwinds that had felled Jibo and Kuri: too early, too expensive, too little capital.</p>]]></content:encoded>
    </item>
    <item>
      <title>Google Allo</title>
      <link>https://www.aigraveyard.org/p/google-allo/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/google-allo/</guid>
      <pubDate>Tue, 12 Mar 2019 00:00:00 GMT</pubDate>
      <description>A messaging app built around Smart Reply suggestions and the first consumer outing of Google Assistant.</description>
      <content:encoded><![CDATA[<p>Allo launched on September 21, 2016 as Google's answer to WhatsApp and iMessage, and was the first app to ship Google Assistant. Its headline AI feature was Smart Reply, which proposed canned responses based on the conversation and on photos sent in the chat. The product never escaped Google's older messaging properties, and a phone-number-only signup with no SMS fallback kept growth flat. In April 2018 Google paused investment and reassigned the team to RCS in Android Messages. Allo stopped working on March 12, 2019, with Smart Reply and Assistant integration folded into Messages.</p>]]></content:encoded>
    </item>
    <item>
      <title>Jibo</title>
      <link>https://www.aigraveyard.org/p/jibo/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/jibo/</guid>
      <pubDate>Mon, 04 Mar 2019 00:00:00 GMT</pubDate>
      <description>The self-described first social robot for the home, a swiveling tabletop companion that recognized faces, held conversations, and told stories.</description>
      <content:encoded><![CDATA[<p>Jibo raised $3.5M on Indiegogo in 2014 and went on to raise around seventy-three million in venture capital, but it did not ship the robot until late 2017, by which point Amazon Echo and Google Home had taken the home assistant market at a fraction of Jibo's $899 price. The company sold its assets to SQN Venture Partners in 2018, and on March 4, 2019 the cloud servers were shut down, with each robot delivering a now-famous farewell speech and final dance before going dark. NTT Disruption later acquired patents and limited assets.</p>]]></content:encoded>
    </item>
    <item>
      <title>Lighthouse AI</title>
      <link>https://www.aigraveyard.org/p/lighthouse-ai/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/lighthouse-ai/</guid>
      <pubDate>Tue, 18 Dec 2018 00:00:00 GMT</pubDate>
      <description>A smart home security camera with 3D sensing and computer vision that could recognize specific family members, pets, and strangers and respond to natural language queries.</description>
      <content:encoded><![CDATA[<p>Two Stanford self-driving car researchers from Sebastian Thrun's lab built a three-hundred-dollar camera packed with depth sensing and on-device recognition, backed by Playground Global, Eclipse Ventures, and SignalFire. The product launched into a market already saturated by Ring, Nest, Arlo, and sub-$100 Wyze cameras, and could not justify its price premium to mainstream buyers. The company shut down in December 2018, offered full refunds through January 25, 2019, and Apple later bought its patent portfolio in March 2019.</p>]]></content:encoded>
    </item>
    <item>
      <title>Kuri</title>
      <link>https://www.aigraveyard.org/p/kuri-mayfield-robotics/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/kuri-mayfield-robotics/</guid>
      <pubDate>Wed, 31 Oct 2018 00:00:00 GMT</pubDate>
      <description>A $700 wheeled home robot with expressive eyes and a built-in camera, made by a Bosch-backed startup to roam the house and film candid moments.</description>
      <content:encoded><![CDATA[<p>Mayfield Robotics was incubated inside Bosch's startup program and showed Kuri at CES 2016, a 700-dollar home robot with animated eyes, mapping, and a camera meant to capture candid family video. It began shipping to pre-order customers in late 2017, into the same indifference that was about to kill Jibo and Anki. In July 2018 Mayfield stopped production, saying it had found no business fit within Bosch to scale the company, and wound the whole operation down by the end of October. Backers were refunded and allowed to keep their robots, which lost features as the cloud services behind them switched off. Kuri was the friendly face of a category buyers admired in videos and declined to purchase.</p>]]></content:encoded>
