What AI has changed
Until recently, automating meant RPA: software that imitated a person’s
clicks on a screen. It worked, but it was fragile — all it took was the supplier moving one
button for everything to stop working.
It also had a hard limit: it only worked for tidy data. An email written by
a person, a delivery note photographed on a phone or an order in free format were left out,
and that is precisely where most of the time goes in a small business.
That is what has changed. Now the untidy can be automated too: reading an
email and understanding what it asks for, pulling the data out of a scanned document,
writing a draft reply.
The change shows in how people search: queries about AI agents have multiplied by
more than fifteen in the last year, while those about RPA have fallen to a third.
It is not a headline fashion: it is that it now works for cases where it did not before.
With judgement, though. AI is used where something has to be interpreted;
the rest of the process runs on direct connections between systems, which are cheaper,
faster and do not fail. You can see how I apply it to the first contact with the customer on
the AI chatbot for business page.