What AI Is Good and Bad At
The jagged-frontier cheat-sheet: when to trust AI's output, and when to slow down and check.
Outcome: know at a glance when to trust AI’s answer and when to slow down and check.
Who it’s for: anyone who’s been burned by a confident wrong answer — or wants to avoid being.
The big idea: AI is jagged. It’s genuinely brilliant at some things and confidently terrible at others — often an inch apart. The skill isn’t trusting it more, or less. It’s knowing which side of the line you’re on. The whole pattern in one line:
AI is good at shaping language you can check. It’s risky with facts and figures you can’t.
✅ Reach for it (good at)
- Drafting & rewriting — emails, posts, first versions of anything.
- Brainstorming — options, angles, names, “what am I missing?”
- Summarizing & reformatting — long → short, prose → table, messy → tidy.
- Explaining & translating — plain-English a hard idea; rephrase for a specific reader.
- Editing your words — tone, clarity, grammar, length.
- Rubber-duck thinking — talk a problem through out loud.
Why these are safe-ish: you can read the result and judge it yourself. The cost of a miss is low.
⚠️ Slow down & verify (bad at / risky)
- Exact math & counting — it predicts text, it doesn’t calculate (it miscounts the r’s in “strawberry”).
- Current events — anything after its training cutoff (every model has one).
- Citations, quotes, dates, page numbers, links — the single thing it fabricates most.
- Niche, local, or obscure facts — your town, a small business, a specialized field.
- Anything you can’t undo — or anything about a real, named person.
Why these bite: the answer looks just as confident either way — so the wrongness is invisible until it costs you.
Why this matters
In a study of 758 consultants, people using AI on a task just past its ability were 19 percentage points less likely to be right than people using no AI at all. The danger isn’t that AI is sometimes wrong. It’s that it’s wrong confidently — and confidence is easy to mistake for correctness.
The move
When you’re working on the right-hand column, run the 60-second Is AI Wrong? Checklist before you act. And remember: this is about the type of task, not the brand of AI — a “smarter” model doesn’t make shaky facts safe (newer reasoning models sometimes make more of them up).
Safety note
Good at a task ≠ safe to paste anything into. Keep private details out regardless — see What Not to Paste into AI.
Next
Pick something from the left column to start with — see Your First Tasks to Try.