AX Is, in Effect, DT
AX is, in effect, DT—redesigned with AI as a fundamental premise.
To be more precise, AX is closer to a form of DT rebuilt and optimized for AI. And in practice, about 80 percent of DT is preprocessing.
Here, preprocessing does not simply mean cleaning up data. It includes defining business rules, identifying exceptions, structuring data, and converting information into formats that different systems can exchange and understand.
Until just one or two years ago, DT and RPA—Robotic Process Automation—were areas that only large corporations with sufficient financial resources or public institutions could realistically access, especially when productivity was measured in terms of time and cost.
Implementation costs were high, and analyzing workflows and turning them into actual systems required considerable time and manpower.
But as AI, particularly LLMs, became widely accessible, the barrier to entry dropped dramatically.
We are now moving beyond a stage where only a handful of companies use these technologies. They are passing the point of becoming common and beginning to approach the status of a standard.
What once seemed like the exclusive domain of a select few is becoming an everyday business tool—much like KakaoTalk became a mainstream communication channel on the back of widespread smartphone adoption.
Ah... right. The introduction got a little long.
There is only one conclusion.
Humans should step out of the preprocessing layer.
Based on my own experiments across several different approaches, I have found that the core of DT lies in minimizing human intervention in preprocessing.
People should optimize the part of the workflow they actually use: the interface.
The human role is to decide what information should be entered, what outputs should be reviewed, and at which points judgment and accountability are required.
By contrast, tasks such as data cleanup and transformation, database construction, and delivery between systems should be delegated to AI as much as possible.
Of course, some basic harness engineering is still necessary to prevent the entire process from collapsing when a model changes or a particular tool is replaced.
But I will not go into that in detail here.
AX, by definition, is a DT process designed around AI.
When humans intervene awkwardly using the same methods as before, AI adoption simply adds another human-centered intermediate step.
In the end, the system built for automation creates yet another layer of manual work and legacy processes.
Saying that humans should step out does not mean abandoning control.
Humans should still own the goals, standards, accountability, and validation.
But there is no reason for them to remain directly involved in every repetitive step of preprocessing, transformation, and delivery.
This is not about excluding people.
It is about delegation.
And ultimately, it is a question of roles and responsibilities.
Perhaps it is something like Season 2 of organizational culture.
The first version was about collaboration among people. The next version is about building a culture of collaboration among people and AIs.
It means clearly defining what humans should do and what AI should do.
Perhaps AX does not begin with choosing a new AI model.
Perhaps it begins with identifying the points in an existing workflow where humans should no longer be involved.
