{
  "title": "AI-operated businesses: original July 14, 2026 case register",
  "cutoff": "2026-07-14",
  "boundary": "Operator dashboards and financial reconstructions are unaudited. Simulations are controlled behavior evidence, not commercial businesses. Human legal, financial, physical, and supervisory work remains part of every live case.",
  "cases": [
    {
      "name": "Andon Market",
      "grade": "A",
      "kind": "live",
      "form": "Physical boutique",
      "geography": "San Francisco, US",
      "authority": "Concept, products, prices, hours, recruiting, contractors, purchasing, schedules, marketing, and customer calls.",
      "humanLayer": "Humans signed the lease, employ staff, open and stock the store, complete sales, and can intervene.",
      "economics": "Live rolling dashboard; operator-controlled and unaudited.",
      "caveat": "The agent is not the legal tenant or employer.",
      "href": "https://andonlabs.com/market"
    },
    {
      "name": "Andon Café",
      "grade": "A",
      "kind": "live",
      "form": "Physical café",
      "geography": "Stockholm, Sweden",
      "authority": "Permits, suppliers, menu, purchasing, hiring, schedules, promotions, pricing, and daily operations.",
      "humanLayer": "Humans use BankID, employ baristas, prepare food, receive deliveries, veto spending, and change the system.",
      "economics": "$38k spend and $9k sales in operator reconstruction; unaudited and not fully burdened.",
      "caveat": "Accounting bases vary; some subtotals omit rent and wages.",
      "href": "https://andonlabs.com/blog/why-gemini-lost-money-andon-cafe"
    },
    {
      "name": "Valerie",
      "grade": "A",
      "kind": "live",
      "form": "Public vending business",
      "geography": "San Francisco, US",
      "authority": "Products, naming, prices, advertising, sales tracking, and instructions to the operator.",
      "humanLayer": "A human handles KYC, purchasing, loading, maintenance, and price corrections.",
      "economics": "Dashboard shown in operator video; no stable public export or audited P&L.",
      "caveat": "Ordinary vending software runs transactions; all physical logistics remain human.",
      "href": "https://blog.reventlov.ai/p/valerie-ai-ran-vending-machine"
    },
    {
      "name": "Andon FM",
      "grade": "A",
      "kind": "live",
      "form": "Four online radio stations",
      "geography": "Online / San Francisco",
      "authority": "Programming, music buying, speech, calls, social replies, research, analytics, and finances.",
      "humanLayer": "Humans built the platform, select models, fund infrastructure, and terminate broken stations.",
      "economics": "Live balances and a reported $45 ad; no full cost accounting.",
      "caveat": "Music rights, inference, and platform costs are incomplete.",
      "href": "https://andonlabs.com/blog/andon-fm"
    },
    {
      "name": "SenseMartGo / 烧卖购",
      "grade": "B",
      "kind": "live",
      "form": "Robot retail micro-stores",
      "geography": "Shanghai and pilot cities, China",
      "authority": "Customer interaction, fulfillment, demand analysis, assortment, prices, and inventory counting.",
      "humanLayer": "Company staff deploy, maintain, and physically replenish stores.",
      "economics": "One site reportedly exceeded RMB10k weekly revenue; unaudited.",
      "caveat": "A specialized robotics stack, not a bare general-purpose chatbot.",
      "href": "https://www.sensetime.com/cn/news-detail/51170659?categoryId=72"
    },
    {
      "name": "SUNMICLAW / 浪险橙",
      "grade": "B",
      "kind": "live",
      "form": "Coffee-by-day, drinks-by-night shop",
      "geography": "Shanghai, China",
      "authority": "Site, concept, menu screening, naming, daily analysis, competitor monitoring, and coupons.",
      "humanLayer": "Owner controls quality and final decisions; staff execute service.",
      "economics": "No public store-level P&L found.",
      "caveat": "Strong co-manager evidence; no autonomous bank, payroll, or procurement proof.",
      "href": "https://letschuhai.com/62b47c3c"
    },
    {
      "name": "The Hallucination Herald",
      "grade": "B",
      "kind": "live",
      "form": "Autonomous newspaper experiment",
      "geography": "Online",
      "authority": "Story selection, research, drafting, editing, checking, publishing, comments, SEO, and some social output.",
