In many tax teams, we see the same pattern emerging. The tools are there, Copilot, agents, experiments, but usage is lagging behind. A few enthusiastic colleagues are already building things, but this is mainly causing a proliferation: agents who no one knows who maintains them, differ in quality and no shared way of working.

In addition, tax teams indicate that AI is their number one priority, while 61% have neither the structure nor the basis to work with it effectively. Our survey of 1,600 organizations shows why: there is little trust in the technology, because results sometimes differ and people don't understand how AI comes to answers. In addition, important data is not yet available to agents, so the promising use cases remain out of reach.

The next thing: employees are afraid of an impact on their roles, causing resistance if AI is “pushed in” from above. Teams often focus on the wrong use cases. They are too complex, too big, or simply not suitable for AI. So the problem isn't that tax professionals don't want to. The problem is that they don't know where to start, what works, and how to do it safely and scalably. And that's exactly what the AI Agent Workshops help with.

The workshops are practical, structured and aimed at tax teams that want to take a first step or want to avoid getting stuck in separate experiments. Let's start with the basics:

• Which use cases are really valuable?
• Which ones aren't and why?
• What can you already build without complicated data projects?

After that, we bring everyone to the same level of knowledge. Not with technical details, but by exploring together what agents do, how they think, and what you need to do it. This creates a shared understanding, which is crucial for trust and ownership.

We show examples of agents who are already working in other tax teams. We provide participants with workable prompt templates, clear guidelines for connecting data and knowledge, and practical tips for getting better results from AI. During the workshop, each participant works on their own use case, prepared with “homework” in advance, including data (real or synthetic), so that it can actually be built. That's how AI becomes something you do, not something you brainstorm about.

‍