Imagine asking a coworker a question. They turn around, whisper it to a stranger in the hallway, then recite the stranger's answer back to you — word for word, full of jargon neither of them understands, with total confidence.
That's a meat proxy: a human who moves AI output from one window into another and adds nothing but delay.
❌ please don't do this
Maya asked Tom because he's been at the company longer — he might actually know. What she got is a wall of text that nobody has read. Including Tom. Maybe it's right, maybe it isn't — that's the problem: neither of them can tell.
Maya still doesn't know if she can keep her vacation days. She has three paragraphs of hedged "it depends" that could have been written about any company, and it's now her job to find out which parts, if any, are true. She asked a question and got homework.
If she'd wanted an AI's best guess, she'd have asked one herself — it's one tab away. She asked a person, because a person can check before answering and stand behind what they say. Skip that part and you're not helping — you're a slower, lossier chatbot.
✅ instead, try this
Tom still used AI — and that's fine. Encouraged, even: it found the right section of a 90-page PDF in seconds. The difference is what happened before he hit send: he read that section, made sure it actually answered the question, and replied in his own words.
Eleven minutes instead of two. But Maya got an answer that's actually about their company, from someone willing to stand behind it.
"but it saves so much time!"
It saves your time by spending the other person's. AI answers are built to sound right: long, confident, sprinkled with terms you'd have to look up. Forward one unread and you're not sharing knowledge — you're outsourcing the fact-checking to the person who asked.
And it compounds. They paste your paste into their own AI to decode it. Now two chatbots are talking to each other through two humans, and nobody in the loop knows what's true.
Chat is just the obvious case. Send a report you haven't read, and your reader becomes the first human ever to check your work. Ship code you haven't read and pipe the reviewer's comments back into the AI, and your reviewer is now the author — while you play modem.
So before you send anything an AI wrote for you: read it, understand it, check it, then say it in your own words. If you can't explain it, don't send it — "that's what the AI said" won't save you. You sent it. It's yours.