01 The trap
The easiest way to apply AI is to point it at the business we already have.
We document the current process, identify its delays and manual activities, and ask where AI could make it faster, cheaper or easier. This is sensible. It can produce real value.
But it also carries a risk: the current process quietly becomes the design brief for the future.
We begin with what exists, improve it and call the result transformation. The business may operate more efficiently, but its underlying architecture—its work, decisions, roles, information flows and organizational boundaries—remains largely unchanged.
We have paved the cow path.
02 The path was shaped by old constraints
A cow path is not deliberately designed. It develops over time in response to terrain, fences, obstacles and repeated behaviour. Paving it makes travel faster, but it does not prove that it is still the best route.
Business processes develop in much the same way.
They are shaped by the technology, information, policies, organizational boundaries and management assumptions that existed when the work was designed. A process may contain a handoff because information could not once move freely. A queue may exist because only one role could make a decision. Work may be divided between functions because coordination was expensive.
The current process is therefore not neutral. It is a record of past constraints and accumulated compromise.
Some of those constraints remain real. Others no longer need to be true.
If we automate the process without understanding what shaped it, we risk embedding the old constraints more deeply. We may become remarkably efficient at operating a business designed for conditions that no longer exist.
The current state should be treated as evidence—not instruction.
03 Automation is useful, but insufficient
This does not mean that every AI transformation must begin with a blank sheet of paper.
That is rarely practical. Organizations need to solve immediate problems, demonstrate value and learn through real use. Automating bounded parts of the current business can be a reasonable place to start.
The problem is not beginning with automation.
The problem is stopping there.
Automation asks how we can perform the existing work better. Transformation must eventually ask a wider question: what work, service or capability could we now design differently?
If AI transformation = business (re)architecture, then automation is one intervention within a much larger redesign. It should create value today while also producing the insight and capability needed to question tomorrow’s business.
The choice is not automation or re-architecture. The opportunity is to use automation as a doorway into re-architecture.
04 Think bigger
As we automate, we should keep widening the frame.
What outcome are we actually trying to produce?
Which constraints shaped the current work?
Which of those constraints still need to be true?
What could the business now do that was previously impossible, impractical or uneconomic?
How might work be divided differently among people, AI, agents and systems?
Which decisions, roles and organizational boundaries would change if we designed the business around what AI now makes possible?
Sometimes this inquiry will confirm that the existing path remains the right one. Sometimes the process will be simplified or removed. In other cases, the process itself may cease to be the most useful unit of design.
AI transformation can start by improving what the business does today. But it cannot be limited by today’s business.
The goal is not to avoid paving every cow path. It is to stop mistaking a paved path for a transformed business.