The first article I wrote about applying AI to organisations dates to 2017 - nearly 10 years ago.
When I was asked to speak on the topic back then, it still landed as science fiction. The room was polite and curious, and quietly sceptical. The practitioners were a small group, and most of us had come to it sideways -developers, physicists, and philosophers.
Nearly a decade later, here's what changed and what didn't.
What changed:
β The same year I wrote that article, a Google paper introduced the Transformer. Almost nobody outside research noticed. It became the foundation of nearly everything we now call AI.
β In 2017, "doing AI" meant a narrow model, a labelled dataset, a data science team and months of work. Today a general-purpose model can draft, code, analyse documents and carry out multi-step tasks from a plain-language instruction.
β The main question has moved from "can it?" to "should it, and who is accountable?" An international management-system standard for AI (ISO/IEC 42001) now exists. In 2017, that would have sounded as futuristic as the technology itself.
What didn't change:
β Bad data still produces confident nonsense, just more fluently.
β Problem framing still decides most of the value.
β Adoption is still a people and process problem more than a technology problem.
β The organisations winning are still the ones that treat AI as a capability to govern, not a tool to buy.
The science-fiction part is over. The hard part (trust, accountability, redesigning work) is just beginning.