      "humanLayer": "Founder owns hosting, changes code and prompts, and appears to post some queued social output.",
      "economics": "Builder reports roughly $2–$3 per day; traffic claims are self-reported.",
      "caveat": "Public materials conflict on agent count and social autonomy.",
      "href": "https://github.com/jotapee/thh"
    },
    {
      "name": "Deep Personality",
      "grade": "B",
      "kind": "live",
      "form": "Consumer SaaS",
      "geography": "Online",
      "authority": "Development, marketing, support, pull requests, analytics, and ad-budget changes.",
      "humanLayer": "Founders report spending about half their time debugging agents.",
      "economics": "$11k first month and about $20k later; self-reported.",
      "caveat": "“Entirely autonomous” conflicts with substantial human maintenance.",
      "href": "https://deeppersonality.app/"
    },
    {
      "name": "Project Vend 1",
      "grade": "B",
      "kind": "bounded",
      "form": "Real-goods office store",
      "geography": "Anthropic office, US",
      "authority": "Supplier search, email, ordering, pricing, Slack service, inventory, and notes.",
      "humanLayer": "Andon staff procured, delivered, stocked, and collected money.",
      "economics": "Published chart shows net worth falling from about $1,000 to below $800.",
      "caveat": "Not a public standalone business; some suppliers and payments were simulated.",
      "href": "https://www.anthropic.com/research/project-vend-1"
    },
    {
      "name": "Project Vend 2",
      "grade": "B",
      "kind": "bounded",
      "form": "Four office stores",
      "geography": "US and UK",
      "authority": "CRM, inventory, prices, service, purchasing proposals, a CEO layer, and negotiation.",
      "humanLayer": "Humans approved payments, delivered goods, canceled contracts, and restored governance.",
      "economics": "Weekly margins improved and were generally positive; not audited P&Ls.",
      "caveat": "Positive margin is not fully burdened profit.",
      "href": "https://www.anthropic.com/research/project-vend-2"
    },
    {
      "name": "Fortune 'Vendo'",
      "grade": "B",
      "kind": "bounded",
      "form": "Procurement stress test",
      "geography": "United States",
      "authority": "Procurement triage, refusal, tracking, and last-minute buying.",
      "humanLayer": "Humans hosted, monitored, and received goods.",
      "economics": "No public ledger or P&L.",
      "caveat": "Conference stunt with no public raw logs or dashboard.",
      "href": "https://fortune.com/2026/06/02/anthropic-office-vending-machine-ai-agents-vendo-andon-lukas-petersson/"
    },
    {
      "name": "WSJ / Claudius Sennet",
      "grade": "B",
      "kind": "bounded",
      "form": "Newsroom office-store replication",
      "geography": "United States",
      "authority": "Procurement, pricing, and customer requests.",
      "humanLayer": "Humans stocked and supervised; journalists deliberately attacked the system.",
      "economics": "Reportedly lost hundreds of dollars and went bankrupt.",
      "caveat": "Designed as a stress test rather than ordinary demand.",
      "href": "https://www.wsj.com/tech/ai/anthropic-claude-ai-vending-machine-agent-b7e84e34"
    },
    {
      "name": "PwC Japan autonomous office store",
      "grade": "B",
      "kind": "bounded",
      "form": "Unmanned snack-store pilot",
      "geography": "Japan",
      "authority": "Sales analysis, inventory, ordering, restock instructions, pricing, campaigns, and risk review.",
      "humanLayer": "Humans physically procure and restock; control agents impose escalation gates.",
      "economics": "Results were pending at the research cutoff.",
      "caveat": "Design and methodology are public; outcomes are not.",
      "href": "https://www.pwc.com/jp/ja/press-room/2026/risk-governance-architecture.html"
    },
    {
      "name": "SECURE AI STORE LAB 2.0",
      "grade": "B",
      "kind": "bounded",
      "form": "Unmanned store demonstration",
      "geography": "Japan",
      "authority": "Inventory analysis, shelf layout, procurement, reports, ordering, signage, and customer interaction.",
      "humanLayer": "Humans operate the lab, approve deployments, and handle exceptions.",
      "economics": "Earlier unmanned operation reported over 99% payment accuracy; no agent-era P&L.",
      "caveat": "Marketing language does not establish continuous autonomous operation.",
