October 15, 2025
The World AI Summit in Amsterdam brought together thousands of AI practitioners, researchers, and industry leaders — yet one of the biggest surprises was what wasn’t being discussed on the first day: AI agents.
Despite the dominance of vendors showcasing platforms (from the likes of AWS’s AgentCore to a host of emerging providers), there was little conversation on the first day about real-world usage. By the second day, however, that began to shift — and the conversations revealed a lot about where applied AI stands today, and what it takes to turn intelligence into action.
Jeroen Van Genuchten of ING offered one of the clearest definitions of agents I heard at the summit: AI that doesn’t just know things — it does things. That addition of action is what separates a static system from something truly transformative.
Van Genuchten’s talk was packed with practical examples of how ING has applied AI over the years, but what stood out was his framing — that the future of AI lies in pairing intelligence with the ability to act autonomously within defined boundaries. It’s a vision that resonates strongly with what we’ve seen in our own work with autonomous customer service agents and decision-support systems.
Hans-Joachim Belz from E.ON shared some fascinating insights into how agentic AI is being applied in customer service. One point that struck me was how LLMs enabled a more natural, free-flowing interaction between customers and systems — something we’ve also observed in our own LLM-based customer service projects.
He described how these systems could handle unseen challenges — a reminder that one of the greatest strengths of large models is their adaptability.
Belz also underscored a point that can’t be repeated often enough: testing and monitoring are critical. As AI systems become more autonomous, the need for continuous evaluation becomes not just a best practice, but a safeguard.
Speaking of testing, LangWatch gave a great presentation on testing methodologies for AI agents — complete with a familiar friend: the testing pyramid. It’s always refreshing to see traditional engineering disciplines being reinterpreted for AI systems. Their approach to structured testing for LLMs and agentic workflows shows how software engineering best practices are evolving to meet new forms of intelligence.
Alejandro Saucedo’s talk from Zalando was another highlight. Zalando has long been a pioneer in AI-driven retail, and Alejandro covered an impressive range of applications — from personalisation and virtual try-on to trend detection. But perhaps the most important takeaway was his reminder that AI’s real power often lies behind the scenes.
To round out the event, Yunpeng Liu’s talk on how Pokémon Go optimises global gameplay was a brilliant reminder that AI extends well beyond generative models. His team uses AI to schedule Pokémon battles on a global scale — a logistical challenge that combines operations research, data science, and real-time decision-making.
One line in particular stood out: “Don’t just optimise for accuracy.” It’s a mantra every data scientist should live by. Real-world impact comes not from perfect predictions, but from meaningful outcomes.
World AI Summit Amsterdam showed that while agentic AI is still finding its footing in production, the groundwork is being laid. The tools are maturing, testing practices are evolving, and organisations are learning what it takes to safely turn AI from an assistant into an actor.
As always, the most interesting work is happening where intelligence meets action — and where AI doesn’t just answer questions, but helps make better decisions.
All business models have their pros and cons. But, when you consider the type of problems we help our clients to solve at Equal Experts, it’s worth thinking about the level of experience and the best consultancy approach to solve them.
If you’d like to find out more about working with us – get in touch. We’d love to hear from you.
