One of the mistakes I think we made early was assuming that trust would emerge naturally if the underlying technology became good enough.

It doesn't.

Trust is partly a property of the technology, but it is also a property of the product surrounding it. People need to understand what the system is doing. They need to know where its boundaries are. They need ways to intervene. They need to recover when something goes wrong. Most importantly, they need to develop an accurate mental model of what the system can and cannot be trusted to do.

That meant some of our best product decisions had very little to do with intelligence itself. We spent time thinking about interfaces, explanations, controls, review mechanisms, and how information was presented. We worked with early customers closely because they exposed problems we could not find in a laboratory.

A system could be technically impressive and still create too much friction. It could produce the correct answer but present it in a way that made the user uncertain. It could automate a task but remove too much control.

Those were not secondary design considerations. They determined whether the product would actually be adopted.

That was another lesson from building companies: the technology creates the possibility, but the product determines whether that possibility becomes useful.