    </item>
    <item>
      <title>Otto</title>
      <link>https://www.aigraveyard.org/p/otto-trucking/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/otto-trucking/</guid>
      <pubDate>Mon, 30 Jul 2018 00:00:00 GMT</pubDate>
      <description>A self-driving truck startup that made the first commercial autonomous freight delivery before being absorbed and shut by Uber.</description>
      <content:encoded><![CDATA[<p>Otto came out of stealth in May 2016 and was sold to Uber in August of the same year for a reported 680 million dollars in stock, before it had shipped a real product. In October 2016 it ran a retrofitted truck 120 miles on a Colorado highway to deliver Budweiser, a stunt that defined early autonomous-trucking hype. Waymo then sued, alleging co-founder Anthony Levandowski had walked out of Google with 14,000 confidential files; Uber fired him in 2017 and settled with Waymo for 245 million dollars in 2018. Uber shut the entire self-driving trucks unit on July 30, 2018, citing a need to focus on robotaxis, and laid off or reassigned the remaining team.</p>]]></content:encoded>
    </item>
    <item>
      <title>Yahoo Aviate</title>
      <link>https://www.aigraveyard.org/p/yahoo-aviate/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/yahoo-aviate/</guid>
      <pubDate>Thu, 08 Mar 2018 00:00:00 GMT</pubDate>
      <description>An Android launcher that rearranged your home screen by time of day, location, and what you were doing.</description>
      <content:encoded><![CDATA[<p>Aviate was built by ThumbsUp Labs and launched as an invite-only Android launcher in 2013, promising a home screen that learned your routine and surfaced the right apps at the right moment. Yahoo acquired ThumbsUp in January 2014 in a deal valued around $80 million, opened the launcher to all users, and bolted on a Smart Stream of contextual cards. Updates slowed after Verizon bought Yahoo's operating business. The last release went out in May 2017, and on March 8, 2018, Yahoo killed the app entirely, telling users it would stop functioning. The contextual-launcher idea was later absorbed into Google's own Now and Assistant work.</p>]]></content:encoded>
    </item>
    <item>
      <title>Facebook M</title>
      <link>https://www.aigraveyard.org/p/facebook-m/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/facebook-m/</guid>
      <pubDate>Fri, 19 Jan 2018 00:00:00 GMT</pubDate>
      <description>Facebook&apos;s Messenger concierge, a personal assistant that mixed AI with human contractors to book tables, order gifts, and plan trips.</description>
      <content:encoded><![CDATA[<p>Facebook introduced M in August 2015 as a Messenger-based assistant meant to out-do Siri and Alexa by handling the hard things: booking restaurants, ordering flowers, planning trips. The trick was that humans did the hard things. Behind the chat window sat contractors fielding the requests the AI could not, on the theory that the model would watch and gradually take over. It never did, a human in every loop did not scale, and M stayed stuck in a small private beta most Facebook users never saw. Facebook announced the shutdown on January 8, 2018 and closed the service on January 19, keeping the M name only for a handful of canned Messenger suggestions.</p>]]></content:encoded>
    </item>
    <item>
      <title>Microsoft Tay</title>
      <link>https://www.aigraveyard.org/p/microsoft-tay/</link>
      <guid isPermaLink="true">https://www.aigraveyard.org/p/microsoft-tay/</guid>
      <pubDate>Thu, 24 Mar 2016 00:00:00 GMT</pubDate>
      <description>Microsoft&apos;s Twitter chatbot designed to mimic an American teenage girl and learn through conversation with users.</description>
      <content:encoded><![CDATA[<p>Tay launched on Twitter on March 23, 2016 and was pulled within sixteen hours after coordinated trolls exploited its repeat-after-me function and conversational learning to produce racist, sexist, and antisemitic tweets. Microsoft had no content filters on the public learning loop. The incident became the textbook example of why production chatbots cannot learn from unfiltered public input.</p>]]></content:encoded>
    </item>
  </channel>
</rss>