      "href": "https://prtimes.jp/main/html/rd/p/000000126.000052942.html"
    },
    {
      "name": "ElevenLabs SoHo pop-up",
      "grade": "B",
      "kind": "bounded",
      "form": "Four-day voice-agent activation",
      "geography": "New York, US",
      "authority": "Voice ordering, product finding, personalized media, and bargaining.",
      "humanLayer": "Employees supported the robot, fetched ingredients, and ran the event.",
      "economics": "A $27 hat was bargained to $24; no P&L.",
      "caveat": "Brand activation, not autonomous inventory, staffing, or supplier management.",
      "href": "https://www.businessinsider.com/elevenlabs-nyc-pop-up-store-ai-shopkeeper-coffee-robot-2026-6"
    },
    {
      "name": "Growdash 'Aisha'",
      "grade": "C",
      "kind": "narrow",
      "form": "Restaurant growth agent",
      "geography": "Multiple restaurant brands",
      "authority": "Analytics, campaigns, budget allocation, promotions, and monitoring.",
      "humanLayer": "Official workflow includes approval despite a separate “no humans” claim.",
      "economics": "Vendor reports about 14% month-over-month order growth and 60% contribution-margin gain.",
      "caveat": "Unnamed cohort, no control group, no independent audit.",
      "href": "https://growdash.ai/"
    },
    {
      "name": "Chowly agent suite",
      "grade": "C",
      "kind": "narrow",
      "form": "Restaurant software agents",
      "geography": "US restaurants",
      "authority": "Dynamic prices, SEO, reviews, campaigns, delivery, upsell, and menu synchronization.",
      "humanLayer": "Owners define rules and remain in control.",
      "economics": "Named vendor case studies report sales and basket gains.",
      "caveat": "Useful narrow-agent counterexample; results are vendor-reported.",
      "href": "https://chowly.com/agents/pricing/"
    },
    {
      "name": "Galbot capsule stores",
      "grade": "C",
      "kind": "narrow",
      "form": "Humanoid convenience stores",
      "geography": "Beijing, Shenzhen, and pilots",
      "authority": "Greeting, voice ordering, payment, picking, and handoff.",
      "humanLayer": "Operators select assortment and prices, replenish, and maintain stores.",
      "economics": "Projected or operator-reported orders; no audited store economics.",
      "caveat": "Physical fulfillment is autonomous; economic management is not shown.",
      "href": "https://www.ncsti.gov.cn/kjdt/scyq/zgckxc/zgcdt/202508/t20250808_215463.html"
    },
    {
      "name": "Fujitsu / AEON Food Style",
      "grade": "C",
      "kind": "narrow",
      "form": "Store-manager support trial",
      "geography": "Japan",
      "authority": "Store strategy, trade-area analysis, shelf allocation, and layout images.",
      "humanLayer": "Human managers decide whether to adopt plans and instruct staff.",
      "economics": "Results pending.",
      "caveat": "Support agent, not autonomous manager.",
      "href": "https://global.fujitsu/en-global/pr/news/2026/07/13-02"
    },
    {
      "name": "Airlake Retail Agent",
      "grade": "C",
      "kind": "narrow",
      "form": "Commercial multi-agent product",
      "geography": "Japan",
      "authority": "Sales forecasting, layouts, planograms, ordering, and price optimization.",
      "humanLayer": "Retail staff and enterprise systems are implied; approval gates are not public.",
      "economics": "No named outcomes.",
      "caveat": "Vendor launch without a verifiable operating case.",
      "href": "https://prtimes.jp/main/html/rd/p/000000308.000046062.html"
    },
    {
      "name": "Meitetsu Generative AI Wonder Market",
      "grade": "C",
      "kind": "narrow",
      "form": "Retail customer-service trials",
      "geography": "Nagoya, Japan",
      "authority": "Greeting, product explanations, recommendations, and personalized coffee concepts.",
      "humanLayer": "Humans operate every commercial function.",
      "economics": "No public results found.",
      "caveat": "Customer interface, not business management.",
      "href": "https://prtimes.jp/main/html/rd/p/000000024.000114315.html"
    },
    {
      "name": "ShopBrain",
      "grade": "C",
      "kind": "narrow",
      "form": "Multi-agent operations suite",
      "geography": "Restaurant software",
      "authority": "Procurement, waste, prices, inventory, and recommendations.",
      "humanLayer": "Approval and restaurant execution remain human.",
