AIYING CaseLab #02
When Should You NOT Use AI?
The emerging professional challenge of deciding what to delegate, what to augment, and what to retain
AIYING CaseLab — Current Challenge → Judgment → Action → Value
AI is becoming easier to access, faster to use and increasingly capable.
For professionals, this creates an interesting paradox.
The first stage of AI adoption was dominated by the question:
“How can I use AI?”
The next stage may require a more sophisticated question:
“When should I not use AI?”
This is not an argument against AI.
It is an argument for better judgment about AI.
And that makes it an ideal challenge for AIYING CaseLab #02.
PART 1 — THE CHALLENGE
A very current signal
On September 9, 2026, Kristen Swanson, who leads AI fluency research and learning at Anthropic, described an increasingly important workplace phenomenon.
Sometimes the effort required to evaluate AI-generated work can exceed the effort required to perform the task yourself.
She calls this the “discernment tax.”
Swanson's broader argument is particularly important for professionals: sophisticated AI users are not necessarily those who use AI for everything. What distinguishes them is their judgment about what should and should not be delegated. (Business Insider)
That changes the meaning of AI fluency.
AI fluency is not merely:
knowing tools + knowing prompts + generating outputs.
It increasingly includes:
Knowing when AI adds value, when it creates additional work, and when human capability is the better choice.
AI is already deeply embedded in professional work
This question matters because AI use is no longer occasional.
The Thomson Reuters Future of Professionals Report 2026, based on a global survey of 1,816 professionals in law, tax, audit, accounting, compliance, risk and global trade, reports that 74% use AI several times a week and 44% use it multiple times a day. (Thomson Reuters)
But the same research reveals another important concern:
48% of professionals fear a negative impact on the development of independent professional judgment. (Thomson Reuters)
This suggests that the professional challenge is no longer simply whether AI will enter the workflow.
It already has.
The harder question is:
What should happen to human judgment once AI enters the workflow?
THE CASE
Consider Ananya, a senior manager preparing for an important leadership meeting.
Her organization is considering a new customer-service initiative.
Over several weeks, Ananya has:
spoken with customers,
reviewed complaints,
examined operational reports,
met frontline employees,
studied competitors,
and discussed implementation constraints with department heads.
She understands the problem deeply.
She now needs to prepare a recommendation for management.
Because her organization encourages AI use, Ananya decides to accelerate the work.
She gives an AI assistant her notes and asks:
“Based on this information, prepare a comprehensive recommendation for management, including priorities, risks and an implementation roadmap.”
Within minutes, AI produces an impressive document.
It has headings.
Tables.
Recommendations.
Risks.
Priorities.
An implementation roadmap.
It looks professional.
Ananya initially thinks:
“This has saved me a lot of time.”
Then she starts reading carefully.
The hidden work begins
One recommendation sounds sensible but doesn't fit the organization's operating reality.
Another overlooks an important customer concern.
Some priorities are generic.
One assumption is not supported by the evidence.
Several nuances that Ananya considered important have disappeared.
The document also sounds more certain than the underlying evidence warrants.
So Ananya begins correcting it.
She rewrites one section.
Then another.
She changes the priorities.
She restores missing context.
She checks the claims.
She restructures the recommendation.
Eventually she wonders:
“Did AI actually save me time?”
And then comes the more important question:
“Did AI improve my thinking—or merely generate something I now have to judge?”
The times and details in Ananya's scenario are illustrative. Ananya is a teaching-case character, not a reported individual.
The invisible cost of AI
When professionals evaluate AI, they often focus on one visible number:
Generation time.
AI produced the document in three minutes.
Wonderful.
But generation time is only one part of the workflow.
The actual work may look more like:
Prompt → Generate → Read → Verify → Detect errors → Restore context → Correct → Rewrite → Decide → Approve
Now consider the total human effort.
The AI-generated output may have been almost instantaneous.
But verification, correction and judgment are not instantaneous.
That is where the “discernment tax” becomes useful as a professional concept. Swanson notes that for tasks someone already performs expertly, evaluating AI's work can consume the time they hoped to save. (Business Insider)
AI can create work as well as remove it
This leads to an important distinction.
Suppose:
Without AI
Human work = 30 minutes.
With AI
Generation = 3 minutes
Verification = 15 minutes
Correction = 15 minutes
Reconstruction = 10 minutes
AI has technically completed the task faster.
But the workflow has not.
These figures are illustrative, not research findings.
The important point is conceptual:
AI speed is not necessarily workflow speed.
And:
AI output is not necessarily AI value.
A second problem: polished output can hide weak thinking
There is another danger.
Generative AI is very good at producing material that looks finished.
