I’m writing a message to my team. Three sentences. Simple update.

I read it back. Does this sound right? Is it too direct? Will they think I’m being dismissive?

I paste it into Claude. Ask if it sounds okay. Get a response. Tweak it. Ask again.

Fifteen minutes later, I hit send.

For a three-sentence Slack message.

Shit, what am I doing?

The Validation I Already Needed

Here’s the thing: I’ve always done this.

Before AI, I’d send important stuff to my manager first. “Does this sound too harsh?” “Did I miss anything?” “Would you add something here?”

Year-end reviews. Retros. Performance feedback. Anything that mattered.

I already struggled with needing validation. The impostor syndrome was already there.

AI just became a second validator. One that’s always available. One that never gets tired of my questions.

And that should make things easier, right?

The Problem Nobody Talks About

It doesn’t make things easier. It makes things worse.

Because AI doesn’t just validate. It rewrites.

I write something in my voice. Run it through AI. It comes back polished. Professional. Clean.

And completely not me anymore.

I send it to my manager for review. He reads it and says:

“This is clearly AI generated. The person receiving it won’t appreciate that.”

Shit.

Embarrassing. But also the relief I needed. Someone finally said it out loud.

When It Started

When AI went mainstream, I started using it. For everything.

Grammar checks. Reframing ideas. Sparring partner for documents.

And it was helpful. It is helpful.

But somewhere along the way, it shifted from “helpful tool” to “I can’t send anything without checking first.”

I don’t even notice it anymore. It’s automatic.

Draft message → paste into AI → tweak → check again → send.

By the time it goes out, is it even my voice?

The Year-End Reviews

I wrote performance reviews this year. Fully myself, start to finish.

Then I ran them through AI. A couple iterations. Some suggestions were good. Some weren’t.

But the output? Machine-ish. Generic. It removed the personal touch that connected what the person actually did during the year.

I read it back and thought: this doesn’t sound like me at all.

So I rewrote it. My voice. But then I ran THAT through AI with the prompt: “Please don’t change my voice.”

And then I rewrote that version too, making it sound like me again.

I ended up with something that worked. But the process was exhausting.

Using AI as an outline helped. Filling in the gaps myself helped more.

But I still needed it. I still couldn’t just write the thing and send it.

The Two Types of Use

I’ve learned where AI actually helps and where it fucks things up.

Good use:

• Boring docs that don’t need personality
• Specs and READMEs
• Grammar and structure
• POCs and quick prototypes

For this stuff, AI is great. It removes 90% of the grunt work. Nobody cares if a technical spec sounds like me.

Bad use:

• Performance feedback
• Personal messages
• Anything where emotion matters
• Situations where someone asked for YOUR opinion

For this stuff, AI strips out exactly what makes it valuable. The tone. The personal connection. The humanity.

I know the difference.

And for the important stuff? I validate with AI, but I don’t just hit send. I read it again. Rewrite it authentically. Make sure it actually sounds like me.

But I still needed AI in the loop. That’s the part that bothers me.

How I Actually Use It

I’ve developed a method that works better.

Voice notes. I talk to AI on my phone. Tell it what I’m trying to say.

I tell it: “Ask me questions first. Don’t assume. Read beyond what I’m saying and pinpoint the end goal so I can validate it hits the nail.”

Then it gives me something back. A couple iterations.

Voice feels more like me than typing. And it’s faster.

I also use multiple AIs sometimes. Claude, ChatGPT, whatever. Cross-check if something feels too AI-generated.

I create personas. “Read this as a senior engineer who’s feeling burnt out. What do you think?” Trying to look beyond what’s on the page.

More context helps. My prompts are heavy with context because I’m trying to get better output.

And honestly? Sometimes it works. The rubber ducking with AI helps me think through things.

As an EM, you’re often lonely. You don’t always have someone to talk to. AI fills that gap.

But it doesn’t mean I have to use whatever it outputs. I should still validate it myself.

That’s the part I forget sometimes.

The Shame

Here’s what I don’t say out loud:

The shame isn’t about using AI. It’s about needing the validation loop at all.

You’re supposed to be capable. You’re the engineering manager. You should know what to say.

But management has no test suite. No passing green bar that tells us our communication landed correctly. The uncertainty is the job.

So I ask AI. And AI always has an answer. AI is never uncertain.

I don’t blindly use what it gives me. I rewrite. I validate with my own brain. I make it sound like me.

But I still needed to ask. That’s what eats at me.

There’s also judgment. People look down on AI-generated content.

“This is clearly AI-written, I’m not even going to read it.”

“This code is AI-generated” - said like it’s a bad thing.

And yeah, sometimes it is a bad thing. When it strips out the human part.

But sometimes it’s just AI-enhanced. It’s still my words. Just rewritten better than I could do alone.

The judgment I feel is mostly from myself though. Not from others.

The Manager’s Lens

After my manager called me out, I started looking at my messages differently.

Does this feel personal? Or does it feel like a template?

Would the person receiving this know it’s from me? Or could it be from anyone?

That lens helps. But it doesn’t stop me from using AI.

It just makes me rewrite more after AI gives me something.

Which is... still using AI. Just with extra steps.

When It Works (And When It Doesn’t)

Sometimes AI-heavy stuff works fine.

I’ve used it to build POCs. Quick prototypes to show engineers what I want. It doesn’t have to be perfect. It just has to demo the idea.

For that? Great. No shame. It’s a tool doing what tools do.

I’ve used it for boring documentation. Status updates. Technical specs.

Also fine. Nobody needs my personality in a README.

But feedback? Performance reviews? Messages where someone asked for my actual thoughts?

That’s where I have to be more careful. Because the thing they’re asking for is me. Not AI’s version of me.

I use AI to structure my thinking. To poke holes in my logic. But then I rewrite it in my voice.

The Loop I’m Aware Of

Here’s the pattern I’m trying to break:

1. Need validation because management has no clear feedback loop
2. Use AI to get that validation
3. AI gives me something that sounds too polished
4. Rewrite to get my voice back
5. Question if it’s good enough
6. Sometimes ask AI again
7. Eventually send it when it feels like me

The loop is still there. But I’m catching it more often now.

No Resolution

I’m not writing this with an answer.

I still use AI as a high-tech rubber duck. I still ask it to poke holes in my thinking. I still run important messages through it.

But I’ve gotten better at catching when it strips out my voice. At rewriting authentically. At validating with my own brain, not just AI’s.

The need for validation hasn’t gone away. Management still has no test suite.

But at least now I’m naming it. Maybe that makes it harder to ignore.

---

Note: Yes, this post is AI-enhanced. I talked through the ideas with Claude, it helped me structure them, and I rewrote it in my own words. The irony is not lost on me.

Let’s connect on LinkedIn - are you asking AI for permission too?