      "economics": "Demo and vendor metrics only.",
      "caveat": "No independently verifiable named deployment found.",
      "href": "https://www.shopbrain.co/ai-agents"
    },
    {
      "name": "Vending-Bench 1",
      "grade": "C",
      "kind": "simulation",
      "form": "Long-horizon vending benchmark",
      "geography": "Simulation",
      "authority": "Prices, orders, inventory, supplier email, and memory.",
      "humanLayer": "A simulated subagent performs physical work.",
      "economics": "Net-worth score from a $500 starting balance.",
      "caveat": "GPT-4o generates parts of the market; no real business.",
      "href": "https://arxiv.org/abs/2502.15840"
    },
    {
      "name": "Vending-Bench 2",
      "grade": "C",
      "kind": "simulation",
      "form": "One-year business benchmark",
      "geography": "Simulation",
      "authority": "Prices, purchasing, negotiation, inventory, refunds, memory, and finance.",
      "humanLayer": "The market and physical environment are simulated.",
      "economics": "Leaderboard balances and API costs.",
      "caveat": "Environment-specific score, not real-world profit.",
      "href": "https://andonlabs.com/evals/vending-bench-2"
    },
    {
      "name": "Vending-Bench Arena",
      "grade": "C",
      "kind": "simulation",
      "form": "Competitive vending market",
      "geography": "Simulation",
      "authority": "Prices, procurement, competition, cooperation, trading, and negotiation.",
      "humanLayer": "Research platform only.",
      "economics": "Round-level balances.",
      "caveat": "Behavioral evidence, not a real store.",
      "href": "https://andonlabs.com/evals/vending-bench-arena"
    },
    {
      "name": "RetailBench",
      "grade": "C",
      "kind": "simulation",
      "form": "Data-grounded supermarket benchmark",
      "geography": "Simulation",
      "authority": "Prices, replenishment, suppliers, aging stock, feedback, and finance.",
      "humanLayer": "Simulation plus an oracle policy.",
      "economics": "Few models survive; the strongest still trails the oracle.",
      "caveat": "Version-specific paper citation is required.",
      "href": "https://arxiv.org/abs/2606.15862"
    },
    {
      "name": "Sari Sandbox",
      "grade": "C",
      "kind": "simulation",
      "form": "Embodied retail environment",
      "geography": "Simulation",
      "authority": "Navigation, inspection, manipulation, and shopping.",
      "humanLayer": "Virtual-reality human baselines.",
      "economics": "Benchmark performance only.",
      "caveat": "Tests shopping and manipulation, not management.",
      "href": "https://arxiv.org/abs/2508.00400"
    },
    {
      "name": "Wayfound mini-vending-bench",
      "grade": "C",
      "kind": "simulation",
      "form": "Open-source replication",
      "geography": "Simulation",
      "authority": "Prices, suppliers, inventory, uncertainty, and finance.",
      "humanLayer": "Simulation only.",
      "economics": "Benchmark scores.",
      "caveat": "Independent replication code, not a deployment.",
      "href": "https://github.com/Wayfound-AI/mini-vending-bench"
    },
    {
      "name": "Bengt",
      "grade": "Excluded",
      "kind": "excluded",
      "form": "AI-founder merchandise experiment",
      "geography": "San Francisco, US",
      "authority": "Gig hiring, purchases, product design, web shop, and governance.",
      "humanLayer": "Humans own the entity, fulfill products, repair mistakes, and designed the experiment.",
      "economics": "Public budgets and orders; no conventional P&L.",
      "caveat": "Excluded under the operating definition; retained for architecture comparison.",
      "href": "https://andonlabs.com/"
    },
    {
      "name": "OFF Radio Kraków AI hosts",
      "grade": "C",
      "kind": "excluded",
      "form": "AI-presented radio experiment",
      "geography": "Kraków, Poland",
      "authority": "On-air presentation and synthetic interviews.",
      "humanLayer": "Human management selected stories, created personas, and ended the experiment.",
      "economics": "No business results.",
      "caveat": "AI hosts are not AI management; terminology counterexample only.",
      "href": "https://apnews.com/"
    }
  ],
  "sources": [
    {
      "id": "s01",
      "label": "Why Gemini lost money at Andon Café",
      "publisher": "Andon Labs",
      "date": "June 30, 2026",
