What can an AI agent actually do for you?
6:19 · Horizontal
"Got it" doesn't mean done
Xiaoming hands a whole outing over to AI, and it goes wrong three times in a row. This episode is about what it takes for AI to actually get things done.
Checking tickets, booking them, reading the map, sending messages: Xiaoming hands the lot to AI and just shows up.
First, it mistakes "request received" for "tickets booked," and Xiaoming gets stuck outside the ticket gate. Second, it reads the notice about a change of address but never updates the destination, so Xiaoming trudges to the old building for nothing. Third, he asks it to write an email explaining things and show him first, and it treats the draft as ready to send.
In the second half, Xiaoting sits down at the computer and gets things back on track: spell out the goal, what it's allowed to touch, and how you'll check the result; try the same approach three times at most; wait for a person's OK before sending anything. An AI that looks at the results to decide its next step, and stops when it hits a boundary, is what this episode means by an agent. Next time AI tells you "got it" or "done," check what it actually did.
It lists twenty steps, every one of them sensible, and you still don't know where to start.
Say thisSwap the [ ] part for yours
Break [this thing] into no more than 5 steps. For each step, tell me what to do, roughly how long it takes, and how I'll know it's done. Flag the hardest step.
