Anthropic and Dust Join Forces to Revolutionize AI Agents in Europe

Exclusive: Anthropic and French AI firm Dust want to save you time at

The partnership comes as Anthropic expands deeper into Europe. US artificial intelligence (AI) company Anthropic, maker of the Claude large language model (LLM), is partnering with the Paris-based AI agent builder Dust as part of its scaled-up investment in Europe. They announced the joint venture on Thursday, sharing details with Euronews Next in an exclusive for Europe. It's the latest step in the race between technology companies to release agentic AI models, or AI agents, which do not just process information but also try to solve problems, make plans, and complete tasks. They differ from AI chatbots, which are designed for conversations with people and serve more as co-pilots than independent actors. Both Anthropic and Dust's co-founders previously worked at OpenAI, which made ChatGPT. Dust's clients include French tech champions Quonto and Doctolib, which TK. In the new partnership, Dust will help companies create AI agents using Claude and Anthropic's Model Context Protocol (MCP), which is an open standard to connect external data sources with AI tools. It can be thought of as a USB-C port for AI applications. The companies say this will create a central operating system where AI agents can access company knowledge and take action independently of human employees, while still keeping their data secure. "Companies have various AI tools across departments operating in isolation with no ability to communicate with each other," Gabriel Hubert, CEO and co-founder of Dust, told Euronews Next. He said that Dust can overcome this issue so that AI agents can work together on different tasks. AI's growing role at work - with or without people In his own work, Hubert said he uses AI agents to help him write job offers, analyse job applications and customer reviews, and other time-saving tasks. "We've given ourselves the possibility to do something that I wouldn't have the time to do otherwise, and I still sign every offer letter that goes out," he said. However, AI and AI agents are relatively new and still tend to make mistakes. For example, in a recent Anthropic experiment, an AI chatbot charged with running a small shop lost money and fabricated information. "Claudius (the AI shop) was pretty good at some things, like identifying niche suppliers, but pretty bad at other important things, like making a profit. We learned a lot and look forward to the next phase of this experiment," Guillaume Princen, head of Anthropic's Europe, Middle East, and Africa team, told Euronews Next. The project with Dust "comes with a lot of power, [and] it comes with a lot of responsibility," he said. Another "tough problem to crack," Princen said, is determining who is at fault if an AI agent does something wrong: the AI company, or the organisation using the tools? "Understanding who's accountable when an agent does a thing sounds easy on the surface, but gets increasingly blurry," he said. In some cases, AI agents could act as someone's "digital twin," he said, while other times they could act on behalf of a specific person, team, or company. Most companies haven't quite decided where they land. "We tend to work with very fast-moving companies, but still on that one, we're realising that there is some education to do," Princen said.

Anthropic is reportedly partnering with Dust in Europe

According to an exclusive report of Euro News, US artificial intelligence company Anthropic has partnered with Paris-based AI agent builder Dust to expand its European presence. The joint venture focuses on developing AI agents capable of problem-solving and task completion, differentiating them from conversational chatbots. This collaboration marks a strategic move for Anthropic, maker of the Claude large language model (LLM), as it deepens its investment in the European market. Both Anthropic and Dust's co-founders previously held positions at OpenAI. Dust's client roster includes French technology firms Quonto and Doctolib. Under this new partnership, Dust will facilitate the creation of AI agents for companies, leveraging Claude and Anthropic's Model Context Protocol (MCP). The MCP is an open standard designed to link external data sources with AI tools, conceptualized as a universal connector for AI applications. This integration aims to establish a centralized operating system where AI agents can securely access company knowledge and operate autonomously, reducing reliance on human intervention. Gabriel Hubert, CEO and co-founder of Dust, detailed the current challenges faced by organizations. "Companies have various AI tools across departments operating in isolation with no ability to communicate with each other," Hubert stated. He emphasized Dust's role in overcoming this fragmentation, enabling AI agents to collaborate on diverse tasks. Hubert personally utilizes AI agents for tasks such as drafting job offers, analyzing job applications, and processing customer reviews, noting these applications save significant time. "We've given ourselves the possibility to do something that I wouldn't have the time to do otherwise, and I still sign every offer letter that goes out," he explained. Anthropic trashed millions of books to train its AI Despite their capabilities, AI agents are an evolving technology and can exhibit errors. A recent Anthropic experiment involving an AI chatbot managing a small retail operation resulted in financial losses and data fabrication. Guillaume Princen, head of Anthropic's Europe, Middle East, and Africa team, commented on the experiment: "Claudius (the AI shop) was pretty good at some things, like identifying niche suppliers, but pretty bad at other important things, like making a profit. We learned a lot and look forward to the next phase of this experiment." Princen acknowledged the dual nature of the partnership with Dust. "The project with Dust comes with a lot of power, [and] it comes with a lot of responsibility," he stated. A complex issue, according to Princen, is assigning accountability when an AI agent errs. "Understanding who's accountable when an agent does a thing sounds easy on the surface, but gets increasingly blurry," he observed. He elaborated that AI agents might function as a "digital twin" in some contexts or act on behalf of an individual, team, or entire company in others. Princen noted that many companies are still determining their stance on this matter. "We tend to work with very fast-moving companies, but still on that one, we're realising that there is some education to do," he concluded.