European agentic-AI startups pass last year's funding total with 7.9bn euros raised

Sifted's two totals are 0.9bn euros apart ^[1]. This year's count is already about 13 per cent above all of last year's ^[2]. It is a year-to-date figure, so the excess arrived before the year was out ^[1].

To test whether the money is outrunning use, you need a second series: a count of agents doing work inside customer companies. The Sifted article does not include one. It offers the vendors instead. Rezonant, cofounded by Emma Burrows, builds each client a tailored 'context graph' that maps scattered information into business categories such as sales, operations or finance ^[4]. Tavily sells an API connecting agents to web information ^[9]. N8n connects LLMs, data sources and business tools ^[7]. Each person quoted works at a company that sells the layer they describe ^[3].

They disagree about order. Against Rezonant's data-first pitch, Lindsay Keim, N8n's vice-president of customer success, said platforms like hers let users "pull data points out of a legacy system, run them through a separate AI model for analysis and move them into a newer system, without actually modernising the whole stack first" ^[6]. The article itself says many organisations mistakenly believe they must modernise their entire tech stack before using AI ^[11]. Those companies are spending on the stack and not yet on agents. Keim's warning is that many of them never start ^[5].

Three outcomes fit what the vendors say. If Rezonant's order holds, early customer budgets go to data-organising vendors, and application startups such as Legora, Tandem Health and Cleo ^[2] sell to customers who are still doing that work. If Keim's holds, agents run on top of old systems and use catches up with funding sooner. In the third, the real limit is permission. "The permissions context is the most difficult and important part of what we do," Burrows said ^[12]. Agents that start out relying heavily on human feedback can "over time, learn routing patterns," she said ^[13].

I think the third is closest to the evidence, because it applies under either of the first two. Staff still review what an agent may touch, whether the data was organised first or piped through a workflow tool ^[14]. On that view the 7.9bn euros ^[1] mostly measures how many companies have been funded to sell agent layers, and customer use grows only as fast as staff approve permissions. The view is wrong if the application companies report customer revenue rising faster than the money they raise. It is also wrong if Sifted's next count shows most funding going to agents that customers already run without a separate data project.