Thoughts / 01
Most people call me about AI. Here is the honest version.
The call usually opens the same way. Somebody has read that AI is going to change everything, their competitor has apparently started using it, and they would like to know what they should be doing.
Here is what I actually think, after having this conversation a lot of times.
AI is genuinely good at two things your business does
The first is producing a first draft. A listing description, a follow-up email, a summary of a long document, the first pass at a job posting. Work where a blank page is the expensive part and editing is cheap. AI is good at this because being roughly right quickly is exactly what a first draft needs to be.
The second is reading messy input. A stack of supplier invoices in fourteen different layouts. Forwarded emails where the information you need is buried in a paragraph. Scanned forms someone filled in by hand. Older software needed the input to be perfectly consistent before it could do anything useful. This is the part that has genuinely changed, and it is less exciting than the headlines, and it matters more.
And bad at one thing that matters
It cannot be trusted without someone checking. Not occasionally. Every time, for anything that carries consequences.
That single fact decides where AI belongs in a business. It can remove typing. It cannot remove responsibility. So it fits work where a person was always going to review the output anyway, and it does not fit work where nobody was going to look.
This is why I am wary when someone wants AI to handle something end to end with no human in the loop. Not because it will not produce an answer. Because it will produce a confident one, and the first time it is wrong will be the time nobody was looking.
The trap
Here is the part that surprises people. The work costing you the most is usually not a thinking problem at all.
It is that the same eight facts get typed onto four different forms. It is that a total gets copied from one system into another by hand every Tuesday. It is that somebody rebuilds the same report from scratch every month because the numbers live in three places.
None of that needs intelligence. It needs software that does the same thing the same way every time. That is a smaller, cheaper, duller project than the one people call about, and it is usually the one that gives them their week back.
A test you can apply yourself
Take the task that annoys you most and ask: does this require judgment, or just accuracy?
If it requires judgment, a person keeps doing it, and AI might make their first draft faster. If it only requires accuracy, a person should probably stop doing it, and what replaces them is ordinary software rather than anything clever.
Most of the tasks people describe to me are the second kind. They came in asking about AI and left with something that never mentions it.
Recognise your own week in that?
Bring the task. Thirty minutes, no cost, and a straight answer about whether it is worth building anything.
Bernot Software