Lesson 2.3: Tasks AI Should NOT Do

Purpose

You've learned what AI can help with. Now you need clear guardrails: tasks where AI should NOT be the primary tool or decision-maker.

This lesson prevents costly mistakes. Using AI on the wrong tasks can damage relationships, create legal problems, or undermine your credibility.

Why This Matters for Managers

It's tempting to assume: "If AI can do X, I should have it do X." This leads to problems.

Real consequences of wrong AI use:

• Employee files a legal complaint about how feedback was handled (AI-drafted without your understanding of context)
• You delegate a judgment call to AI, make a poor decision, and damage team trust
• You use AI for something confidential, creating security issues
• You rely on AI's analysis and miss a critical issue

The stakes are real. You're responsible for whatever you send, decide, or communicate—even if AI helped.

Core Concepts

The Four Categories of "Don't Use AI"

Category 1: High-Stakes Decisions with Real Consequences

These are decisions that affect people's livelihoods, careers, or legal standing.

Why not AI:

• Judgment, not analysis, is required
• Consequences are irreversible
• Accountability is entirely yours
• Context (relationships, history, politics) is crucial

Examples:

• Who to hire or promote
• Who to terminate or lay off
• Performance evaluations with consequences
• Compensation decisions
• Career development decisions
• Role assignments

What to do instead:

• Use AI to organize information or present options
• Make the decision yourself with full knowledge
• Have HR/legal review for compliance

Example (wrong):

"I'm deciding between two candidates. Have AI evaluate them and recommend who to hire."

Example (right):

"I have two strong candidates. Help me think through the tradeoffs. What should I consider?" Then you decide, informed by AI's analysis but guided by your judgment about fit, culture, potential, and team dynamics.

Category 2: Anything With Legal or Compliance Implications

These are tasks where getting it wrong creates legal exposure.

Why not AI:

• AI is not a lawyer
• Legal advice is protected professional work
• Mistakes have real legal consequences
• You have liability for what you communicate

Examples:

• Employment law decisions (contractor vs. employee classification, accommodation decisions)
• Discrimination issues
• Compliance decisions
• Intellectual property questions
• Confidentiality breaches
• Any communication that might become evidence

What to do instead:

• Consult with legal/HR for actual decisions
• Use AI to understand concepts ("Help me understand what employment law defines as 'accommodations'")
• Have legal review any output before using it

Example (wrong):

"I'm unsure if an employee's situation requires reasonable accommodation. Have AI tell me what to do."

Example (right):

"I'm unsure if accommodations apply here. Let me talk to HR." Or: "Explain what reasonable accommodation means," then talk to HR.

Category 3: Sensitive HR or Interpersonal Communication

These require psychological awareness, context, and relationships.

Why not AI:

• AI lacks understanding of how words land emotionally
• Context (relationship history, person's circumstances) is critical
• Tone and approach matter more than content
• Mistakes damage relationships irreparably

Examples:

• Difficult feedback about behavioral issues
• Communicating terminations
• Handling sensitive personal situations
• Conflict resolution
• Addressing discrimination or harassment
• End-of-life career conversations

What to do instead:

• Use AI to prepare your thinking ("How should I structure this conversation?")
• Deliver the communication yourself, with your voice and understanding of the person
• Have HR review sensitive communications before sending

Example (wrong):

"Draft the message telling Sarah she's being moved to a different team" and send it as-is.

Example (right):

"Help me think about how to approach this conversation with Sarah. Here's the situation [describe]. What should I consider?" Then have the conversation with her personally.

Category 4: Decisions Requiring Ethical Judgment

These involve values, trade-offs, and what's right.

Why not AI:

• Ethics requires values that AI doesn't have
• Trade-offs require judgment about what matters
• AI can present options but can't make moral choices
• You bear responsibility for ethical decisions

Examples:

• Whether to disclose a mistake to customers
• How to balance team welfare vs. business needs
• Allocating limited resources fairly
• Deciding whether to challenge an executive's questionable decision
• Environmental, social, or ethical trade-offs
• Prioritizing integrity vs. expediency

What to do instead:

• Use AI to explore implications ("What are the stakeholder impacts of each approach?")
• Make the ethical choice yourself
• Consider involving stakeholders in the decision

Example (wrong):

"What should I do about this quality issue? Should we disclose it?" (Treating it as a data problem)

Example (right):

"Here's a quality issue. AI can analyze the scope and impacts. But the decision about whether and how to disclose is an ethical decision I'm making: we disclose because integrity matters more than cost."

Tasks That Blur the Line (Use With Extreme Care)

These are tasks where AI might seem helpful but carry hidden risks.

Performance Evaluations

The temptation: AI can summarize feedback, identify patterns, and even draft narrative.

Why it's risky:

• Evaluations have legal consequences (discrimination claims)
• Performance expectations are contextual (AI doesn't know what actually matters for this role in your organization)
• Tone can unintentionally communicate bias
• If negative, it affects compensation, career trajectory, employment

Better approach:

• Use AI to organize feedback ("Here's feedback from 5 sources. What are the themes?")
• Write the evaluation yourself, informed by themes but shaped by your knowledge of the person
• Have HR review for compliance and fairness

Red flags: If using AI feels like it's replacing your judgment, don't use it.

Hiring Decisions

The temptation: AI can screen resumes, score candidates, identify top prospects.

Why it's risky:

• Hiring biases are real (gender, age, educational background bias)
• AI reproduces biases from training data
• Context (what you actually need, team dynamics, growth potential) is critical
• Legal risk if decisions appear to discriminate

Better approach:

• Use AI to organize candidate information
• Use AI to standardize interview questions
• Make hiring decisions yourself with full knowledge and diverse perspectives
• Have HR review for compliance

Red flags: If you're using AI to "be objective" in hiring, reconsider. Informed human judgment is more fair than seemingly objective automation.