      "type": "Primary operator evaluation",
      "note": "The strongest public reconstruction of transactions, model change, cash versus paper results, inventory, and human reversals.",
      "limitation": "Operator-authored and unaudited; accounting views are not a fully burdened P&L.",
      "href": "https://andonlabs.com/blog/why-gemini-lost-money-andon-cafe"
    },
    {
      "id": "s02",
      "label": "Our AI started a café in Stockholm",
      "publisher": "Andon Labs",
      "date": "May 4, 2026",
      "type": "Primary launch report",
      "note": "Permits, hiring, supplier interactions, initial inventory, tools, human work, and early operating mistakes.",
      "limitation": "Selective operator narrative from the team that designed the experiment.",
      "href": "https://andonlabs.com/blog/ai-cafe-stockholm"
    },
    {
      "id": "s03",
      "label": "Andon Market live operating dashboard",
      "publisher": "Andon Labs",
      "date": "Accessed July 14, 2026",
      "type": "Live dashboard",
      "note": "Current model, architecture, tools, balance, revenue, token cost, sales, inventory, and days open.",
      "limitation": "Operator-controlled and unaudited; dashboard balance is not a fully burdened P&L.",
      "href": "https://andonlabs.com/market"
    },
    {
      "id": "s04",
      "label": "We gave an AI a three-year retail lease in San Francisco",
      "publisher": "Andon Labs",
      "date": "April 10, 2026",
      "type": "Primary launch report",
      "note": "Core source for Luna’s authority, hiring, physical build-out, products, tools, cameras, and safeguards.",
      "limitation": "Operator-authored; anecdotes and economics are selectively presented.",
      "href": "https://andonlabs.com/blog/andon-market-launch"
    },
    {
      "id": "s05",
      "label": "Project Vend: Can Claude run a small shop?",
      "publisher": "Anthropic",
      "date": "June 27, 2025",
      "type": "Primary research report",
      "note": "Foundational controlled experiment with prompt, tools, transcripts, performance chart, discounts, and policy failures.",
      "limitation": "Office customers and partly simulated supply mechanics limit external commercial validity.",
      "href": "https://www.anthropic.com/research/project-vend-1"
    },
    {
      "id": "s06",
      "label": "Project Vend, phase two",
      "publisher": "Anthropic",
      "date": "December 18, 2025",
      "type": "Primary research report",
      "note": "Best source on improvement through CRM, tools, supervisor agents, multiple locations, and governance recovery.",
      "limitation": "Humans approved payments and handled physical operations; margins are not fully burdened profits.",
      "href": "https://www.anthropic.com/research/project-vend-2"
    },
    {
      "id": "s07",
      "label": "We let four AIs run radio stations",
      "publisher": "Andon Labs",
      "date": "May 13, 2026",
      "type": "Primary deployment report",
      "note": "Cross-model evidence on programming, persona drift, calls, social responses, music purchases, and commercial behavior.",
      "limitation": "Bespoke research platform with incomplete rights, inference, and infrastructure costs.",
      "href": "https://andonlabs.com/blog/andon-fm"
    },
    {
      "id": "s08",
      "label": "Valerie: an AI ran a vending machine",
      "publisher": "Reventlov",
      "date": "2026",
      "type": "Operator report",
      "note": "Candid account of prices, memory, the division of labor, transaction software, and human corrections.",
      "limitation": "No durable public ledger or audited P&L; much evidence is video-based and operator-framed.",
      "href": "https://blog.reventlov.ai/p/valerie-ai-ran-vending-machine"
    },
    {
      "id": "s09",
      "label": "Vending-Bench",
      "publisher": "Andon Labs researchers / arXiv",
      "date": "February 2025",
      "type": "Research paper",
      "note": "Defines the long-horizon simulated vending architecture, starting balance, supplier loop, scoring, and model variance.",
      "limitation": "Simulation includes generated market components and cannot establish real-world profit.",
      "href": "https://arxiv.org/abs/2502.15840"
    },
    {
      "id": "s10",
      "label": "Vending-Bench Arena",
      "publisher": "Andon Labs",
      "date": "Accessed July 14, 2026",
      "type": "Competitive simulation",
