Not everything that impressed in a demo survived contact with real work.
Some promises looked huge until simple questions came up.
What information does it rely on?
What happens when the case is not what it expected?
Who reviews the result?
How does it fit into the operation already in place?
What does it cost to maintain after the demo?
What evidence remains when it gets something wrong?
The hype is rarely that the technology itself is fake.
It is presenting a partial capability as if it solved the whole problem.
A fast response is not a finished process.
A prototype is not an operation.
Generated code is not software ready to support a business.
And a productivity figure without a before, an after, and a shared definition is just an attractive number.
This year confirmed something I already knew: a tool deserves attention when it makes it past the demo and finds a clear place in real work.
Artificial intelligence has greatly expanded what we can try.
It has also made it more important to distinguish between an interesting possibility and a capability we can actually operate.