Analysis of Sensitive Data

The temptation: AI can analyze employee data, engagement data, team metrics quickly.

Why it's risky:

• Employee data is confidential
• Some analysis shouldn't leave your organization
• AI systems may store or log your data
• Privacy and legal risks

Better approach:

• Check organizational policy on sensitive data before using external AI
• Use on-premise or private AI tools for sensitive data
• Anonymize or aggregate data before analyzing
• Never include identifying information in AI prompts

Red flag: If you're analyzing identifiable employee data in a public AI tool, stop.

Practical Managerial Use Cases

Case Study 1: Hiring (What NOT to Do)

Wrong approach:

"I have 50 resumes. Have AI score them by qualification and recommend the top 5 to interview."

Why it fails:

• AI learns from historical hiring data (which may encode bias)
• "Qualifications" might miss potential, cultural fit, or growth trajectory
• You lose human judgment about who could succeed in your role
• Discrimination risk if candidates from protected classes are screened out

Right approach:

"I have 50 resumes. Help me organize them by experience level and key skills. Then I'll screen them for interviews based on what I'm actually looking for."

Case Study 2: Performance Review (What NOT to Do)

Wrong approach:

"Here's feedback from my team on Sarah's performance. Have AI draft her review."

Why it fails:

• You're outsourcing judgment about fair evaluation
• AI's tone might unintentionally sound harsher or lighter than intended
• If Sarah contests the review, you need to own and explain every word
• It affects her compensation, promotion potential, career

Right approach:

"Here's feedback on Sarah from multiple sources. Help me identify the key themes. Now let me draft the review based on my knowledge of her, what I expect in her role, and her growth trajectory."

Case Study 3: Sensitive Communication (What NOT to Do)

Wrong approach:

"Sarah's moving to a different team. Have AI draft the message I'll send her."

Why it fails:

• This is news that affects her career and her feelings
• Tone matters. A cold, AI-drafted message damages relationship
• Context matters (why the move, what you think about her capability, what this means for her)
• She needs to hear it from you, in your voice

Right approach:

"I'm moving Sarah to a different team. Help me think about what she needs to hear. What should I communicate? How should I frame it? What does she need from me right now?"

Then have a conversation with her personally.

Case Study 4: Ethical Trade-Off (What NOT to Do)

Wrong approach:

"We have a deadline conflict and a team member is struggling. What should I do?" (Asking AI)

Why it fails:

• This is an ethical question about values
• The "best" answer depends on what you believe is right
• AI can present options but can't make the ethical choice

Right approach:

"Here are the options and their impact on the deadline and the team member. What do I believe is the right thing to do?" (Deciding yourself)

Anti-Patterns / Misuse Risks

Misuse Risk 1: Delegating Judgment Under the Guise of "Automation"

"I'll have AI decide this so it's objective."

Why it's risky: AI isn't more objective. It reproduces biases. And it has no judgment. You do. Own the decision.

Misuse Risk 2: Avoiding Difficult Conversations

"I'll draft this in email and have AI write it so it sounds professional."

Why it's risky: Some conversations need to happen in person or with you clearly present. AI messages feel cold.

Misuse Risk 3: Treating AI as More Fair Than You

"AI should make this decision because it's unbiased."

Why it's risky: AI is not unbiased. It's biased differently than humans, but it's biased. Your informed, thoughtful judgment is often fairer.

Misuse Risk 4: Confidentiality Breaches

"I'll paste the employee data into AI to analyze it."

Why it's risky: You don't know what happens to the data. It might be stored, logged, or used to train future models. Confidentiality violation.

Human Judgment Checkpoints

Before using AI on anything sensitive, ask:

1. Is this a judgment call? If yes, it's human work, not AI work.
2. Are there consequences? If yes, add review layers.
3. Could this become public or legal? If yes, be very careful.
4. Is sensitive information involved? If yes, check policy and use private tools.
5. Would I be comfortable explaining this later? If no, don't do it.

Responsible AI Considerations

Maintaining Accountability

Remember: You are responsible for whatever you send or decide, even if AI helped. This is actually a good thing—it keeps you in charge and forces you to think carefully.

Protecting Confidentiality

Not all information should go into AI tools. Employee information, financial data, strategic secrets—these belong in private systems or your own judgment, not in AI systems you don't fully control.

Recognizing the Value of Human Judgment

Some of the most important managerial work requires your judgment, your values, your relationships. Don't optimize all of it away. Human judgment is your competitive advantage.

Being Transparent About Limitations

When you're not using AI for something that sounds like it could use AI, that's often the right choice. "This is too sensitive to delegate to AI" is reasonable and honest.

Practice / Reflection Prompts

1. Your Red Lines: What 3 tasks in your role should never be delegated to AI? Why?

1. The Temptation: What task do you do that might be tempting to have AI handle, but you shouldn't? Why is it tempting? Why shouldn't you?

1. Legal Check: Are there legal implications in your role that you should talk to HR/legal about before using AI?

1. Confidentiality Review: What information do you work with that's confidential? How would you protect it with AI?

1. Decision Ownership: Think of a recent decision you made. Could you have delegated it to AI? Should you have? Why or why not?

1. Team Trust: If you delegated more decisions to AI, how might that affect team trust in you?

Key Takeaways

1. High-stakes decisions remain yours. Hiring, firing, evaluations, compensation.
2. Legal and compliance matters need professional review. Don't rely on AI for legal advice.
3. Sensitive communication requires your voice. Difficult conversations need you, not AI.
4. Ethical judgment is human work. AI can present options; you decide what's right.
5. Confidentiality is your responsibility. Protect sensitive information.
6. Some judgment shouldn't be automated. Your relationships, your values, your accountability.