Structure, vocabulary, formatting and fluency can create the impression that the underlying thinking is equally strong.
But:
Presentation quality and reasoning quality are not the same thing.
Swanson's comments included a revealing workplace example: the CEO of legal-AI company Spellbook had told employees that, for some idea work, he preferred seeing rough original thinking rather than AI-polished proposals because he wanted visibility into the thinking itself. (Business Insider)
That raises a significant management question.
If AI converts an immature idea into an impressive presentation, can managers still distinguish:
good thinking presented simply
from
weak thinking presented beautifully?
The professional judgment problem
Now imagine Ananya's working day contains ten tasks:
researching competitors;
summarising a 50-page report;
finding patterns in customer feedback;
generating alternative approaches;
making a strategic recommendation;
checking calculations;
drafting a routine email;
deciding departmental priorities;
preparing presentation slides;
evaluating a sensitive customer escalation.
Should she use AI for all ten?
Probably not.
Should she avoid AI for all ten?
Certainly not.
So the real professional capability is neither:
Use AI everywhere
nor:
Keep important work away from AI.
It is:
Know what to delegate, what to augment and what to retain.
THE DECISION
Imagine you are Ananya's manager.
What would you recommend?
A. Use AI more extensively
Perhaps Ananya simply needs better prompting.
B. Stop using AI for strategic work
Important decisions should remain entirely human.
C. Let AI produce the work and require humans to verify it
AI provides speed; humans provide quality control.
D. Decide which parts of the task should be delegated to AI, which should be augmented by AI, and which should remain primarily human
AI involvement varies according to the nature of the work.
E. Something else
Before continuing, make your choice.
And ask yourself:
What exactly are you optimizing—AI usage, speed, quality, professional judgment, or value?
PART 2 — THE AIYING SOLUTION
The AIYING judgment
Our strongest starting point is D.
But the reason matters.
The problem is not whether Ananya should use AI.
The problem is determining the right relationship between human capability and AI capability for each part of the work.
AIYING therefore introduces a useful CaseLab decision lens:
DELEGATE → AUGMENT → RETAIN
These are not permanent labels attached to particular professions or tasks.
They are judgment choices.
As technology, context, risk and human capability change, the appropriate choice can change too.
1. DELEGATE
Let AI perform most of the task while the human establishes boundaries and verifies appropriately.
Good candidates often involve work that is:
repetitive,
high-volume,
structured,
time-consuming,
relatively easy to verify,
or primarily transformational rather than judgment-intensive.
For Ananya, examples might include:
summarising lengthy documents;
organising meeting notes;
extracting recurring themes from customer comments;
reformatting information;
producing an initial categorisation of complaints.
Here, AI may remove substantial low-value effort.
But delegation does not mean abandonment.
The professional remains responsible for determining the level of verification required.
2. AUGMENT
Human and AI work together.
This is likely to be the richest category for professional work.
The human brings:
context,
experience,
intent,
values,
accountability,
domain knowledge,
and judgment.
AI contributes capabilities such as:
pattern exploration,
alternative generation,
comparison,
analysis,
critique,
simulation,
and synthesis.
For Ananya's strategic recommendation, instead of asking AI:
“Write my recommendation.”
she might first develop her own position.
Then use AI to challenge it.
AIYING Power Prompt
Here is my current recommendation and the evidence supporting it. Act as a critical review partner. Identify assumptions I may be making, evidence that does not fully support my conclusion, plausible alternatives I may have overlooked, implementation risks and questions senior management is likely to ask. Do not rewrite my recommendation. Help me test my reasoning.
Notice what changed.
AI is no longer replacing Ananya's thinking.
It is creating additional surfaces against which her thinking can be tested.
That can be much more valuable.
3. RETAIN
Keep the task primarily human when human participation itself constitutes much of the value.
Examples may include situations dominated by:
accountability,
deep contextual knowledge,
sensitive relationships,
ethical responsibility,
negotiation,
trust,
original judgment,
or decisions where consequences are significant and difficult to reverse.
For Ananya, the final strategic recommendation may belong largely here.
AI can contribute evidence, alternatives and critique.
But Ananya may need to own the judgment.
This aligns with Thomson Reuters' broader finding that AI can act as a force multiplier while judgment, relationships and accountability remain human responsibilities in professional work. (Thomson Reuters)
Apply TAJ
Now put the decision through the core AIYING discipline:
THINK → ASK → JUDGE
THINK
Before opening an AI tool, understand the task.
Ask:
What am I trying to accomplish?
What part requires expertise?
What part is repetitive?
What part requires context?
What happens if the answer is wrong?
How easy will the AI output be to verify?
This alone may tell you whether to:
Delegate, Augment or Retain.