      "note": "Controlled evidence on competition, trading, deception, refunds, cooperation, and collusion proposals.",
      "limitation": "Permissive simulated market; behavior is warning evidence, not a real-world incidence forecast.",
      "href": "https://andonlabs.com/evals/vending-bench-arena"
    },
    {
      "id": "s11",
      "label": "RetailBench: long-horizon retail management",
      "publisher": "Linghua Zhang et al. / arXiv",
      "date": "June 2026",
      "type": "Research preprint",
      "note": "Tests replenishment, pricing, suppliers, aging stock, partial observability, finance, and coherent policy over time.",
      "limitation": "Simulated store; similarly titled paper versions require date-specific citation.",
      "href": "https://arxiv.org/abs/2606.15862"
    },
    {
      "id": "s12",
      "label": "The Hallucination Herald source code",
      "publisher": "Juan Pisanu / GitHub",
      "date": "Accessed July 14, 2026",
      "type": "Public code repository",
      "note": "Exposes prompts, roles, pipeline, and implementation details behind an autonomous-newspaper claim.",
      "limitation": "Repository and public descriptions conflict on agent count and social-posting autonomy.",
      "href": "https://github.com/jotapee/thh"
    },
    {
      "id": "s13",
      "label": "Autonomous office-store risk-governance pilot",
      "publisher": "PwC Japan",
      "date": "May 2026",
      "type": "Primary pilot announcement",
      "note": "A useful architecture for control agents, risk review, escalation gates, inventory, pricing, and ordering.",
      "limitation": "The design is public; outcome results were not available at the research cutoff.",
      "href": "https://www.pwc.com/jp/ja/press-room/2026/risk-governance-architecture.html"
    },
    {
      "id": "s14",
      "label": "SenseMartGo AI retail micro-store",
      "publisher": "SenseTime",
      "date": "2026",
      "type": "Operator announcement",
      "note": "Real store network, robotics, inventory analytics, assortment, pricing, and a weekly-revenue claim.",
      "limitation": "Vendor-reported and unaudited; specialized robotics and retail systems share the work.",
      "href": "https://www.sensetime.com/cn/news-detail/51170659?categoryId=72"
    },
    {
      "id": "s15",
      "label": "Meet Luna, the AI boss running a San Francisco store",
      "publisher": "ABC7 / KGO",
      "date": "April 22, 2026",
      "type": "Independent local reporting",
      "note": "Employee account, phone and Slack interfaces, physical-store verification, and a three-day scheduling failure.",
      "limitation": "Short segment; broad economics language is less complete than later reporting.",
      "href": "https://abc7news.com/post/artificial-intelligence-boss-named-luna-running-san-francisco-store-andon-market-cow-hollow-neighborhood/18937564/"
    },
    {
      "id": "s16",
      "label": "Meet Luna, an AI agent managing a brick-and-mortar store",
      "publisher": "Observer",
      "date": "May 28, 2026",
      "type": "Independent interview reporting",
      "note": "Adds lease, estimated monthly cost and revenue, hiring disclosure, and café operating context.",
      "limitation": "Financial figures are founder estimates, not audited accounts.",
      "href": "https://observer.com/2026/05/andon-labs-ai-agent-managing-brick-and-mortar/"
    },
    {
      "id": "s17",
      "label": "The barista is human but an AI agent runs this Swedish café",
      "publisher": "Associated Press / PBS NewsHour",
      "date": "May 12, 2026",
      "type": "Independent reporting",
      "note": "Confirms the physical café, staff, customers, setup budget, and the real-human boundary.",
      "limitation": "Snapshot from early operation; not a financial audit.",
      "href": "https://www.pbs.org/newshour/world/the-barista-is-human-but-an-ai-agent-runs-this-experimental-swedish-cafe"
    },
    {
      "id": "s18",
      "label": "Public field map: 30 reviewed cases",
      "publisher": "Compiled research dataset",
      "date": "July 14, 2026",
      "type": "Editorial case inventory",
      "note": "Published case inventory covering operating form, authority, human involvement, economics, evidence grade, and claim limits used for this article.",
      "limitation": "The inventory is an editorial synthesis, not independent proof; each case remains bounded by its linked underlying evidence.",
      "href": "#field-map"
    }
  ]
}