AI Agent Manages Experimental Cafe in Stockholm

An experimental cafe in Stockholm is being managed in large part by an AI agent called Mona, according to a blog post from Andon Labs and coverage by AFP, Euronews and other outlets. Per Andon Labs' blog, the startup signed a lease for premises at Norrbackagatan 48, delivered the lease and some starting capital to Mona, and observed the agent execute a prioritized checklist that included food-business registration, supplier contracts and staff recruitment. Reporting by Yahoo/AFP and Euronews documents that Mona posted job listings, conducted phone interviews and selected at least two human baristas to work on site. Several outlets report that Mona is powered in part by Google's Gemini model. Local reporting and staff comments describe operational mistakes, such as unnecessary bulk purchases; AFP quoted barista Kajetan Grzelczak saying, "ordering isn't really her best suit."

Per Andon Labs' public blog post, Mona carried out document analysis (lease and permits), generated prioritized tasks, and interacted with external services (job sites, supplier contact forms, municipal registration pages). Andon Labs published sample task lists and conversational logs showing the agent synthesizing deadlines and vendor data during setup. Industry observers have demonstrated comparable agent patterns in prior experiments: autonomous agents chaining web actions, API calls and human handoffs to complete multi-step workflows. Editorial analysis: agents that combine a large-language backbone like Gemini with scripted tool integrations commonly excel at planning and information synthesis but struggle with real-world procurement nuance and cost optimization, which can produce the kind of redundant purchases reported at the cafe.

Editorial analysis: this deployment is an incremental but visible example of putting AI agents into everyday operational roles that require coordination across bureaucracy, procurement and hiring. It mirrors previous experiments where agents were given budgets and tools to run small businesses; the novelty here is the public, customer-facing retail setting in a European regulatory environment. Reporting highlights ethical and labor questions: Euronews and AFP quoted Andon Labs team members framing the project as an experiment to surface questions about AI employing humans. For practitioners, the case illustrates practical integration points (document parsing, automated hiring workflows, inventory automation) and the friction points (supplier negotiation, regulatory compliance subtleties and procurement mistakes) that surface when agents act outside tightly controlled testbeds.

Editorial analysis: observers should track:

• •post-launch operational data if Andon Labs publishes outcomes such as revenue or inventory loss
• •how the team constrains the agent's financial authority and error-correction mechanisms
• •any regulatory or labor responses from Swedish authorities or worker groups. On-site dashboards shown by reporters appear to display real-time revenue and balances, but neither Andon Labs nor local outlets have published a comprehensive post-mortem on profitability or safety controls to date

Why it matters