I’ve also had moments when the writing sounded human—but the judgment was missing

Last Tuesday night, I was revising a client proposal in a Hangzhou café. I used to organize the ideas myself, cut the filler, and check whether the tone was honest. But I was rushing, so I ran it through AI first. It read better and felt smoother, but it also felt like it had lost some of my judgment. I got nervous: if we outsource even proposal writing, do we slowly lose the ability to express our own position?

The issue isn’t whether AI can write, but who remains responsible for the words

The Pulse from Pragmatic Engineer raised a question that felt close to my work: could Amazon’s writing culture change as AI-assisted writing becomes more common? Amazon is known for complete, accurate written communication, and many teams rely on writing to record decisions and hand off tasks. But when AI can generate a credible-looking email in seconds, the important question changes: is text still expressing a position—or merely packaging a vague idea so it looks like a conclusion?

I thought of Sarah, a friend in remote operations. In Berlin, she explains complex outages to customers. She used to write her judgment first, then add the evidence and limitations. Later, she started using AI to tidy her sentences, but she still verified the timeline and conclusions herself. The machine wasn’t speaking for her; it was reducing repetitive work. The article doesn’t conclude that “Amazon has already completely changed.” It feels more like a mirror: we can use AI to improve writing efficiency, but we shouldn’t let it quietly take over our positions.

Replicating this approach is inexpensive today

I’d start with a low-risk experiment rather than handing every team conversation to AI right away. The cost can start at $0 per month: use the chat tool you already have, or try an AI writing tool with a free allowance. The time commitment is about 30 minutes: collect three real emails, ask AI to tighten the expression, then verify every sentence yourself. The technical barrier is low, and no coding is required. What does matter is being able to distinguish facts, tone, and promises. The first step is simple: open the AI tool you already use, start a new chat, upload an internal message you’ve already written, and tell it to “polish only; don’t add facts.”

I messed this up before. The first time I asked AI to “make this sound more professional,” it added a delivery date the customer hadn’t agreed to. That mistake taught me to redact sensitive information from real emails and personally confirm every sentence involving a promise. There’s no urgency to try this either; not every writing workflow is suited to automation.

How I’d set it up at each stage

If I’m just getting started, I’d keep writing personally: ten minutes a day, using AI as a tone checker rather than a ghostwriter. I might pick the email I’d least like to write, ask for three clearer versions, and choose one myself.

If I already have one or two clients, I’d create a “human confirmation required” list covering pricing, timing, promises, and disputed points. Whenever I use AI to revise a draft, I’d let it change the structure but not supply facts. That saves time on repeated polishing while reducing the risk of sending the wrong commitment.

If I’m scaling, I’d first standardize the team’s writing template: lead with the conclusion, follow with evidence, and close with the owner and next step. Then I’d use AI for tasks such as organizing meeting notes and generating first drafts. Not everyone needs the tool, especially when the team still relies on live discussion to move forward. Once the writing process is stable, I’d consider expanding what gets automated.