Last year, the focus was still on AI as a copilot, a smart assistant that thinks along with you. This year, the message on the IncentroCon stage is different. Agentic AI doesn’t just think along, it takes action. Maurits Berger and Kelian Mos of iPaaS platform Workato explain what that requires from your infrastructure.
The room is full of people already working with AI. When Kelian asks who uses AI tools, hands go up everywhere. ChatGPT, Copilot, Claude, most attendees already have experience with them. Yet the real value often remains limited, Maurits quickly points out. According to him, there is a proliferation of AI initiatives with little return.
“Everyone is doing it, but the return is barely there.” He argues that the issue is not the technology itself, but how organizations approach it. Many applications remain stuck in experiments or isolated use cases, without clear objectives or a connection to the business.
From talking to doing
When it comes to impact, the two speakers highlight a clear difference between ‘regular’ AI and agentic AI. A chatbot can retrieve information or formulate an answer. An agent goes a step further and actually performs actions. That sounds simple, but in practice it turns out to be a difficult step. Kelian illustrates this with a simple example.
“Ask an AI tool to pull your top customers and plan a route. Chances are you still have to dive into other systems to gather the data. You’re still copy-pasting. How much time have you really saved?”
The real bottleneck is in infra
According to them, the issue does not lie in AI itself, but in the infrastructure around it. AI works well as long as you stay within a single ecosystem. But as soon as you start using multiple systems, such as ERP, CRM or local software, things become more complex. Integrations are missing, security concerns arise and governance becomes an issue.
Maurits: “Would you allow ChatGPT to update your SAP system? Over my dead body, most people would say.” This highlights the core tension. Organizations want to adopt AI, but at the same time keep their core systems tightly protected.
A brain without hands and feet
Kelian uses a clear metaphor to explain the problem. A language model is the brain. It can think, analyze and reason. But without access to systems, it cannot execute anything. “It’s like having a brain without hands and feet.”
Only when AI is connected to the systems where processes and data come together can it truly act. It can then not only provide answers, but also trigger actions, execute workflows and automate processes.
The missing link between AI and systems
The rest of the session focuses on that missing link. How do you give AI safe and controlled access to your business processes? Workato addresses this with a middleware layer, the Model Context Protocol, which acts as the connection between AI and enterprise systems. A kind of digital glue that enables agents not just to think, but also to act.
This layer not only handles integrations, but also takes care of security, governance and control. Who is allowed to do what, which data is used and how actions are logged becomes critical. Once agents start acting autonomously, you need full visibility into what is happening and on whose behalf.
Autonomy requires choices
With the right infrastructure in place, the level of autonomy required for agentic AI suddenly becomes achievable. But that does not mean everything should be automated. Through concrete examples, Kelian shows that autonomy does not mean AI can do anything it wants. “Through that layer, you can strictly control which actions an agent is allowed to perform. Retrieving data might be fine, but modifying an order or sending an email only under the right conditions.”
Accountability also plays a key role. You don’t want it to say somewhere that AI did it, you want a person’s name attached to it. It is not about maximum autonomy, but about finding the right balance.
The step that is needed now
A key message from the session is that many organizations are still in the experimental phase, while the next step is already required. Not another tool or use case, but a focus on fundamentals. How do you connect AI to your existing systems? How do you handle governance and security? And how do you ensure that AI not only advises, but actually delivers value in processes?
Maurits concludes: “Think strategically and prevent AI from becoming a collection of isolated experiments.” Only when AI can truly act within your organization, it will evolve from a helpful assistant into a digital colleague.
* This blog is based on a presentation by Maurits Berger and Kelian Mos of Workato at IncentroCon Agentic ’26.
4 min read
Will AI replace all of us? Philosopher Bas Haring on the role of AI agents
3 min read
Mathijs Bouman: is AI a lifeline for our aging population?
3 min read
Scaling AI agents without making a mess: with Agent Orchestration
4 min read
Deploying AI where it adds value? It's all about the what, the how, and the why.
