Don't Run Blind, Don't Sit Still: Lessons from Our Tax Authority Research on Agentic AI

In recent months, commissioned by the Corporate Directorate for Innovation and Strategy of the Dutch Tax and Customs Administration (Belastingdienst), we interviewed eleven leading experts on the question: what does agentic AI mean for an organization that processes millions of tax returns annually, handles millions of phone calls, and executes complex regulations?

The result is a report we are publishing today: "Agentic AI: The Outside World Speaks." Not a blueprint, but a compass. Below are the most important insights.

The experts at the table

We deliberately chose diversity. The panel included Durk Kingma (researcher at Anthropic, formerly OpenAI and Google DeepMind), Robert Engels (Head of Gen AI Lab at Capgemini), Deborah Nas (Professor at TU Delft), Sanne Manders (President of Flexport), Winifred Andriessen (VP AI Excellence at KPN), Geert-Jan van der Snoek (CEO Sdu Lefebvre), Jorissa Neutelings (CDO ABN AMRO), Jeroen van Glabbeek (CEO CM.com), Marijn Pijnenborg (co-founder Funda), Bas Haring (philosopher), and Douwe Groenevelt (founder Viridea, former ASML).

Technologists, scientists, executives, and thinkers. That combination proved essential to look beyond the hype.

The common thread: the hype is real, and so is the opportunity

One thing stood out in almost every conversation: the warning not to succumb to the marketing machine surrounding AI. Deborah Nas was the most outspoken; in practice, she sees hardly any organizations where agents are truly operational in critical processes. Robert Engels referred to it as the "Bermuda Triangle" of agentic AI the tension between autonomy, agency, and authority. You communicate in natural language with a system that is fundamentally probabilistic. That is fine for brainstorming, but risky for processes with legal consequences.

At the same time, Jeroen van Glabbeek provided nuance: the hallucination problem from a few years ago has largely been solved. At CM.com, 85% of helpdesk questions are handled fully autonomously with higher customer satisfaction than before.

The art lies in the middle ground: neither naive enthusiasm nor paralyzed skepticism, but informed experimentation.

From tools to processes: where the real value lies

The most concrete example came from Sanne Manders at Flexport, which has achieved 51% automation in its core operations. He described three layers of automation: classic software engineering, the "Excel stack" (work that was too variable for traditional automation), and a new third layer where you tackle that middle layer using LLMs and low-code.

What the Tax Authority can learn from this: the greatest value is not in chatbots, but in automating processes that currently run manually when they don't have to. Flexport now audits 100% of all customs transactions instead of a random sample of a few percent. That fundamentally changes what you can know as an organization.

The human doesn't disappear, but the role changes

Philosopher Bas Haring shared the best example. A man at the municipality of Tilburg, over sixty, used to help people write reports. He doesn't do that anymore AI does it better. But he has just as much work. Why? Because people want to drink coffee with him. They want a human across from them who listens, thinks along, and provides support. The same amount of work, but a very different kind of work.

This is the core question for every organization: what do you do with the freed-up capacity? The report argues for using it not for more efficiency, but for better service. The citizen who gets stuck, the entrepreneur with a complex situation, the bereaved person who cannot find their way that is where you need people.

The agent as customer: a paradigm shift

Jorissa Neutelings of ABN AMRO sketched perhaps the most far-reaching scenario. What if the citizen no longer visits the Tax Authority website themselves, but sends their own AI assistant? She introduced the concept of the "liquid enterprise": an organization that breaks its services into small, connectable blocks that can be queried by both humans and machines.

For the Tax Authority, this means: you must not only think about how AI improves your processes, but also about how you handle a world in which the citizen sends an agent instead of calling themselves.

Governance as a foundation, not a brake

The lesson from the childcare benefits scandal (toeslagenaffaire) resonated through several interviews. Every AI implementation must be designed from day one with logging, audit trails, and explainability. Not as an afterthought, but as a foundation. Geert-Jan van der Snoek of Sdu Lefebvre structurally spends 5 to 6 percent of all development time on reliability and governance.

Robert Engels summarized it poignantly: you don't build trust with technology alone, but with transparency, explainability, and the ability to correct errors.

Seven recommendations

The report concludes with seven concrete recommendations:

1. Start with the end vision, not the technology. What does the ideal taxpayer experience look like?
2. Start with low-risk experiments to learn. Not primarily to deliver production value, but to understand.
3. Invest substantially in AI literacy. Not as a one-off course, but as an ongoing program at all levels.
4. Design governance from day one. Logging, audit trails, and explainability as a requirement for every pilot.
5. Keep humans central where it matters. Use freed-up capacity for better service delivery.
6. Remain flexible in technology choices. The toolset changes rapidly avoid vendor lock-in.
7. Create space for innovation "on the side." A separate team with the freedom to work radically differently.

The conclusion

The eleven experts agreed wholeheartedly on two things. Sitting still is not an option technology is developing, the private sector is embracing it, and citizens will expect comparable service levels. But running blind is equally unwise. Success goes not to those who are first, but to those who learn best.

Or, in the words of Bas Haring: "Fiddle around a bit. But fiddle with attention, with reflection, and with the larger goal in mind."

The full report "Agentic AI: The Outside World Speaks" was written on behalf of the Corporate Directorate for Innovation and Strategy of the Tax and Customs Administration by Remy Gieling and Job van den Berg of The AI Group (ai.nl). The report can be downloaded here.

Ready to get started with AI yourself? Check out our e-learning AI Agents or AI workshops for teams.

// About the author

Remy Gieling

Mede-oprichter, AI-expert & bestseller-auteur

Tech-expert (1988) gespecialiseerd in kunstmatige intelligentie en mede-oprichter van ai.nl, The Automation Group, Proxies en eBrain.ai. Oud-hoofdredacteur van diverse zakenmerken en daardoor een geoefend verteller op het podium en in de media. Verzorgt jaarlijks 150+ AI-keynotes in binnen- en buitenland en is gastdocent aan Nyenrode. Co-auteur van zeven boeken, waaronder 'Handboek AI Strategie' en 'AI Agents', en bekend als presentator op radio en RTL Z. Reist langs de labs van OpenAI, Nvidia en Tencent en vertaalt de nieuwste doorbraken naar inzichten die leiders direct kunnen toepassen.

LinkedIn