What I notice, question, and learn.

The systems I trust most still have a person in the loop. On a submarine that person could stop a procedure. On a railway it should be the operator who decides whether a warning is a defect, a false alarm, or something that can wait until the next possession.

I do not treat that as nostalgia for analogue work. Models are useful. They are also confident, incomplete, and silent about the cost of being wrong. The job is to make the uncertainty visible, name who decides, and leave a trail that another professional can inspect.

That is the thread from radiation safety, through enterprise GenAI, into TrackSense and the advisory work I do now. If a product cannot say who is accountable when the answer is uncertain, I do not think it is ready.