ASK
If AI has a useful role, define that role precisely.
Don't automatically ask:
“Do this task for me.”
You might instead ask:
“Help me explore.”
or:
“Challenge my assumptions.”
or:
“Find patterns.”
or:
“Generate alternatives.”
or:
“Check this against these criteria.”
or:
“Summarise this without making recommendations.”
The quality of AI collaboration depends partly on defining what role AI should play.
JUDGE
Now evaluate both the output and the collaboration itself.
Two levels of judgment are needed.
Output judgment
Is it:
accurate?
relevant?
complete?
supported?
appropriate?
safe?
Value judgment
Did using AI:
save time?
improve quality?
expand possibilities?
reduce effort?
surface something I missed?
improve the final decision?
Or did it create:
more checking?
more correction?
more confusion?
false confidence?
loss of nuance?
unnecessary cognitive effort?
This second level is often overlooked.
From “Can AI do it?” to “Should I delegate it?”
AI capability is expanding rapidly.
That makes a capability-only question increasingly inadequate.
Can AI perform this task?
may soon receive “yes” for a very large number of tasks.
But professional judgment requires additional questions:
Should I delegate this task?
What value would delegation create?
What new work will delegation create?
What judgment must remain mine?
Those are fundamentally different questions.
The AIYING AI Value Test
We can now make the hidden cost visible.
For a particular task, consider:
AI Value = Benefit Gained − Delegation Cost − Verification Cost − Correction Cost − Risk
This is primarily a decision lens, not a universal accounting formula.
Benefit gained
Time saved, additional insight, improved quality, new possibilities, reduced repetitive work.
Delegation cost
Time and effort required to explain the task, provide context and prepare appropriate inputs.
Verification cost
Effort required to determine whether AI's output is trustworthy.
Correction cost
Effort required to repair, rewrite or reconstruct inadequate output.
Risk
Potential consequences of inaccurate, inappropriate, insecure or poorly judged AI output.
If the combined costs exceed the benefit:
Using AI has not added value merely because AI completed the task.
Apply the test to Ananya
Suppose Ananya originally used AI to generate the entire management recommendation.
She experiences:
Benefit: faster first draft.
But also:
Delegation cost: assembling context and explaining the situation.
Verification cost: checking every recommendation.
Correction cost: rewriting generic or inappropriate sections.
Risk: important nuances disappearing from a consequential decision.
The AIYING judgment might therefore be:
Don't delegate the recommendation.
Instead:
Retain the judgment.
Augment the reasoning.
Delegate the supporting processing.
That gives us a much more intelligent human-AI workflow.
The redesigned workflow
BEFORE
Research → Human analysis → Recommendation → Presentation
Poorly designed AI workflow
Research → Give everything to AI → AI recommendation → Human checks and repairs → Presentation
AIYING workflow
AI-assisted information processing
↓
Human understanding
↓
Human initial judgment
↓
AI challenge / alternatives / pattern exploration
↓
Human evaluation
↓
Human recommendation
↓
AI-assisted communication
↓
Human approval
AI is used extensively.
But not indiscriminately.
This is not “less AI”
That distinction is important.
Delegate → Augment → Retain should not become an argument for minimizing AI.
The objective is not:
more AI
or
less AI.
The objective is:
the right AI involvement for the desired accomplishment.
And the answer is not fixed.
Swanson makes another important point: what should be delegated can change as AI capabilities improve. Professionals therefore need to periodically revisit tasks that AI previously handled poorly rather than freezing their understanding of AI at an earlier capability level. (Business Insider)
So professional AI judgment itself must remain dynamic.
There is also an organizational dimension
Organizations should not simply tell employees:
“Use AI.”
They should help employees develop judgment around:
where AI is appropriate,
how outputs should be verified,
where human expertise should override AI,
and
how AI-assisted work should be validated before consequential use.
Thomson Reuters explicitly argues that AI training needs to go beyond technical instruction and include opportunities for professionals to develop judgment by reviewing AI-assisted work, encountering edge cases and learning when human expertise should override the technology. (Thomson Reuters)
That is a major implication for AI education and professional development.
AI fluency needs another layer
Traditional AI literacy often emphasizes:
What is AI?
How does it work?
How do I prompt it?
What tools should I use?
These remain useful.
But professional AI fluency increasingly needs another capability:
AI delegation judgment.
The professional needs to recognize:
This I can delegate.
This AI should help me think through.
This judgment I should retain.
That is not a tool skill.
It is a professional capability.
ADD VALUE
How would Ananya know whether the redesigned AI relationship is actually better?
Measure the complete workflow—not merely generation speed.
Possible indicators include:
Total task time
not just AI response time.
Correction effort
How much AI-generated material required modification?
