What AI can and cannot do

AI is strong at transforming patterns—summarizing, classifying, drafting, and generating options—but it does not automatically know current facts, intent, accountability, or whether an answer is safe in your context.

At a glance

Question	Practical answer
When is it useful?	AI can draft three versions of a customer reply, while a person must confirm policy, private data, tone, and the final decision to send it.
What should you do?	Choose five recurring tasks and classify each as automate, assist with human review, or keep human-owned; state the reason and failure cost.
How do you know it worked?	Each task has an owner, an input boundary, a verification method, and a stop condition before AI output can cause an external action.
Common failure	Fluency is not evidence: confident language can still contain invented facts, broken code, or inappropriate advice.

flowchart LR
  A[Question] --> B[What AI can and cannot do]
  B --> C[Small example]
  C --> D[Evidence]

The important idea is not to stop at a definition: connect the concept to a small example and observable evidence.

Worked example

AI can draft three versions of a customer reply, while a person must confirm policy, private data, tone, and the final decision to send it.

Before acting, write the success signal. Change one condition at a time, observe the result, and record assumptions. For What AI can and cannot do, this separates what you know from what you are merely guessing.

Practice in 20–30 minutes

Goal: Choose five recurring tasks and classify each as automate, assist with human review, or keep human-owned; state the reason and failure cost.

1. Record the starting state and your prediction.
2. Implement the smallest version without adding unnecessary tools.
3. Change exactly one input or constraint and repeat.
4. Save a command, screenshot, output, or checklist as evidence.

Expected result: Each task has an owner, an input boundary, a verification method, and a stop condition before AI output can cause an external action.

What can go wrong

Fluency is not evidence: confident language can still contain invented facts, broken code, or inappropriate advice.

When the result differs from your prediction, do not change many things at once. Check inputs, versions, environment, permissions, and logs, then repeat from the smallest example.

Definition of done

• I can explain the concept in my own words.
• I completed the small example and kept evidence.
• I know one failure mode and how to check it.
• Someone else can repeat the work without guessing missing steps.

Go deeper

Use the linked resource or repository at the end of the page when you need a full implementation. Check current versions before applying commands to a real project.