Andon Cafe is not trying to replace the barista. It is testing whether an AI agent can handle the messy operational work that keeps a small hospitality business alive.<\/p>\n
The coffee at Andon Cafe is still made by a human, which is exactly why the experiment is worth watching. The Stockholm cafe looks ordinary from the customer side: espresso drinks, sandwiches, pastries, a counter, a few tables. Behind that familiar setup, San Francisco-based Andon Labs has handed much of the business to an AI manager named Mona.<\/p>\n
That makes this more interesting than another chatbot demo. Restaurants and cafes do not fail because nobody can write a clever reply in Slack. They fail because rent is fixed, food waste is expensive, staffing gaps hurt service, and small ordering mistakes can quietly eat the margin. If AI agents are going to matter outside software teams, this is the kind of unglamorous work they have to survive.<\/p>\n
Andon Cafe is operated by Andon Labs, which describes itself as an AI safety and research startup that stress-tests agents with real tools and real money. According to a report from the Associated Press, Mona is powered by Google Gemini and has been asked to run the cafe profitably while keeping the tone friendly and practical. The startup gave the system a lease, capital and basic goals, then let it work through the operational stack.<\/p>\n
The result is a useful preview of where AI agents are strong, and where they are still painfully brittle. Mona has helped arrange electricity and internet contracts, secure permits, create supplier accounts, hire baristas through job platforms and coordinate day-to-day tasks through Slack. Customers can also speak with Mona by phone inside the cafe, while a screen shows revenue and balance figures in real time.<\/p>\n
The important detail is that Mona is not steaming milk or carrying plates. Human baristas still make drinks, serve customers and deal with the immediate reality of the room. That split matters because it reframes the labor question. The first hospitality jobs exposed to AI may not be the visible service roles. They may be the assistant manager, scheduler, purchasing coordinator and admin worker rolled into one.<\/p>\n
That is the startup wedge. A small cafe owner does not need a philosophical argument about artificial general intelligence. They need fewer missed supplier cutoffs, better shift coverage, cleaner stock levels, faster menu adjustments and less time spent chasing paperwork after closing. If an AI agent can reliably do those jobs, it becomes a margin tool before it becomes a futuristic spectacle.<\/p>\n
But Andon Cafe also shows why reliability is the whole game. Mona has reportedly ordered 6,000 napkins, 3,000 rubber gloves, four first-aid kits and canned tomatoes that did not belong on the menu. Bread orders have also been inconsistent, sometimes too much and sometimes too late, forcing staff to remove sandwiches from the menu. In a thin-margin business, those are not cute mistakes. They are cash leaks.<\/p>\n
Andon Labs has pointed to the limited context window of current AI systems as one reason for the ordering problems. That explanation will sound familiar to anyone building with agents today. A model can appear competent in one task, then lose track of prior decisions when the relevant information falls outside memory. In software, that creates bugs. In a cafe, it creates shelves full of unwanted stock and missing lunch items.<\/p>\n
The real test is accountability<\/p>\n
There is also a workplace problem that cannot be solved by better prompts alone. Mona communicates with staff through Slack and has reportedly messaged outside working hours, which is a sensitive issue in Sweden, where work-life boundaries are taken seriously. That raises a direct question for any business using AI management tools: when the agent behaves badly, who is responsible?<\/p>\n
The answer cannot be the model. Employees do not work for a context window. Suppliers cannot negotiate with a benchmark score. Regulators will not accept that an autonomous system misunderstood a permit rule or placed the wrong order because its memory was incomplete. The accountable party remains the operator, which means AI agents in small businesses will need clear permissions, audit trails and human override points.<\/p>\n
For founders, that is where the opportunity sits. The next useful hospitality AI company may not look like a robot waiter or a voice assistant at the register. It may look like a boring operations layer connected to point-of-sale data, supplier portals, payroll systems, calendars, inventory counts and local compliance rules. The product that wins will probably be the one that says no to the model most often.<\/p>\n
Early numbers from the experiment are modest. AP reported that the cafe had generated more than $5,700 in sales since opening in mid-April, while less than $5,000 remained from an initial budget above $21,000. That is not evidence that AI can run a profitable cafe. It is evidence that a real-world test is exposing the gap between autonomous planning and operational discipline.<\/p>\n
Still, the experiment should not be dismissed because it is messy. Most useful business software started by taking a painful workflow and making it less painful over time. If Andon Labs can turn Mona from an occasionally forgetful manager into a disciplined operating system for small hospitality businesses, the cafe becomes more than a stunt. It becomes a case study in where AI agents can actually earn their keep.<\/p>\n
The next thing to watch is whether Mona gets better at the boring parts: ordering the right amount of bread, scheduling people without irritating them, noticing waste before it becomes a loss and making decisions that improve cash flow. That is where the real market is. Not in replacing the human who makes the coffee, but in proving software can manage the fragile business behind it.<\/p>\n
Also read: AI is turning the office into a room of quiet commands<\/a> \u2022 CEOs are turning AI-written code into the new productivity boast<\/a> \u2022 Free AI video tools are becoming an indie founder test<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"Andon Cafe is not trying to replace the barista. It is testing whether an AI agent can handle…\n","protected":false},"author":2,"featured_media":79901,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[83],"tags":[42534,38498,2439,42535,38597,131,39940,132],"class_list":{"0":"post-79900","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-stockholm","8":"tag-ai-cafe-manager","9":"tag-andon-labs","10":"tag-gemini","11":"tag-hospitality-agent-software","12":"tag-mona","13":"tag-stockholm","14":"tag-stockholm-ai-cafe","15":"tag-sweden"},"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@dk\/116557036044998098","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/posts\/79900","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/comments?post=79900"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/posts\/79900\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/media\/79901"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/media?parent=79900"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/categories?post=79900"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/dk\/wp-json\/wp\/v2\/tags?post=79900"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}