14/10/2026
Last week, Nelu, our Chief Innovation Officer, spent two days at World Summit AI in Amsterdam. He wanted to find out what actually gets AI projects from pilot into production, and the answer had less to do with the models than he expected. Especially in banking and insurance, what mattered most was being able to show that an agent did the right thing, and that a person stepped in when it counted.
That is exactly the road we have been walking. Specs are written before an agent touches the code, our engineers hold the gate at the merge, and every rejected change teaches our knowledge base something.
But a good conference should also challenge you, and this one did. Nelu brought home three questions:
๐ท How do we check the AI in our products as carefully as the AI we build with?
๐ท What does it really cost to get one useful result?
๐ท And what changes when the next user of our APIs is someoneโs agent rather than a person?