Give an AI a pile of your old messages and ask it to sound like you, and it will. It will also, sooner or later, say something you never would have said. That is not a bug you can patch. It is what the tool is built to do.
A twin fills the gaps
An AI twin works by prediction. It reads how you have written before and guesses the next word you would probably use. Most of the time the guess is close. The problem is what happens when the question lands somewhere you never wrote about. The twin does not stop. It cannot say "I do not know" because it was built to always produce a reply. So it fills the gap with the most likely sounding words, and those words go out under your name.
You will not see it happen. The message will read like you. The tone will be right. Only the content will be invented, and only the person who asked, acting on it, will find out later that you never actually said that.
The risk is not tone, it is commitment
People worry that an AI will sound robotic. That is the small risk. The real risk is the opposite: it sounds exactly like you while committing you to something you did not decide. A price you did not approve. A date you did not agree to. A yes you would have made a no. The better the imitation, the more your teammate trusts it, and the harder the walk-back later.
When a stranger sends a bad message, people forgive the stranger. When a message arrives under your name, in your voice, they hold you to it. Sounding like you raises the stakes of being wrong.
Speaking from a record is different
There is another way to be away and still answer. Instead of predicting what you would say, a system can answer only from what you already said. If you wrote it down, it can repeat it and point to where it came from. If you did not, it says so and stops. No guessing, no filling the gap, no words invented in your mouth.
The difference is simple to test. Ask it something you never wrote about. A twin gives you a confident answer that sounds like you. A record gives you "that is not written down anywhere, here is who to ask." One protects your voice by using it carefully. The other spends your voice on a guess.
Whose name is on it
When a tool guesses wrong, the tool does not carry the cost. You do. The apology is yours to write, the trust is yours to rebuild, the deal is yours to save. A system that speaks for you should be judged by one question: when it does not know, what does it do? If the answer is that it makes something up in your voice, it is not standing in for you. It is putting words in your mouth and handing you the bill.
Why this matters more when you are off
The whole point of standing in for you is that you are not there to correct it. You are asleep, or on a plane, or with your family. Whatever goes out goes out without your review. That is exactly the moment you do not want a tool improvising in your name. You want it to answer where you have an answer and to be honest where you do not.
That is how StandIn works. You write one short brief at the end of your day. While you are off, it answers your teammates from what you actually wrote, with a source under each reply, and when the answer is not in the record it tells them so instead of inventing one. It never speaks as you. It repeats what you already said, and it stops where you stopped.
When you're off, your StandIn is on.
It answers your teammates' questions from work you've already done, in your words, with a source under every answer.