What AI Is Good At, and What It Still Handles Badly

What AI is usually good at

• Turning rough notes into a structured draft.
• Rewriting text for clarity, tone, or length.
• Creating checklists, outlines, and example scenarios.
• Summarizing documents when a person can check the result.
• Explaining unfamiliar concepts in simpler language.

What AI handles badly

AI struggles when the answer depends on exact truth, fresh information, private context, accountability, or values. It may invent citations, miss exceptions, assume facts not in evidence, or produce a generic answer that ignores the situation. It can also make sensitive work seem easier than it is.

The review-cost test

Ask how hard it will be to review the output. If reviewing takes less time than creating from scratch, AI may help. If reviewing requires expertise you do not have, AI may create false confidence. For example, using AI to draft a polite refund email is easy to review. Using AI to interpret a contract clause is not easy if you are not qualified to evaluate the clause.

The mistake-cost test

Ask what happens if the answer is wrong. Low-cost mistakes can be fixed: awkward wording, incomplete brainstorms, rough outlines. High-cost mistakes can harm people, money, reputation, safety, or compliance. The higher the cost, the more you need expert review and reliable sources.

A balanced workflow

Use AI early when exploring and drafting. Use people late when deciding and approving. This keeps speed where it helps and judgment where it matters.

How to apply this guide

Use this guide for repeated workplace tasks where clarity, review, and privacy matter more than speed alone. The point is not to automate everything. The point is to let AI handle a draft or structure while a person keeps the business context.

• Write the task in one sentence before opening an AI tool.
• Decide which parts need human review: great for drafts, weak at accountability, risk depends on review.
• Remove private or unnecessary context before prompting.
• Check whether the final output changes a fact, promise, number, date, or decision.

Before using the output, check whether it changes a commitment, exposes private details, adds facts that were not in the source, or removes a nuance that your team actually needs.

A safer prompt to try

Use this starter prompt when you want help with the idea in this guide but still want the model to show its limits.