Most articles about AI’s limits talk about entire jobs and industries. That is not very useful day to day. Here are five specific, everyday tasks where reaching for AI first is genuinely the wrong call, and a human should do the thinking instead.
1. Delivering Bad News to a Specific Person
AI can draft the words for a redundancy letter, a client termination or a difficult performance conversation. It cannot carry the accountability of having said them, or read the room in real time and adjust. Use AI to organise your thoughts beforehand, then have the actual conversation yourself, in your own voice, present for the reaction.
2. A Genuinely Novel Strategic Decision
AI is strong at summarising known approaches and precedent. It is weaker when a decision has no real precedent, involves conflicting priorities with no clean answer, and where you will personally live with the consequences either way. A tax lawyer needs to know the law. A business owner deciding whether to pivot the entire company needs judgement AI cannot supply, because AI has no stake in the outcome.
3. Resolving an Angry or Distressed Customer Complaint
An AI chatbot handles routine questions well. A genuinely upset customer, especially over something serious such as a safety issue or a significant financial loss, needs to feel heard by a person who can make a real judgement call on the spot, such as an exception to normal policy. Escalating quickly to a human, rather than looping a frustrated customer through automated responses, protects the relationship far more than trying to resolve it with AI alone.
4. Final Review of High-Stakes Factual Claims
AI hallucination rates have fallen, but they have not reached zero. For anything with real consequences if wrong, such as a legal filing, a medical claim or a financial figure quoted publicly, a human needs to verify the specific facts against a primary source before it goes out. Treat AI-generated facts in high-stakes documents as a first draft requiring a genuine independent check, not a finished answer.
5. Building Trust With Someone New
A first client meeting, a job interview or a difficult negotiation all rely on a kind of trust that builds through genuine, in-person presence, not just correct answers. AI can help you prepare talking points beforehand. It cannot build the relationship itself. Use it as preparation, not as a substitute for showing up.
The Underlying Pattern
Each of these tasks shares something in common: real consequences that a person has to live with, and no fully objective right answer available to check the output against. AI works best on tasks with a clear right answer and low personal stakes. It works worst exactly where judgement, accountability and genuine human presence matter most.
Frequently Asked Questions
What kinds of tasks is AI actually bad at?
AI struggles most with tasks involving personal accountability, genuinely novel decisions with no precedent, and situations requiring real human trust or presence, rather than tasks with a clear, checkable right answer.
Can AI help prepare for a difficult conversation, even if it should not deliver it?
Yes. AI is useful for organising your thoughts and drafting talking points before a difficult conversation, provided a human actually delivers the message and stays present for the response.
Why does AI hallucination matter for high-stakes documents?
Even with falling hallucination rates, AI can still state incorrect facts confidently. For legal, medical or financial content with real consequences, a human should independently verify specific facts against a primary source before publishing.
Is AI generally getting better at judgement and decision-making?
AI is improving steadily at well-defined reasoning tasks, but genuine judgement in situations with competing priorities and personal consequences remains a distinctly human capability, since it requires a stake in the outcome that AI does not have.
Should a small business avoid AI entirely for customer complaints?
No. AI handles routine, low-stakes questions well. The key is recognising when a complaint is serious or emotionally charged enough that it needs quick escalation to a human who can exercise real judgement.
Curious about AI’s growing capabilities too? Read our explainer on what agentic AI actually means or browse AI News & Updates.
