Why AI Agents Can Mess Up Real Tasks

AI agents can goof — like ordering 100 pizzas instead of 1, or emailing the wrong person. Always supervise!

Some examples

• An AI agent once booked a flight for the wrong day.
• Some AIs typed the wrong amount when paying for things.
• AI might 'agree' to scary terms in fine print without reading.
• AI doesn't always know when something looks suspicious.

Try it!

Imagine you asked AI to order one t-shirt. What should you check before AI clicks 'buy'? Make a checklist of 3 things!

The types of mistakes AI agents most commonly make

AI agents make mistakes in pretty predictable patterns — once you know the patterns, they're much easier to catch. The most common mistake is misunderstanding the scope: the agent does more than you asked (orders 10 items instead of 1) or less (only edits the first document instead of all three). The second most common is wrong assumptions: the agent assumes your contact named 'Chris' is the one at school, not the one at church. The third is cascading errors: one wrong step leads to the next wrong step, and by step five, the result is completely off track. The good news is that all of these mistakes are more likely to happen on big, vague tasks than on small, specific ones. 'Book me a flight' is much riskier than 'Find me three flights from Atlanta to Chicago on June 5th under $300 and show me the options.' Specificity is your best protection against AI agent mistakes.

• Scope mistakes: agent does too much or too little compared to what you intended
• Assumption mistakes: agent fills in ambiguous details with wrong guesses
• Cascading mistakes: one wrong step causes every subsequent step to be wrong
• Confident wrong answers: agent presents a mistake as if it's certain — always verify important facts