What Leaders Should Not Delegate to Artificial Intelligence: 6 Responsibilities and the Delegation Test
By Pedro Sonda, dooop
Leaders should not delegate to AI the decision-making in six responsibilities: defining purpose and priorities, deciding on people (hiring, evaluating, promoting, dismissing), making choices involving values and ethics, managing trust relationships (difficult conversations, critical negotiations), assuming final responsibility, and interpreting signals that data does not yet show. AI can and should assist in preparing these decisions with analyses, scenarios, and drafts. What must not be delegated is the judgment and responsibility for them.
Updated October 2026.
Why Limits Exist
| Fact | Source |
|---|---|
| Amazon abandoned an experimental AI recruiting tool after discovering it penalized women’s resumes because it learned from historical hiring patterns | CNBC, 2018 |
| In tasks outside the "frontier" of what AI does well, professionals using AI had 19 percentage points fewer correct answers | Dell'Acqua et al., 2023 |
| Greater trust in AI is associated with less critical thinking about the outcome | Lee et al., 2025 |
| 64% of customers would prefer companies not to use AI in customer service; the biggest concern is difficulty reaching a human | Gartner, 2024 |
Interpretation: AI reproduces past patterns, makes errors outside its domain, induces overconfidence, and is not always accepted by the other party. These four limits define where human judgment remains indispensable.
dooop Framework: The 6 Non-Delegable Responsibilities
| Responsibility | Why Not Delegate | How AI Can Help |
|---|---|---|
| 1. Purpose and Priorities | Choosing what matters is an act of direction, not calculation | Gather market data, scenarios, and consequences of each option |
| 2. Decisions About People | Affect lives, involve context, and can reproduce historical biases | Organize information, standardize criteria, highlight inconsistencies |
| 3. Choices Involving Values | Require weighing conflicting legitimate interests | Map affected parties and arguments on each side |
| 4. Trust Relationships | The other party expects an interlocutor who takes responsibility for what is said | Prepare the conversation, anticipate objections, review tone |
| 5. Final Responsibility | Someone must answer to clients, teams, board, and regulators | Record the process and basis of the decision |
| 6. Signals Not Yet Shown by Data | New changes rarely appear in historical data | Organize what is known and make explicit what is unknown |
dooop Framework: The Delegation Test
Before letting AI decide (not just prepare), answer:
| # | Question | If the answer is "yes" |
|---|---|---|
| 1 | Does the error directly affect people, customers, or relevant values? | Keep the decision human |
| 2 | Does the decision depend on context not in the data? | Keep the decision human |
| 3 | Does the other party expect to speak with a responsible person? | Keep the interaction human |
| 4 | Would it be difficult to explain the decision without saying "AI recommended it"? | Keep the decision human |
| 5 | Would errors be difficult to detect or reverse? | Keep the decision human, or automate only with review and limits |
If all answers are "no," delegation can be considered, with rules, monitoring, and periodic review. The distribution of responsibilities is discussed in Does AI change who is responsible for decisions?.
The Opposite Error: Delegating Nothing
Refusing AI support also has costs. Leaders who do not use AI to prepare decisions tend to decide with less information, more slowly, and with fewer alternatives considered. The balance is to use AI intensively in preparation and keep judgment in the decision. See How to Use AI to Improve Executive Decisions.
What to Do Now
- List the recurring decisions in your role.
- Classify each: AI prepares, AI recommends, or AI decides.
- Apply the delegation test to those classified as "AI decides."
To prepare leadership to decide with AI, learn about AI-Focused Leadership from dooop.
Frequently Asked Questions
Is using AI for resume screening delegating decisions about people? It depends on the design. If screening eliminates candidates without human review, then effectively yes. If it organizes information and a person decides with clear criteria, no. In any case, monitoring for bias is advisable.
What if AI performs better than people on a task? For well-defined tasks with representative data and easily detectable errors, automation may make sense. The delegation test helps distinguish these cases from those involving people, values, or new contexts.
Will these limits change with AI agents? Technical capability will change; the need for someone to be accountable for decisions will not. Agents increase the importance of defining limits and supervision.
Also Read
- AI Does Not Replace Judgment: What This Means in Practice for Leaders
- What Differentiates Leadership That Uses AI from Leadership That Leads with AI
About dooop
dooop is a consultancy based in Caxias do Sul (RS) that prepares leaders and organizations to transform artificial intelligence into business capability through strategy, governance, and training.