The rise of AI agents
What European scaleups have learned so far
See how European scaleups got their AI agents past proof-of-concept by fixing the content layer first
See how European scaleups got their AI agents past proof-of-concept by fixing the content layer firstMost AI projects don't stall because the model is wrong. They stall because the content the model needs is scattered across drives, inboxes, and point solutions — ungoverned, and invisible to your agents.
Sifted spoke to founders, CTOs, and CIOs across Europe to find out what actually separates working agents from expensive pilots. The answer is rarely the model. As Box CEO Aaron Levie puts it, "Context is king for AI agents."
The numbers back it up. Fewer than half of AI agents make it past proof-of-concept, according to a Dataiku poll of business leaders, and a widely cited 2025 MIT paper found 95% of generative AI pilots fail to deliver a return on investment. Meanwhile 95% of leaders admit they couldn't fully trace an AI decision end to end if a regulator asked.
The teams getting real value are doing something simpler. At Irish aviation software company Cloudcards, agents handle the extraction of commercial and technical terms across dozens of documents while human experts focus on strategic oversight. At consultancy Yonder, agents surface knowledge that was "sitting in old decks, case studies, and deliverables that was effectively invisible before."
Read the report to learn:
- What separates a real agent from a chatbot or a workflow with a new name
- How lean teams get work done without adding headcount or budget
- Where agents break, and the guardrails European teams put around them
- Why the content layer, not the model, decides whether agents deliver