A short observation. I think it’s a small one, but I keep noticing it everywhere now, so I want to write it down.

Over the last year, the way people communicate with each other at work has shifted in a specific way. The shift is small and the cause is obvious in hindsight, but it has consequences worth thinking about.

People are starting to write to other people the way they write to AI.

The pattern

You’ve probably seen this. Maybe you’ve done it. I’ve definitely done it.

The structure looks like this:

Hi Sarah,

Could you please send me the following:

  1. The Q3 client report (PDF version)
  2. The campaign performance data for September
  3. A list of all active campaigns with their budgets

Please format the response as a clear summary with each item separately addressed.

Thanks!

What’s wrong with this? Nothing, really. It’s clear. It’s structured. It’s polite. Sarah will know exactly what’s being asked.

But two years ago, this same email would have looked different. It would have been:

Hey Sarah, can you dig out the Q3 report and the campaign data from September? Also a list of what we’re running right now and the budgets. Bit of a headache, sorry — partner meeting tomorrow.

The second version is shorter, warmer, and gives Sarah something the first one doesn’t: context. She knows why she’s being asked. She might suggest something useful Mike hadn’t thought of, or push back on the partner meeting framing, or ask a clarifying question that saves an hour.

The first version doesn’t invite any of that. It’s a structured request expecting a structured response. It’s a prompt.

Why this is happening

The cause is the same as the cause of a lot of things this year: we’ve all started using AI for a meaningful chunk of our work, and the habits we develop while prompting AI start leaking into the habits we use with each other.

Prompts work best when they’re explicit, numbered, structured, and free of context the model doesn’t need. They’re optimised for a system that doesn’t care how you feel about the request, can’t push back productively, and will produce output that exactly matches the shape of the input.

When you spend hours a day writing in that shape, it becomes the default shape. So when you turn to write an email to a human colleague, the same shape comes out.

Why it’s a problem

Here’s the thing: humans are not LLMs. The properties that make a prompt good for an LLM are exactly the properties that make a message bad for a human.

LLMs don’t have context you need to share. Humans do. The chunk of “context the model doesn’t need” that you trim out of an AI prompt is exactly the chunk that makes a human request feel like it came from a person.

LLMs don’t push back. Humans should. If you write to a colleague the way you write to ChatGPT, you’re implicitly asking them not to push back, not to question, not to suggest alternatives. Just to comply with the structure you’ve given them. Most colleagues will pick up on this signal and just comply, even when they shouldn’t.

LLMs don’t have relationships with you. Humans do. Every interaction either deepens or thins those relationships. Bot-shaped messages thin them. They communicate that you’re treating the colleague as a system, not a person.

What it actually costs

The cost shows up slowly, in three places.

First, the quality of the work goes down. Because nobody pushes back. Because nobody volunteers the unsolicited improvement. Because everyone defaults to the literal request rather than the underlying need.

Second, the workplace gets colder. Not in a single dramatic moment. In a thousand small ones, where the warmth that used to be in workplace messages just isn’t there anymore. The team feels less like a team. The work feels more transactional.

Third, the best people leave. The people who are good at their jobs are the ones who add the most value through unprompted judgement. They notice things, suggest improvements, ask the right questions. When their environment systematically asks them not to do those things — by treating them as systems — they get bored. They leave. The ones who stay are the ones who are happy to behave like systems, which is the wrong selection effect for the long-term health of the work.

What to do

I don’t have a clever fix. I do think we should notice when it’s happening and write differently on purpose.

The next time you’re writing a message to a colleague, before you hit send, read it back and ask yourself: would you have written this two years ago? If the answer is no, and the reason is structural rather than substantive, consider rewriting.

Add the missing context. Drop the numbered list if you don’t need it. Use the word “I” more than you use the word “the” or “please.” Be a person. Sound like one.

If the request is genuinely simple, don’t dress it up to look complicated. If it’s complicated, give the colleague enough context to push back if they need to.

Save the prompt-shaped messages for the actual prompts. The colleagues will appreciate the difference. The work will be better.