TL;DR - AI is brilliant at producing and reshaping language, and weak at live facts, exact math, citations, and judgment. Match the task to the tool and you're instantly more effective.
Why it matters
Most people either over-trust AI (and ship its mistakes) or dismiss it (and miss huge wins). The fix is knowing the boundary - then you delegate the right things and verify the rest.
Great at
- Drafting - emails, posts, outlines, first versions of anything.
- Transforming - summarize, rewrite, simplify, translate, change tone or format.
- Explaining - break a topic down to any level.
- Brainstorming - many options, fast.
Weak at (today)
- Live facts - prices, news, "what time is it" (unless the tool browses).
- Exact math & counting - it approximates; verify numbers.
- Sources - it can fabricate realistic citations.
- Judgment - ethics, strategy, people calls stay human.
Worked example - sort these
"Draft a reply to this email" -> great fit.
"What's our current stock price?" -> needs live data, not the model's memory.
"Total this column of 38 numbers exactly" -> verify, or use a real calculator/Sheets.
"Decide who to promote" -> human judgment.
Same tool, four very different reliability levels.
Steal this - the delegation rule
AI produces and reshapes. You supply the facts and the judgment.
Before using output, ask: would I bet money this fact is right?
If yes-but-unsure -> verify. If it's a judgment call -> that's your job.Common mistakes
- Asking it to be a calculator. Use Sheets, or AI's data/code tools that actually compute.
- Trusting a cited source without opening it. Open the link; it may not exist.
- Outsourcing the decision, not just the draft.
Good to know
Claude and ChatGPT are strong at the "produce and reshape" zone - drafting, summarizing, explaining. For numbers, ChatGPT's data-analysis mode and Gemini in Google Sheets actually run calculations rather than guessing. Pick the mode that fits the job.