Step 2 of 5 · Wave 1 to wave 2
Why scripted bots break, and what replaces them
A scripted bot knows where to click, not what it is looking at. That is its strength and its limit. It runs a process exactly as recorded, and it has no way to notice that the process around it has changed.
So bots break in ordinary ways. A supplier sends a different invoice layout. A web page moves a button. A policy adds a step. Each change means a person rewrites the script, tests it and puts it back, and a growing estate of bots becomes a maintenance team of its own.
The third wave changes the material. A language model reads. It can take a document it has not seen before, pull out what matters and write a reply in plain language. Work a script could not touch, because it needed understanding rather than clicks, comes within reach.
The first form of it most people meet is the chat assistant: a general model you ask, which drafts, summarises and answers. It does not break when a layout changes, because it was not following a layout in the first place.
The trade is that it waits to be asked. An assistant has no place in your process until a person opens it, pastes in the context and copies the answer back out. The work is faster, but a person still carries it from step to step. That is the gap the next wave tries to close.
Promenaut
