The AI's Reply Was Still at the Top

Last week a colleague forwarded me an email he'd drafted with AI.

How’d I know? Because before I reached the body of the email, I read the AI's reply to him. You know, that short note these tools write before handing you the draft.

It was still sitting at the top, above the message he meant to send me.

It read “Yes — I’d make the confirmation explicit. Here’s a slightly more polished version:…”

That slip is the kind of thing any of us could make on a busy afternoon. What I kept thinking about, though, was what it said about the order of things. He had clearly read enough to know the draft was what he wanted.

He just hadn't read enough to notice it was carrying something extra.

I wondered if I had ever done the same thing. He didn't skip the review. He did the review most of us do now, the one that checks whether a draft looks done.

Polish works on a draft the way a pressed suit works on a job candidate. It gets you past the first glance, and the first glance is where a lot of us stop.

Like you, I keep seeing organizations ask, and increasingly expect, people to work AI into everything they do. The guidance on how to use it well and ethically hasn't kept up, so people optimize for the thing they can see, which is output.

Think about the last AI draft that went out under your name. Maybe a reply or update to a client or your leadership team, or, like me, a follow-up after a tricky meeting.

It probably came back so clean you changed a word, maybe two.

You read the first paragraph carefully, skimmed the middle, and read the last line to make sure it ended politely.

But…

Could you name its weakest claim right now, the one a sharp colleague would push on first?

Researchers at Microsoft asked 319 knowledge workers about 936 real tasks where they'd used generative AI. The more people trusted the tool, the less critical thinking they reported doing.

Confidence in themselves ran the opposite way. People who felt sure they could do the task on their own reported thinking more critically. Granted, that's self-reported, but it matches what I keep seeing.

If you're an expert, that second finding probably sounds like good news, and in my experience it holds up.

Hand most experts an AI comparison of vendors or a summary of a dataset, and they'll check it line by line. The guard comes down a little when it’s the email that frames the analysis or a message sent between meetings.

It's just an email, so it gets a skim. Yet that email is often the only part of your thinking other people ever read.

The even quieter cost is that when you start from the AI's draft, you end up grading its argument instead of building your own.

You can end up signing off on a conclusion you'd never have reached, simply because it arrived well organized. The thinking you skipped was the part that would have made the work yours.

People can feel this happening.

In a, 60% of employees said they worry AI is eroding their skills, and critical thinking was the one they named most often.

For an expert, that's also where influence starts to leak.

People act on your recommendations because they trust that your name on a document means you did the thinking. A draft you didn't really read borrows against that trust, and the colleagues who notice rarely say so.

The people deciding where your career goes next are watching for exactly this.

The same IBM study found 71% of HR leaders prioritize the ability to supervise, validate or override AI output, compared with 38% of employees. That gap is where an expert who actually reads the draft stands out.

I've started putting a small speed bump between the draft and the send.

Let’s get it…

It takes about sixty seconds and a few questions.

For the drafts that deserve more than a minute, I use the free tool, The Expert's Second-Pass Discipline

Sixty seconds would have caught my colleague's stray note.

More to the point, it would have given him the chance to ask whether the email said what he actually meant

The draft my colleague sent looked finished.

That was the problem.

Polish tells you a draft is readable. Only you can tell whether it's right.

Your influence rests on people trusting that your name means you did the thinking. The minute before you hit send is where you keep that trust.

Thanks for reading. Be easy!