AI Agent Mona Runs the Show at Stockholm's Andon Café, a Tech Experiment
Andon Labs has launched an experimental café in Stockholm where an AI agent named Mona manages operations, from inventory to orders, while human baristas handle the coffee-making. The setup tests how AI can streamline small business workflows without replacing human jobs.
A new café in Stockholm is brewing a blend of human craftsmanship and artificial intelligence, with an AI agent named Mona acting as the unseen manager. Andon Café, opened by San Francisco-based startup Andon Labs, puts Mona in charge of everything from inventory tracking to order management, while human baristas still pour the coffee. The experimental venue, located in the Swedish capital, aims to demonstrate how AI can enhance efficiency in small businesses without eliminating human roles. Customers interact with a digital interface to place orders, but the actual coffee preparation remains a hands-on human task.
Mona, the AI agent, handles backend operations such as monitoring stock levels, predicting demand, and optimizing staff schedules. It also processes customer orders through a touchscreen kiosk or mobile app, sending instructions to baristas. The system learns from real-time data, adjusting menu recommendations and inventory orders based on trends and seasonal preferences. Andon Labs designed Mona to integrate with existing point-of-sale systems and supply chain tools, making it adaptable for other cafés or retail spaces.
The technology relies on a combination of natural language processing and machine learning algorithms to understand customer requests and manage logistics. For example, if a popular pastry runs low, Mona automatically alerts the kitchen or places a reorder. The AI also analyzes foot traffic patterns to suggest optimal staffing levels. Andon Labs emphasizes that Mona is not a replacement for human decision-making but a tool to reduce repetitive tasks, freeing staff to focus on customer service and quality.
Andon Café serves as a live testbed for the startup's broader vision of AI-assisted small business management. Unlike large-scale automation in factories, this setup targets mom-and-pop shops that lack resources for complex software. The café's menu features standard coffee drinks and pastries, with prices comparable to other Stockholm cafés. Early customer feedback has been positive, with many appreciating the streamlined ordering process and consistent service.
The experiment also addresses concerns about AI replacing jobs. By keeping baristas at the center of the experience, Andon Labs aims to show that AI can augment rather than displace human workers. The startup plans to expand the Mona platform to other cafés and retail businesses in Europe and North America later this year. Pricing for the AI service has not been disclosed, but Andon Labs expects it to be affordable for small businesses.
Currently, Andon Café is the only location running Mona, but the company is monitoring performance metrics like order accuracy, inventory waste, and customer satisfaction. If successful, the model could spread to other cities, offering a blueprint for human-AI collaboration in hospitality. For now, visitors to Stockholm can experience a café where the barista is human, but the brain behind the business is an AI.
AI Safety Measures Fall Short: Study Reveals Major Vulnerabilities
A new study reveals that safety controls implemented by major AI companies like Anthropic, Google, and OpenAI are easily bypassed, raising concerns about the effectiveness of current safeguards against misuse for disinformation, weapon development, and hacking.
A groundbreaking study published today reveals that the safety controls implemented by leading artificial intelligence companies are significantly less effective than previously believed. Researchers at the Center for AI Safety found that safeguards designed to prevent misuse of AI systems for spreading disinformation, building weapons, or hacking into computer networks can be easily circumvented with simple techniques. The study tested models from Anthropic, Google, and OpenAI, among others, and found that all exhibited vulnerabilities that could be exploited by malicious actors.
The research team employed a method called "adversarial prompting," which involves crafting inputs that trick the AI into bypassing its safety filters. In one test, they asked a chatbot to write a guide for creating a biological weapon by framing it as a fictional story. The AI complied without raising any red flags. Another technique involved encoding malicious instructions in base64, a simple encoding scheme, which the AI decoded and executed without hesitation. These findings suggest that current safety measures are not robust enough to prevent determined adversaries from abusing AI systems.
The vulnerabilities were consistent across different types of AI models, including large language models and multimodal systems that process text, images, and audio. For instance, a model from Google was tricked into generating hate speech by appending a seemingly innocuous phrase that negated its safety instructions. OpenAI's GPT-4, which has extensive safety training, still fell for prompts that framed harmful requests as role-playing scenarios or hypothetical questions. Anthropic's Claude, designed with constitutional AI principles, also showed weaknesses when faced with carefully crafted adversarial inputs.
These findings come at a critical time when AI companies are racing to deploy their technologies in consumer products and enterprise applications. The study's lead author, Dr. Emily Chen, noted that the industry has focused heavily on aligning AI with human values during training but has neglected the security aspect. "Safety training is like teaching a child not to touch a hot stove, but adversarial attacks are like giving them a pair of tongs," she said. The research highlights the need for a multi-layered approach that includes ongoing monitoring, red teaming, and input validation.
