The next challenge is not AI adoption. It is workflow translation.
A lot of discussion about AI still sits at the tool level: which model, which assistant, which platform, which licence. That matters, up to a point. But it is not the hardest part.
The harder part is translating real work into something structured enough to be understood, governed, improved, and, where it makes sense, supported by machines.
That is a bigger shift than people sometimes admit.
Why most workflows are still more informal than they look
Most organisations still depend on workflows that are only partly formal. On paper, there is a process. In practice, there is usually a mixture of habit, judgement, workarounds, local knowledge, and quiet improvisation.
People know who to ask. They know which spreadsheet is the real one. They know where the workaround lives, which step can be skipped, and when the written process stops being useful because the actual process sits in somebody’s head. In plenty of places, Excel is not just a tool. It is doing a fair bit of the architectural heavy lifting.
People can work around that. Machines cannot. And, to be fair, neither can good governance.
Where AI ambitions start to slow down
That is why AI efforts often begin quite well. There is curiosity, some access to tools, maybe a few early experiments with assistants or chatbots. The drag tends to appear when organisations try to connect that interest to real operational workflows.
At that point, the real issue comes into view.
It is not simply whether AI can support the workflow reliably. It is whether the workflow is legible at all. Once you look properly, the mess usually becomes harder to ignore: inconsistent inputs, unclear decision points, fragile handoffs, undocumented exceptions, local shortcuts, and no single shared view of how the work actually functions.
That is where a lot of the friction sits.
The real shift now underway
The transition is not just from not using AI to using AI. It is from informal, human-held workflow to structured workflow that can stand up to scrutiny, support automation where it is sensible, and meet basic expectations of control.
That is a more demanding piece of work than adopting a tool.
It means making tacit work visible. It means being clearer about decisions, dependencies, exceptions, risks, and ownership. It means turning “this is roughly how we do it” into something people can actually explain, test, govern, and improve without relying on whoever happens to know how it works this week.
Why this matters even more in regulated environments
In regulated organisations, this matters more, because workflows are not just about efficiency. They are often tied directly to accountability, auditability, safety, and trust.
A workflow that is too messy to explain is usually too messy to automate well. Quite often, it is too messy to govern well too.
That is where a good deal of the real work will sit over the next decade: helping organisations move from informal practice to something more explicit, more resilient, and more defensible. Not for the sake of neatness, but because that is the condition for improving it with control.
The next challenge is not AI adoption. It is workflow translation.