AI Transformation · Adaptability
AI Can Remove the Work That Builds Judgment
The work AI removes may be the work that taught your next generation of leaders how to decide.
Key Takeaways
- Judgment is built by making imperfect decisions under real consequence, repeatedly, with feedback. Much of the work organizations most want to automate is exactly that, which is why it looked automatable in the first place.
- Friction and apprenticeship are indistinguishable on a process map. Both are slow and both are done by junior people. They differ on one question: does the person have to decide something they could get wrong?
- The cost arrives years after the saving, so no productivity measure will show it. What shows it is asking how long it now takes someone to become able to handle an exception unsupervised, compared with three years ago.
How judgment is actually acquired
There is unusual agreement across otherwise unrelated literatures about how expertise forms. Dreyfus and Dreyfus described the movement from rule-following to fluent situational recognition. Ericsson and colleagues established that the driver is deliberate practice rather than exposure, meaning repeated attempts at something slightly beyond current ability with feedback attached. Gary Klein's field studies of firefighters and commanders found expert decisions being made by recognition rather than deliberation, drawing on a store of accumulated cases that could only have been built by living through them.
Kahneman and Klein, who disagreed about a great deal, agreed on the conditions: intuition becomes trustworthy where the environment is regular enough to contain learnable patterns AND the person has had prolonged opportunity to learn them through practice with feedback. Both halves are required. An organization can leave the first intact and quietly remove the second.
Lave and Wenger supplied the organizational mechanism. Newcomers learn by legitimate peripheral participation: doing the real work at its edges, where the stakes are contained but the work is genuine. The edges are precisely where automation lands first, because that is where the work is most bounded and most repetitive. The economics of automation and the geography of apprenticeship point at the same tasks.
Friction and apprenticeship look identical
This is why the mistake is easy rather than careless. On a process map, rekeying an invoice and preparing a first-pass credit assessment are both slow, both junior, both obvious candidates. One produces nothing but delay. The other produces a person who, in four years, can tell when the numbers are wrong before anyone can say why.
The distinguishing question
Does the person doing this have to decide something they could get wrong, and do they find out? Two yeses make it apprenticeship. Automate it and something has to replace it. Anything else is friction, and removing friction is straightforwardly good.
Why review is not practice
The standard reassurance is that people will move up: the system drafts, the human reviews, the human is now doing higher-value work. Sometimes true. Often not, because reviewing begins from an answer, and the parts that build judgment happen before the answer exists. The options considered and rejected, the moment of not knowing which way it goes, the commitment made under uncertainty: none of those are in a review.
Lisanne Bainbridge described the same trap in 1983 for process control. Automating the routine leaves the human with only the exceptions, which are the hardest cases, while removing the routine practice that would have made them competent to handle one. She called it an irony because the better the automation, the worse the position of the person expected to take over.
Redesigning around graduated authority
Automate the gathering, keep the deciding
Most apprenticeship value sits in the judgement at the end of a task, not the collection at the start. A system that assembles the case and stops short of the call preserves the learning while removing the drudgery, and it is usually the cheaper build.
Graduate authority rather than granting it at a threshold
Widen the class of decisions a person owns as they demonstrate calibration, in steps small enough that being wrong is survivable. This is what the removed work used to do implicitly; doing it explicitly is what replaces it.
Make the person decide before they see the system output
Where a model will produce an answer anyway, have the human commit to theirs first. The comparison is the feedback loop, it costs almost nothing, and it converts a review task back into a practice task.
Give the outcome back to the person who decided
Kahneman and Klein’s condition is practice WITH feedback. An organization that automates the work and also loses track of who called what has removed both halves, and will not notice until it needs someone to handle an exception.
The Judgment Formation Map
One row per task under consideration. The last column is the one that does the work, and an answer of "they will review the output" is the failure case rather than the plan.
| Column | The question it forces |
|---|---|
| Task | What is the unit of work being considered for automation? |
| Decision | What does the person doing it currently have to decide? If nothing, it is friction and should go. |
| Ambiguity | How much of that decision is underdetermined by the available information? |
| Consequence | What happens when they get it wrong, and does it reach them? |
| Feedback | How and when do they learn the outcome? Delayed or absent feedback means the task was never teaching anyway. |
| Current human role | Who does it now, and where are they in their development? |
| Proposed AI role | Which part is the system taking: the gathering, the drafting, the deciding, or the whole thing? |
| Retained learning experience | If the deciding goes, what deliberately replaces it? An answer of "they will review the output" is the failure case, not the plan. |
The identity question underneath
Resistance to this kind of change is usually read as fear of replacement. It is often something more specific and more reasonable: the value of an expertise someone spent years acquiring has changed, and nobody has said what they are now being prepared to become. That is a rational objection to an incomplete plan rather than an emotional reaction to a good one, and it will not be answered by reassurance. It is answered by naming the next capability and the path to it.
What to measure
Training hours measure nothing here. Three things do. How many people below the executive tier made a consequential decision under ambiguity in the last year and were told the outcome. How long it now takes someone to reach the point of handling an exception unsupervised, against the same figure three years ago. And which roles have had their decision content removed while keeping their title, because a role can be hollowed out completely without any change to the org chart or the headcount, and that is the version nobody sees.
The Four A's reading
This is Adaptability, and it is the least visible of the four failures. Adaptability asks whether an organization can still learn and change. An organization that has automated its apprenticeship has not lost that capacity today; it has lost the pipeline that would have supplied it in five years, while every current indicator improves. March's distinction between exploration and exploitation names the shape: the returns to exploitation are immediate and certain, the returns to exploration are delayed and diffuse, and organizations reliably over-invest in the first.
It reaches Authority as well, because graduated authority is the deliberate replacement for what the removed work did implicitly, and an organization that will not widen a person's decision rights has no mechanism left to build judgment at all.
About the Author
Dan Flynn
Creator of The Four A's of Organizational Readiness™ · Enterprise Transformation Executive · Author, Builders Build
Dan Flynn has spent thirty years inside federal, defense, and commercial organizations: diagnosing the invisible conditions that determine whether capable people produce extraordinary results. He is the creator of The Four A's of Organizational Readiness™ framework, has reached more than 11,000 professionals across corporate, civic, and national security contexts, and took a federal data platform from one release every six months to seventy-two every two weeks by changing organizational conditions: not people.
His book, Builders Build: The Four A’s of Organizational Readiness™, is forthcoming.