From AI Experiments to Agentic Workflows in Eindhoven
More than 30 founders, product leaders, engineering leads, and AI practitioners came to our first One Horizon event yesterday. They did not need another demo proving that agents can produce output. They wanted to know what it takes to trust them with real work.
That was the useful part of From AI experiments to agentic workflows.
It was the first event we hosted as One Horizon, together with the Stripe Eindhoven Community, at the AI Innovation Center on High Tech Campus Eindhoven.
The subject looked technical on the program. The room made it much more practical: how do you make an agent dependable, where should people stay involved, and how do you turn one useful demo into work a company can repeat?
The gap is no longer access to AI
Stefan Samba of Triagen and I shared two different views of the same shift.
Teams can already get useful results from individual agents. A developer can hand one a coding task. A product manager can generate a prototype. A founder can automate a repetitive process.
Then the experiment meets the company.
The agent needs the right context. It needs permission to use specific tools and data. Someone needs to own the outcome. Another person may need to review it. If the work pauses, fails, or changes hands, the next person or agent must be able to see what happened.
That is where many experiments stall.
Autonomy is an operating and design problem
Choosing a stronger model helps. It does not answer the operational questions around the model.
What information can the agent use? Which actions can it take without approval? Who decides whether the result is good enough? Where are decisions and artifacts stored? How does someone step in when the workflow reaches a judgment call?
These are not edge cases. They are the difference between an AI interaction and a workflow a team can trust.
The panel discussion with Tijn van Daelen, Murat Özmerd, and Eline Schoonen of Be-Inc moved into a different question: How can we keep AI human when it becomes autonomous?
Eline brought the discussion back to responsibility. The EU AI Act sets transparency duties for providers and deployers in defined situations, including certain interactive and generative AI systems. Her point went beyond compliance: a brand or company cannot hand responsibility to the model. If AI acts under your name, you still own how people experience it and what its decisions do.
That makes transparency a UX and UI design requirement. As AI systems make more decisions and take more actions, interfaces need to show people when AI is involved, what it is doing, and when human judgment is needed. Autonomy should not make the work feel less understandable or remove the moments where a person needs to take control.
Bias is part of the same responsibility. Models inherit patterns and biases from their training data and design. Teams need to identify and reduce harmful bias, make limitations visible, and help people recognize when AI may be shaping a decision.
The audience brought practical perspectives to that discussion. The challenge is not only to make autonomous systems capable. It is to design the relationship between those systems and the people affected by their decisions.
The demo made the argument concrete
Tijn also premiered our new One Horizon product demo.
It shows how an idea can move through a governed workflow toward a release, with agents and people working from the same context. That shared record matters more as execution becomes distributed across models, tools, and teams.
A workflow should survive a model switch, an agent handoff, or a human review without losing why decisions were made.
You can watch the complete One Horizon demo.
This was the first one
Thank you to everyone who joined, asked direct questions, and shared what is and is not working inside their companies.
Thank you to Stripe and Mathew Krawczyk for working with us and sponsoring the drinks and snacks. And thank you to the AI Innovation Center, High Tech Campus Eindhoven, and Steven Feenstra for giving us the space to bring everyone together.
One afternoon did not answer every question. It gave us enough material for the next one.
We plan to host more of them.
Follow the One Horizon LinkedIn page if you want to hear about the next event.