Verification effort
How difficult was it to establish trust?
Decision quality
Did AI surface meaningful alternatives or risks?
Rework
Did downstream corrections increase or decrease?
Human capacity
Was Ananya's time released for higher-value work?
Confidence with evidence
Could she explain and defend the final recommendation?
Now AI value becomes observable.
PROVE
Run the two approaches across comparable recurring tasks.
Don't rely entirely on the professional saying:
“AI felt faster.”
Measure:
time spent preparing inputs;
AI interaction time;
review time;
correction time;
total completion time;
quality/error indicators;
and, where possible, outcome quality.
This reveals whether AI genuinely improved the workflow.
AIYING VALUE RECORD — CASE #02
Challenge
Professionals may delegate tasks to AI even when verification, correction and lost context reduce or eliminate the expected benefit.
Intervention
Classify important tasks using:
Delegate → Augment → Retain
and apply Think → Ask → Judge before and after AI involvement.
Before
AI is used largely because it is available, fast or organizationally encouraged. Success is often inferred from output speed.
After
AI involvement is deliberately matched to the nature, risk, context and verification requirements of each task.
Value
AI is concentrated where it can reduce effort, expand insight or improve outcomes while avoiding unnecessary verification, correction and cognitive overhead.
Human Judgment
Humans determine what should be delegated, evaluate AI outputs, retain accountability and own consequential professional decisions.
Evidence
Measure total task time, verification effort, correction/rework, quality, errors, outcome improvement and released professional capacity.
A larger lesson for organizations
Case #01 asked:
Are we creating measurable value from AI?
Case #02 takes the question one level deeper:
Are we using AI in the right places to create that value?
These two cases connect beautifully.
Organizations cannot simply increase AI adoption and assume value will follow.
And professionals cannot simply increase AI usage and assume productivity will follow.
Both need judgment.
Recent McKinsey research reinforces the broader point that value comes from redesigning work rather than merely layering AI onto existing processes. Its 2026 State of AI research finds that nearly three-quarters of AI high performers report fundamentally redesigning workflows because of AI use, compared with about one-quarter of other respondents. (McKinsey & Company)
McKinsey's September 9 analysis similarly argues that AI value depends on intentional operating-model and workflow redesign rather than simply accumulating disconnected use cases. (McKinsey & Company)
The AIYING Management Takeaway
The first generation of workplace AI advice encouraged us to:
Find more things AI can do.
The next generation needs to ask:
Which things should AI do?
And professional maturity may increasingly be demonstrated not by the number of prompts someone writes, but by the quality of the decisions surrounding those prompts.
So CaseLab #02 gives us a simple discipline:
Delegate what AI can efficiently handle.
Augment where AI can improve human thinking.
Retain where human judgment itself creates the value.
And then:
Measure whether the combination actually improved the accomplishment.
AIYING CaseLab #02 — Final Principle
More AI is not automatically more value.
The professional advantage lies in discovering the right division of work between human intelligence and artificial intelligence.
Before your next prompt, therefore, don't immediately ask:
“What should I ask AI?”
First ask:
“Should I delegate this, augment myself with AI, or retain it?”
That small moment of judgment may be one of the most valuable AI skills a professional can develop.
About AIYING CaseLab
AIYING CaseLab takes current challenges faced by professionals and institutions and turns them into practical decision cases.
Each case moves through:
Current Challenge → Judgment → Action → Value
and, where applicable, concludes with the frozen:
AIYING Value Record
Challenge → Intervention → Before → After → Value → Human Judgment → Evidence
The objective is not merely to know more about AI.
It is to accomplish something valuable with AI—and prove it.
AIYING — Accomplishing with AI
Source & Fact-Check Note
The September 9, 2026 current signal about the “discernment tax,” knowing when not to delegate to AI, and periodically reassessing tasks as models improve comes from Business Insider's interview/reporting on Anthropic AI-fluency leader Kristen Swanson. (Business Insider)
The professional-adoption and judgment evidence comes directly from the Thomson Reuters Future of Professionals Report 2026, based on a survey of 1,816 professionals conducted in March–April 2026. It reports 74% using AI several times weekly and 48% concerned about its impact on independent judgment development. (Thomson Reuters)
The workflow-redesign evidence comes from McKinsey's 2026 State of AI research and its September 9 operating-model analysis. (McKinsey & Company)
Ananya, her organization, workflow and any numerical examples attributed to the teaching scenario are fictional/illustrative. They are deliberately separated from externally verified research findings.
This makes CaseLab #02 complementary rather than repetitive: #01 asks whether AI creates value; #02 asks whether we are allocating AI to the right work in the first place.
No comments:
Post a Comment