Keep the Group Chat Human

My phone now offers to write my texts for me. It suggests replies to my sister, rewrites my apology to a friend in a "warmer tone" and offers to summarize the 140 unread messages in my oldest group chat so I don't have to read them.

I've started saying no to all of it, and I want to make the case that you should too. Not because the tools are bad — some are genuinely impressive — but because the group chat is one of the last places where inefficiency is the entire point.

The value of a bad reply

Think about the best message you got from a friend this year. I'd bet it wasn't polished. Mine was a blurry photo of a dog wearing a traffic cone, sent at 1 a.m. with the caption "this is you." Zero craft. Perfect message. It said I was thinking about you at a time when nobody makes you think about anybody.

Group chats run on that kind of signal. The typo that tells you someone typed fast because they were excited. The weirdly formal message from the friend who's anxious. The three-day silence from someone who's clearly going through something. Every quirk is information about a person you love.

An AI-smoothed reply erases that information. It produces messages that are a little better and a lot less legible. When everyone's texts are fluent, warm and well-punctuated, you lose the ability to notice when someone isn't okay.

Summaries are the real threat

Suggested replies are the obvious worry, but I'm more concerned about summaries.

"Here's what you missed: Jess got the job, the weekend trip moved to Saturday, Amir shared photos." Efficient. Accurate. And it turns your friends into a news feed. The summary keeps the facts and throws away the texture: the eleven messages of escalating capital letters when Jess got the job, the running joke that took over the trip planning, the specific photo Amir sent that made everyone go quiet for a moment.

If a chat has become so busy that you need a summary of it, the problem isn't the reading. It's that you're in too many chats, or the wrong ones. A summary treats the symptom and makes the underlying disease more comfortable.

The counterargument, taken seriously

I can hear the objection, and it's a good one. Some people find texting genuinely hard. People with dyslexia, people writing in their second or third language, people with anxiety that makes a simple reply feel like a final exam. For them, a tool that drafts a first attempt can be the difference between replying and disappearing.

That's real, and I don't want to moralize about it. There's a difference between a tool that helps you say what you mean and a tool that says something for you. Using AI to check a message you've written before you send it is a crutch in the best sense of the word. Tapping a suggested reply without reading your friend's message properly is something else.

The line, for me, is attention. If the tool helps you pay more attention to the people you love, use it gladly. If it helps you pay less, be suspicious.

A small manifesto

Here's what I've settled on, for whatever it's worth:

• Read the messages. All of them, eventually, in the chats that matter. If you can't, leave the chat.
• Write your own replies, even short ones. "Ha!" from you is worth more than a paragraph from a model.
• Let your typos live. They're proof of a human on the other end.
• Use help when you need it, not when it's offered. Turn off suggestions; turn them on deliberately.
• Keep one chat completely unassisted. The oldest one, preferably. Treat it like a garden.

We're going to spend the next decade deciding where AI belongs in our lives, and in most places the answer will be "more than you'd think." Work, school, paperwork, medicine. Fine. But it's worth keeping a few rooms in the house where the machine isn't invited. The place where your friends send you dogs in traffic cones at 1 a.m. seems like a good one to start with.