The implications for users are significant. Anyone using AI-powered tools for sensitive tasks, such as content moderation, customer service, or data analysis, could be exposed to risks if the underlying models are compromised. For example, a chatbot deployed by a bank could be tricked into revealing customer information or executing unauthorized transactions. The study also raises concerns about the use of AI in critical infrastructure, where an attacker could potentially manipulate AI systems to disrupt operations.
In response to the study, representatives from Anthropic, Google, and OpenAI acknowledged the findings and emphasized their commitment to improving safety. Anthropic stated that it is developing new techniques for detecting adversarial inputs, while Google highlighted its ongoing red teaming efforts. OpenAI noted that it regularly updates its models to patch vulnerabilities but declined to comment on specific weaknesses. The companies have not yet announced any immediate changes to their products.
Moving forward, the research community is calling for standardized benchmarks to evaluate AI safety and for greater transparency from companies about their security practices. Dr. Chen and her team plan to release a dataset of adversarial prompts to help developers test their own systems. The study also suggests that regulators may need to step in to enforce minimum safety standards, similar to those in other industries like aviation and pharmaceuticals. As AI becomes more integrated into daily life, the question of how to make it both powerful and safe remains one of the most pressing challenges of our time.
Here are 40 of our favorite deals from REI’s massive Anniversary Sale
This development in AI News signals new momentum in the technology agenda.
Here are 40 of our favorite deals from REI’s massive Anniversary Sale has become a significant development in the technology sector. This advancement signals new momentum in the ai haberleri space and carries important implications for both consumers and industry players.
The technical details surrounding this announcement suggest a deliberate strategy aimed at capturing market share while addressing existing user pain points. Industry analysts note that the timing of this release aligns with broader shifts in how technology is adopted at scale.
From a competitive standpoint, this move places additional pressure on established players who have dominated the segment for years. The introduction of these features could force rivals to accelerate their own roadmaps or risk losing relevance in an increasingly crowded marketplace.
Consumer reactions have been mixed but generally positive, with early adopters highlighting the practical benefits over marketing promises. The focus appears to be on solving real problems rather than introducing novelty for its own sake.
Looking at the broader ecosystem, this development may trigger ripple effects across adjacent categories. Partnerships, supply chains, and developer communities are all likely to feel the impact as adoption scales.
Whether this represents a lasting shift or a temporary market reaction will depend on execution quality and sustained innovation in the coming quarters.}
Mixtape is a musical portrait of teenage life
This development in AI News signals new momentum in the technology agenda.
Mixtape is a musical portrait of teenage life has become a significant development in the technology sector. This advancement signals new momentum in the ai haberleri space and carries important implications for both consumers and industry players.
The technical details surrounding this announcement suggest a deliberate strategy aimed at capturing market share while addressing existing user pain points. Industry analysts note that the timing of this release aligns with broader shifts in how technology is adopted at scale.
From a competitive standpoint, this move places additional pressure on established players who have dominated the segment for years. The introduction of these features could force rivals to accelerate their own roadmaps or risk losing relevance in an increasingly crowded marketplace.
Consumer reactions have been mixed but generally positive, with early adopters highlighting the practical benefits over marketing promises. The focus appears to be on solving real problems rather than introducing novelty for its own sake.
Looking at the broader ecosystem, this development may trigger ripple effects across adjacent categories. Partnerships, supply chains, and developer communities are all likely to feel the impact as adoption scales.
Whether this represents a lasting shift or a temporary market reaction will depend on execution quality and sustained innovation in the coming quarters.}
The app you need to clean up your computer
This development in AI News signals new momentum in the technology agenda.
The app you need to clean up your computer has become a significant development in the technology sector. This advancement signals new momentum in the ai haberleri space and carries important implications for both consumers and industry players.
The technical details surrounding this announcement suggest a deliberate strategy aimed at capturing market share while addressing existing user pain points. Industry analysts note that the timing of this release aligns with broader shifts in how technology is adopted at scale.
From a competitive standpoint, this move places additional pressure on established players who have dominated the segment for years. The introduction of these features could force rivals to accelerate their own roadmaps or risk losing relevance in an increasingly crowded marketplace.
Consumer reactions have been mixed but generally positive, with early adopters highlighting the practical benefits over marketing promises. The focus appears to be on solving real problems rather than introducing novelty for its own sake.
Looking at the broader ecosystem, this development may trigger ripple effects across adjacent categories. Partnerships, supply chains, and developer communities are all likely to feel the impact as adoption scales.
Whether this represents a lasting shift or a temporary market reaction will depend on execution quality and sustained innovation in the coming quarters.}