Why being told what to do is not enough
At the close of a recent meeting, a colleague made a striking observation:
physicians, he suggested, often want to be told what to do.
There is truth in this.
Modern clinical practice is built around recommendations, pathways, and protocols. Much of what we do each day involves selecting and applying them.
In a previous essay, I argued that evidence is most useful when it follows, rather than precedes, clinical reasoning.
But even when evidence is used at the right moment, a deeper question remains.
What parts of clinical work can actually be reduced to instruction?
The appeal of instruction
There are good reasons to want clear guidance.
Clinical medicine is complex.
Time is limited.
The cost of error is high.
Well-constructed recommendations reduce cognitive load.
They standardize care.
They allow clinicians to act with confidence.
In this sense, the desire to be told what to do is not a weakness.
It is an adaptation.
The view from above
Modern evidence platforms perform an extraordinary function.
They synthesize vast bodies of literature into concise, actionable recommendations.
They reduce variation.
They make knowledge accessible at the point of care.
From this vantage point, clinical medicine can begin to look different.
Variability appears as error.
Uncertainty appears as a problem to be resolved.
Decision-making appears as something that can be standardized.
The work becomes one of organization.
Of distillation.
Of clarity.
And in that clarity, something important recedes.
The prior work of defining the problem becomes less visible.
Not because it is unimportant.
But because it is not captured by the system.
From this perspective, it is natural to conclude that what clinicians need most is guidance.
That they want to be told what to do.
The view from the bedside
But the view from the bedside is different.
Patients do not arrive as populations.
They do not conform neatly to inclusion criteria.
Their risks are not symmetrical.
Their values are not interchangeable.
A patient with thrombocytopenia may fit the outline of immune thrombocytopenia.
Or may not.
The problem is not only to know what is recommended.
It is to decide what kind of problem this is.
The prior act
Before any recommendation becomes relevant, a cognitive act has already taken place.
Signals have been filtered.
Patterns have been recognized.
Mechanisms have been considered.
Possibilities have been weighted.
Certain risks have been elevated above others.
This is not retrieval.
It is construction.
It is the formation of a working representation of the situation.
And it determines which questions are worth asking.
The asymmetry
It is easy to follow a recommendation once the problem has been defined.
It is harder to decide which recommendation applies.
It is hardest to recognize when none apply.
That last case is the most revealing.
When the patient does not fit the frame.
When the available pathways feel slightly misaligned.
When the recommendation is correct in general, but wrong here.
Recognizing that moment requires stepping outside the system that produced the recommendation.
It requires judgment.
This asymmetry is where clinical expertise lives.
The limits of delegation
Many elements of care can be standardized.
Testing pathways.
Dosing strategies.
Monitoring plans.
These are well suited to protocols.
They can be supported by evidence platforms.
They can, in principle, be automated.
But the act that makes those elements meaningful—defining the problem to which they apply—cannot be completed in advance.
It cannot be reduced to instruction.
Reframing the claim
It is true that clinicians often want to be told what to do.
But only after they have decided what kind of problem they are facing.
When that prior step is incomplete, instruction does not simplify care.
It misdirects it.
The real task
The challenge in modern medicine is not simply to provide better answers.
It is to preserve the work that makes answers meaningful.
If that work becomes invisible, it is not only neglected.
It is at risk of being replaced.
Evidence can guide action.
But it cannot determine the question.
And if the question is wrong, the answer—no matter how well supported—will be applied to the wrong problem.