Your AI agent has a tell, and it's not what you think

Everyone's checking punctuation while the actual tell hides in plain sight.

Amos Bar Joseph June 18, 2026

I’m Amos Bar Joseph**, co-founder of Swan, the first Autonomous Business OS. At Swan, we’re building what we call the Autonomous Business: a company that scales to $10M ARR per employee with no bloat, no assembly lines, no Cog Culture. Just humans in their zone of genius, amplified by AI agents.*** I write The Autonomous Age to share contrarian insights from that journey, on GTM, leadership, and the future of work. If you want to understand how GTM evolves beyond playbooks and assembly lines, this is where the story unfolds. Connect with me on Linkedin or X. If that's not the game you're playing, reply to unsubscribe.

You’re hunting em dashes to catch AI-written messages, but you’re looking at it wrong and it’s costing you. After analyzing thousands of AI messages, i can tell you it’s not what you’d expect.

The real fingerprints are structural, and they survive every model update:

-first you have the famous “It’s not x, it’s y”.. researchers named that pivot ‘negative parallelism’ and we’re starting to get allergic to it. -then somehow everything in threes. Always threes. “$1m ARR generated. Zero paid ads. With no SDRs”…. -and don’t get me started with “Honestly?”, “Here’s how:”, “no small feat”, “Impressed”… It shows up in AI cold emails so often it already became a meme.

The crazy part is that our outbound had all of it.

So last night I went down the GitHub rabbit hole, reading every open-source “humanizer” skill I could find.

One catalogs 33 AI writing patterns. Another runs 8 editing passes. And every single one of them hit the same wall:

Stripping the tics produces text that’s just as dead. clean, polite, voiceless. A different flavor of nobody.. BORINGGGGG!

The mature skills all converged on the same fix: feed it samples of YOUR writing first (or someone else you look up to).

Human subtraction isn’t the answer. Injection is.

So I built our own skill on that principle. We call it Humanizer.

Pass one deletes the structural fingerprints. Pass two rebuilds every message from emails I actually wrote and sent.. my rhythm, my word choices, my styling etc.

2 weeks after? Replies doubled. Prospects felt like someone really put the effort, and that goes a long way.

It’s not that my AI got smarter, it just finally got one specific person to imitate.

Reply “Humanizer” and I’ll send you the skill.

If you hate replying for lead magnets, then tell me about it…..

*[getswan.com

](https://getswan.com/?utm_source=theautonomousage&utm_medium=referral&utm_campaign=your-ai-agent-has-a-tell-and-it-s-not-what-you-think)

AI's signature isn't in the words it picks. it's in how it thinks: pivot then resolve, stack in threes, wrap it in filler. You can't edit your way out of a thinking pattern.

Build your own injection pass**

Stop stripping tics, start feeding it you instead.*

• Pull three emails you actually wrote and were proud of sending. That's your training material.

• Feed those into whatever you use to draft outreach, and ask it to match the rhythm, not just the tone.

• Send one message built that way. That's the only fix that's ever worked.

What's your AI's biggest tell, the thing it does in every single message no matter how you prompt it?

Stop editing out the AI, and start writing in the human.

-Amos

[Read more.

](https://www.linkedin.com/posts/amos-bar-joseph_youre-hunting-em-dashes-to-catch-ai-written-share-7470483342292058113-eXTp/?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAYm8JkB3GSdkfgU5-3_tdW6c48fsSkXvCs)