Design Philosophies That Retain Value in the AI Era

*This article is not a critique or criticism of any specific company, individual, service, or event.
It is strictly a reflection and examination based on universal trends and structures, exploring future possibilities and challenges, as well as my own past initiatives, practical knowledge, and experiences.

Part 1: Before Delegating to AI, How Well-Defined Is Human Judgment?

Using generative AI makes many things faster.

Writing text.
Summarizing documents.
Organizing meeting notes.
Generating ideas.
Creating comparison tables.
Drafting responses to inquiries.

Tasks that used to take several hours a short while ago can now take shape in just a few minutes.

This is a very significant change.

However, what I have been feeling again recently is that what truly retains value in the AI era may not just be the ability to 'create quickly'.

Rather, as AI becomes capable of creating things faster, different skills will be called into question.

These are:
What to delegate to AI.
Where humans should make judgments.
At what point to stop.
What to treat as a premise and what to handle while still undecided.

It is that design which exists before the output is generated.

AI is convenient.

But precisely because it is convenient, it is easy to delegate everything to it.

It produces plausible text.
It produces clean comparison tables.
It produces convincing proposals.
It produces summaries that seem to have few omissions.

As a result, it looks like things are moving forward at first glance.

However, whether that output is truly usable is not determined solely by the quality of the AI's writing.

Before that, how well-organized are our own questions?
Are there criteria for judgment?
Is it decided what to prioritize?
Is it clear where humans should take over the thinking?

If those points remain ambiguous, AI can sometimes end up neatly wrapping up our own confusion.

Being able to create quickly will gradually cease to be special

From here on, I believe that the ability to create something using AI will become increasingly commonplace.

Generating project proposals.
Polishing text.
Summarizing.
Creating structural outlines.
Thinking of ideas for images and videos.

These things are, of course, important.

However, it is becoming harder to differentiate oneself with those alone.

This is because using similar tools allows many people to produce output at a certain standard.

When that happens, the ones whose value remains will not be those who can "make," but rather
those who can discern what should be made
.

To put it another way, it is
those who can shape the form of their judgment before creating
.

When you ask AI to do something, there are actually many judgments made beforehand.

What is the purpose?
Who is it for?
At what level of granularity should it be presented?
How specific should it be?
What should not be stated definitively?
What should be compared?
What should be excluded this time?

Even if these premises remain vague, AI will return something.

But having something returned is different from it being useful for making a decision.

Looking at the state of the question before the accuracy of the AI

When it comes to AI utilization, the conversation tends to focus on "accuracy."

Can it answer more accurately?
Can it write more natural text?
Can it understand intent better?
Can it handle more specialized content?

Of course, those are important.

However, when imagining actual situations where AI is used, there are problems that cannot be solved by accuracy alone.

For example, when the requester does not yet know what they want to decide.
When the criteria for judgment are not aligned among stakeholders.
When it is unclear whether the question is suited for AI or should be thought through by a human in the first place.

If you throw it at AI in this state, the AI will provide a plausible answer.

However, after seeing that answer, you might feel this way:

"It certainly feels correct, but I don't know if I can make a decision based on this."

This sense of discomfort might be a sign that the premise of the question is not yet ready, rather than the AI being at fault.

Before asking AI, I believe what is needed is not to think about
what you want it to answer
, but rather
what state you want to be able to make a judgment in
.

Rather than "design for delegation," use "design for controlled delegation"

When considering the introduction of AI, the topic often centers on "how much can we delegate?"

Which tasks can be automated?
How much can we reduce manual labor?
Which operations can be replaced?
How much efficiency can be gained?

This is a natural question.

However, just as important is the question of
where to draw the line and not delegate
.

Organization that is fine to delegate to AI.
Drafting that is fine to delegate to AI.
Comparison that is fine to delegate to AI.
Hypothesis generation that is fine to delegate to AI.

On the other hand, judgments that humans should oversee.
Decisions for which humans should take responsibility.
Conditions under which to stop if the AI is uncertain.
Areas that should not be automated while still ambiguous.

If you introduce AI without distinguishing these, it may be convenient at first, but you will face difficulties later on.

You don't know why that conclusion was reached.
You don't know who verified it.
You don't know where the AI's suggestion ends and human judgment begins.
It cannot be reproduced in similar situations.
It cannot be handed over to someone else.

In other words, while you have what the AI produced, it does not remain as organizational wisdom.

I feel this will become a very significant issue in the future use of AI.

Will judgment remain, rather than just output?

When you use AI, the output remains.

Text.
Tables.
Summaries.
Proposals.
Meeting notes.
Comparison documents.

However, what we truly need to leave behind may not be just the output itself.

Rather, what is important is the trail of judgment:
why it was shaped that way
what premises were considered
where the hesitation occurred
what was set aside this time
how to make a judgment when the same situation arises next time
.

Design that retains value in the AI era is not just about producing good answers, but perhaps
leaving room for the next person to rethink things
.

If only the output remains, it can be reused.

But if the structure of the judgment remains, it can be reproduced.
It can be improved.
It can be handed over.
It can be applied to other situations.

I think this difference is significant.

Pause for a moment before using AI

Using AI itself will become increasingly common from here on out.

That is precisely why it may become important to pause for a moment before using it.

What is this question intended to decide?
Do I really want to automate this task?
Or do I want to visualize the structure first?
What is the extent of the scope that can be left to AI?
Where is the place that humans must check at the very end?
If it were to fail, what kind of failure would likely occur?

These kinds of questions do not have immediate answers.

However, if you introduce AI without this preliminary organization, the AI, which should be convenient, may instead increase confusion in the workplace or organization.

What retains value in the AI era is not just being able to use a lot of AI.

Before leaving things to AI,
how well can we organize the human side of decision-making?

I believe that this design philosophy is what will become increasingly important from now on.

In Part 2, I would like to break down this concept of "organizing decisions" a bit more concretely.

In particular, aligning premises, putting things into an explainable format, and having conditions for stopping AI.

It may be in these areas that the real differences in the AI era will emerge.