What AI can't do

The same AI that drafts a polished essay in seconds can't reliably count the letters in “strawberry”, add two large numbers, or tell you today's news. Its limits aren't where you'd guess.

A jagged edge, not a smooth line

We expect ability to slope evenly — easy things easy, hard things hard. AI doesn't work that way. Its skill is jagged: superhuman at some tasks, oddly broken at others, with no smooth line between. The tasks that look hard to us (write a poem, translate) are often its strengths; the ones that look trivial (count, calculate, recall today) are where it stumbles.

Where the edges come from

Each weak spot has a plain reason. It reads words as chunks, not letters, so counting letters is genuinely hard for it. It has no calculator by default, so long exact arithmetic drifts. Its knowledge stops at a training cutoff, so it can't know today. And it's built to answer, so instead of “I don't know” it produces a confident guess — a hallucination. Knowing why an edge exists tells you when to expect it.

Working with the jaggedness

You don't need to memorize every edge — you need the habit of asking which side of the jag a task sits on. Is this a language task (its strength) or a counting, calculating, or current-events task (its weakness)? For the weak side, give it a tool or the facts: paste the numbers, attach the source, let it use a calculator or search. Matched to its strengths and propped up on its weaknesses, it's remarkable. Assumed to be evenly capable, it will surprise you at the worst moment.

• —AI ability is jagged — brilliant and brittle sit side by side.
• —Weak spots have causes: chunked text, no calculator, a knowledge cutoff, a bias to answer.
• —For weak-side tasks, hand it a tool or the facts instead of trusting it raw.